This plus the pinebook pro, I really hope that it will help the software availability on armv8 to skyrocket!
Also I really hope that someone will come with a new case to use it as tablet and laptop. Like a mix between the old pi-top and a surface book (or a PineTab).
If anyone has experience developing on the raspberry pi 4 (4GB), I really want to hear from you, especially regarding the processor performance.
It sounds like pine64 devices are miles ahead in terms of upstream kernel support, though. The remaining issue was with GPU support, but that's being taken care of with the Lima/Panfrost project, which now runs great.
Of course, you are free to pick whatever you want from their store, but comparing a raspberrypi with a PineTab is a bit unfair, as these two products are quite different.
Moreover, I disagree with your suggestion that 2GB is useless in a tablet. It really depends on the use-case, and I have rarely seen a capability range similar to the pineTab in a tablet from-factor.
I spent a few months doing much of my development on the 4 GiB Raspberry Pi 4 (gently overclocked). My experience was that of a slightly older notebook - totally fine for most things. The 4 GiB memory was hardly ever an issue, but the pathetic IO (storage and to some degree network) eventually drove me away. Thus, 8 GiB is a great option, but it makes it seem even less balanced now.
EDIT: I haste to add that having the full 64-bit Ubuntu 20.04 was what made this interesting to me.
No one sane uses the microSD for anything but the boot partition :) I used a fast USB key and contemplated using a true fast USB 3.1 SSD, but I really didn't like the bulk (the SSD would have been as big as the Pi). I never lost any data (but I also only accessed it remotely - it was head-less).
> If anyone has development experience using the raspberry pi 4 (4GB), I really want to hear from you, especially regarding the processor performance
I've been developing on the platform for a few months now. Though it's difficult to compare performance between architectures, the new BCM2711 compares favorably in CPU bound tasks to Intel's Cherry Trail Atoms from early 2016, primarily the Atom x5-z8350.
The Pi 4 also overclocks easily from 1.5 GHz to 2.0 - 2.1 GHz, where it exceeds Atom performance in some cases. The A72 cores are a huge leap in performance compared to the A53 cores of the RPi 3. These are architecturally "big" cores, as compared to the more power efficient and lower throughput "little" A53 cores of the previous model.
As someone currently working on supporting application and firmware code on this platform, this is pretty interesting. I'd also like to point out that Buildroot (https://buildroot.org) and upstream U-boot have first class support for building and running your own 64-bit kernel and userland based on included defconfigs.
If you have an application you'd like to ship on a Raspberry Pi image, especially if you want to address all 8 GiB from a single process, check out Buildroot.
8GB is enough to run ZFS with memory ARC backing for TB sized disks behind raidZ2. I've run this with Intel 64bit CPUs, I can't see a reason this won't also run in reasonable time for arm. (Both memory shuffles and the checksum computation)
With the right PiHAT this makes raid FS much nicer.
Before anyone goes "OMG I could use it as a desktop" $75 is close to what you would pay for couple year old ~4 faster laptop with broken screen, sodimm and PCIE SSD slots. Quick ebay check shows things like $90 'Dell XPS12 i7-3537U 8gb 128GB SSD broken screen'
I do agree with that thought, as I personally use these Dell Optiplex i7 3rd gen, 32gb desktops, I'm not a mega gamer but I can throw in a 4GB 1050ti and play some recent games at medium quality. It's definitely not $75 more like $500 altogether but still... what I do like about this is it seems like a less interesting thing to steal if you're concerned about that.
Generally though it is like a "full computer" so I don't know what I would buy one of these for, I currently have 3 Pis that have been on for a few years now continuously they're running web servers/automated tasks mostly Pi Zeros and a 3B+.
I think it's crazy 8GB is becoming like the new 4GB... 32GB is almost not enough personally.
Not sure if it is more fun to use as a desktop a machine without a normal SSD, questionable video hardware, incompatible with x86. It is certainly an interesting toy, but as a daily driver - nah.
Maybe in the US. In my EU country(Austria) second hand hardware prices are through the roof. There's no way you're getting anything half decent from the last decade for $75.
And then there's the cooling and power consumption difference between a 10 year old system and a RPi. It's not even close.
I don't know where you are in the EU but UK refurbishers constantly sell post-lease desktops and laptops on ebay for around £50, shipping to any EU country is usually £10-20 on top of that, if anything.
Is plentiful and easy to buy, again - ships anywhere in the EU.
As for the power consumption - I have to disagree again, I have an SFF like this and in idle it uses about 10W of power, 40W under load.
Rpi4 4GB uses 1-3W in idle, and up to 15W under full load. Yes, the numbers for the Pi are lower, but I wouldn't say "it's not even close". It's close enough that I would absolutely consider getting the larger PC instead.
Yes second hand cars in the UK are cheaper, factors being that there has been a consumer rise for a while, seeing more second hand cars. This with the island effect makes for dispersal of those second hand goods more curtailed and for example you may be able to get cheaper cars in Germany second hand if the geography curtailed those easily shifting to nearby countries who have had less of a consumer drive.
Which is why my friend in Bulgaria will buy a second hand car in the UK, drive it back, sell it for twice what he paid for it. Then in Bulgaria Walnut wood is about 1/5th of the price of what it demands in the UK. But most opportunities in life are going the extra mile.
Yes UK is still technically an EU member during the transition period. How exporting/importing of cars pans out, well as long as the we don't end up with trade loop-holes - then I'll be happy and by that. Say EU has import tax of 30% on UK cars, new or second hand. Yet no such taxes for imports from say canada. Then the UK has no tax import on cars from UK to canada, so would be possible to export to Europe, via Canada to skirt such import taxes.
That kinda thing worries me as environmentally, not good. But could happen.
Well, the obvious issue you're missing here is the fact that the steering wheel is on the right/wrong(heh) side, hugely limiting the potential for exporting your second hand cars elsewhere. Even then, I do know about a few successful businesses, where people have been exporting cars from UK to Poland, moving the steering wheel to the other side, and yet still making money in the process, the value of second hand cars is so much greater there(in the UK a car that is 6+ years old is basically ancient, in Poland that's still considered "nearly new").
I love that HN's culture is one where people can gracefully admit that they were wrong, and that's that (no further dogpiling to call them out on the earlier error).
Well, it was an example, but our technology such as laptops seems to sell cheaper second hand, same with articles of clothing or furniture.
I believe it stems from a tradition of not disposing of things, and instead selling them on (via "car boot" sales) or handing them down. I believe (and this is pure conjecture) that it's a recent thing for brits to dispose of things with such frivolity, and this is somewhat echoed in the car-boot sales of late being chock-full of brand new mostly chinese cheap crap.
And maybe if you have an overabundance of second-hand goods then maybe the price goes down as all of the friction to sell even things considered "tat" are gone.
And most of the things we would consider "tat" are not "tat" to everyone else, for instance, I bought my first computers for 5gbp from a boot-sale, for such a token amount of money it is likely a person would not have sold it another way. Thus, I get a cheaper computer than I would have had to pay otherwise.
Similar issue in my country (Portugal). On our largest second-hand market (OLX.pt) the laptops in the 60-80 EUR range are pretty much all described as "for parts", and their specs are not comparable to this RPi model.
The snide comments like the GP's are due to them living in the american bubble.
EDIT: also note the comment you're replying to is using a traditional workstation "slab" as an example. I'd rather use a small RPi that frees up more desk space than buy something that will make my work environment at home even more cramped!
The EU single market has no positive or negative effect here as it's a private purchase not subject to any regulations.
I am free to buy and sell to and from any country in the world, the issues is not everyone is willing to deal and pay for international shipping which ends up quite high for a cheap SFF tower system.
> not everyone is willing to deal and pay for international shipping
In principle, shouldn't one of the consequences of the EU project be that intra-EU shipping rates should, for the same distance, be identical between within the same country and between EU countries? (Of course, I realise that outcome may not yet have actually been achieved in practice.)
Somewhat. Population also tends to be centered around the country so it even if transportation is better people still have their lives setup such that practically shipping in their country is cheaper.
The EU is a single market, but that doesn't stop all the countries having different languages and consequentely different keyboard layouts. So you are looking in your country, perhaps in a second, which shares your keyboard layout.
About laptops, I mostly agree. (While the wrong keyboard layout is something you could learn to live with or workaround, in practice it may well be too unpleasant to be worth it.)
For desktops, it shouldn't really matter. One quite likely already has a spare USB keyboard with the right layout, and if one doesn't, it shouldn't cost much to buy one (whether new or second-hand).
Big markets like US, UK and Germany have low prices on consumer products and in small markets like Austria and BeNeLux countries consumer products cost more.
Supply and demand or in this case, a bottleneck hold on a supply network that has shifted from warehousing months towards a just in time production/distribution model and with that, more susceptible to any bumps. This causes shortage spikes and price rises of new, which drive up the corresponding second hand offerings.
Same way hard drives a few years back increased in price, supply and demand as a factory went out of action for a while, curtailing market supply. Also why external USB drives became source of HD's as they had a larger supply buffer as less popular at the time and more in warehouses/stores.
On the subject of mini-PCs spesifically, You can buy Intel Atom / Apollo lake system like Beelink x45 for a similar price on Aliexpress, but it's going to be significantly more performant and comes with real storage options.
The PI kicks ass if you want to play with hardware and actually use it's pins to build robots or whatever, or you want to use it's amazing camera and libraries. For pure compute its not a great deal.
I agree that the $75 is misleading. That doesn't include shipping, a power supply, a small SSD, etc.
And the used market is great. My favorite for a very small desktop are the Lenovo USFF devices like the M92P, M93P, etc. You have to search a bit, but you can find 8GB models with an i3/i5, small SSD, and Windows 10 for $110 or so on eBay.
I cannot imagine why anyone would feel anything other than unalloyed delight at the prospect of doing their computing on a used, broken laptop purchased at auction.
Wow 8GB is nice impressive. It's too bad to us it's easy to use these things but comparatively easier to say "here's a laptop" to someone less inclined. I was trying to suggest these to people as replaceable computers that you don't have to worry about getting stolen eg. just keep cheap monitor/keyboard/basic phone charger... anyway that's neat. Basic phone charger may have to get bumped up for more amp(s) but yeah.
Does anyone know if the rPi4 can play HD (1080p) video in a browser smoothly yet? I bought one to use as a media PC after reading that it could play 4K video. Apparently it can’t even do 1080p in the browser though (YouTube etc)
Yes it can do full hd, ever since at least rPi3. Problem here is chromium. Use omxplayer and the rPi4 will do 4k in full glory. I think you can even use omxplayer with youtube somehow. Chromium doesn't even do hardware decoding on x86 desktops, you just don't notice it, because the cpu's are fast enough to do it nowadays.
> Chromium doesn't even do hardware decoding on x86 desktops
Which is a ridiculous waste of energy... :( you notice it too now using Microsoft Teams (Electron) and Google Meet versus Facetime. Eats your battery in no time.
There is no Linux like solution for android phone that runs Xserver and OpenJDK reliably. Samsung DEX emulation was the only usable solution, but that was killed off. Termux can not run OpenJDK, some weird issues with crashing.
Also most devices do not have enough cooling capacity, max performance can be only delivered in short bursts before CPU gets throttled down. If you hack it to run at full speed for hours, it freezes.
PI is simply the only option at this size, with enough CPU power and good Linux support.
This is cool. Whats really holding it back at this point is it's use of a sd card.
I find that its pretty difficult to use 4gb of ram on a pi practically, and its often bottlenecked by I/O. It might just be my SD card, but executing trivial commands sometimes takes forever and reminds me that I'm in a pi. I've heard ssds can improve speeds a lot, but you still cant boot off usb.
Running from RAM would also be great for having an effectively static system -- just reboot to get a clean slate; the hardware storage needn't even be available after boot.
It's fairly trivial to build an entire OS into an initramfs to run from memory. Also, the poor performance of most microSD cards is primarily due to their flash quality, and designed use case in phones and cameras for external storage. Using a quality thumb drive or SSD over USB 3.0 would yield substantially better performance.
The OS uses free RAM as a cache for file reads anyway. Temporary directories that are not meant to persist after reboots are often mounted as a ramdisk however.
That’s been available on earlier versions of the Pi (2B v1.2, 3B, 3B+).
It was not available at launch for the Pi 4, but was released in beta firmware in late September, and is currently part of the critical firmware update.
It works well, allowing separate boot images for each individual Pi on your network.
When using netboot, the boot rom will pull the files it needs from the tftp server. It will first look for them in a subdirectory with the Pi’s serial number as its name.
If you’re netbooting without any sdcard installed, I’d suggest disabling sd polling.
It does, but I think what the Pi does is connect to the TFTP server handed out by PXE and then look for a folder named serial-number. If it doesn't exist in the path PXE gave it it just boots whatever is in the tftproot.
Might or might not. It does DHCP request and server-side, you can do decision on what it returns on any tag you want, often client architecture (0x93); you can then point to the right tftp server, or if you have UEFI client, you can also use http.
Intel should be making an Optane sdcard for embedded use, potentially with a built in power down capacitor to make sure all acknowledged writes make it to Optane.
I’m sure that’s true. I’ve only used this for Rpis, which have it as a neat pull-based capability. Just add a subdirectory with the right name for each one in the tftp root.
You actually can boot straight from USB as of very recently.
Something you've always been able to do is to keep just the boot partition on a small SD card and the root on whatever USB device you like. And yes, even with the supposedly best-performing SD cards, they easily become a problem both with performance and failure.
> You actually can boot straight from USB as of very recently.
IITC that is still not officially supported (as a beta-test feature?), though many report it works as intended.
In the meantime if people don't want to use a beta feature the boot-from-SD-mount-root-elsewhere option works well too. You could also run it of an NFS share instead of a USB stick, though the performance metrics will differ.
I did an article on the topic after the Pi firmware update made the situation a lot better. A bare Pi is okay without a fan, but if you want it in a case, the Flirc case is your best option (IMO): https://www.jeffgeerling.com/blog/2019/best-way-keep-your-co...
Great article, thanks. The Flirc case looks good and for $15 it's an obvious buy.
Having just spent a bit north of $1k on a new CPU, RAM, and mobo for my desktop, I'm having a bit of sticker shock (in a good way) at how low these RPi prices are.
For my use case, I need access to the GPIO pins. The blurb on the Flirc says that access is possible but I don't see any opening. Can you confirm that GPIO access I possible?
The gpio access requires a header cable routed out a slot in the base of the case.
I have several pi in flirc cases, one is a pihole and one runs CI jobs (slowly) for PhotoStructure, and they don't get too hot even under sustained cpu load.
With a $10 case/fan combo mine idles around 30c and hits about 60c under load (this is in a room at about 21c). The Pi doesn't start throttling til about 80c
I have the same case and share your opinion. It gets warm, but CPU temperature is well within limits. And somehow the idea of passive cooling is very attractive.
I have a fan shim on mine, and it keeps it well within limits. The shim is controlled by the on die temperature, so it runs slowly most of the time, but if you run the CPU/GPU hard for any length of time, the fan really cranks up. You have to add a driver to have the automatic control.
I had a passive aluminum block heat spreader on it, and that works if you have good thermal bonding, unless you are really working it for a long duration. The problem is, if you hammer it long enough to throttle, that same mass takes a long time to cool back down.
This is one thing I always appreciated more about the BeagleBone but the Pi being more popular I've gone with it as my main SBC of choice for obvious reasons. The Pi has always been cheaper as well.
A new spec for microSD cards (microSD Express) was announced last year, which would let them use similar protocols to SSDs. I don't know if this means they'll support things like TRIM, without which their performance may initially be fast but degrade over time. Still, it would be a significant improvement.
For folks considering running 64-bit on the pi - I don’t think you need 64-bit userland. I’ve been running a 64-bit kernel on my pi with 32-bit userland perfectly fine. The few things that actually need 64-bit (I’m looking at you mongo) run in 64-bit mode via docker. That said, I’m not 100% sure how much memory overhead I have in this setup vs just running the 64-bit userland.
Technically there’s a very interesting “hybrid” abi you can try - 32 bit pointers on a 64 bit ISA. The main benefit is the extra registers that get unlocked in 64-bit mode while keeping memory usage lower. It seems there’s an attempt to do this in Debian [1]. From what I read it’s not a big win for arm though but could be big for x86 - since the ISA is very different between x86 and x86_64. Probably doesn’t matter when ram is so cheap nowadays.
Separately I’ve now got a serious case of buyers remorse. I just bought the 4gb version last week. I would have easily shelled out an extra $20 for double the ram.
NightMKoder's comment was referring to the x32 mode where an application uses the x86_64 ISA and registers but with a 32-bit ABI (ILP32 - int, long, pointers are 32-bit). The kernel is full 64-bit, so there is no overhead like in PAE where the kernel is 32-bit and needs to temporarily map physical addresses above 4GB (using highmem support). On the application side, x32 has some benefits over x86_32 since it has access to a larger register bank and wider registers if needed (e.g long long).
On ARM, there are patches implementing ILP32 with an A64 ISA (AArch64 mode and registers). While there probably is an improvement over the A32/T32 ISA, the benefit is not that great since the 32-bit ARM architecture already had 14 general purpose registers (AArch64 has 30).
The reason the arm64 ILP32 hasn't made it into mainline Linux is not fully justifying the software porting/maintenance cost over the benefit it brings. Full 64-bit software stack is in general a better option unless you run in a very constraint environment but, in that case, an M-class ARM CPU is probably a better option anyway.
Has anyone quantified these performance differences between full 64 bit vs 64 bit OS but 32 bit memory per process?
At 8 GiB RAM, allowing a single process to "only" use half of the available RAM really doesn't seem too limiting for most workflows. Especially considering that there's 4 CPU cores. If it's a notable hassle-free performance improvement it seems like a no-brainer for most use cases.
Depending on where the kernel/user memory map split is, each process may only be able to map 1-3GB. Unless things have changed and all mappings share a single address space?
I bought a 4gb Pi4 today, got the shipping confirmation email from the supplier, then straight away got another email from the same supplier launching sales of the 8gb one... talk about timing.
I really wish Microsoft would release a version of Windows 10 (the full version, not IoT) for the Pi 4 so developers could easily test their apps on an actual ARM device. Being able ship a Raspberry Pi 4 preloaded with Windows 10 to developers could bring a huge boost in the number of apps that support Windows on ARM.
The lack of suitable USB driver for the USB A ports is the main holdup to any reasonably usable setup (with varying workarounds which each have their own limitations) followed by the GPU driver - neither are easily fixed unfortunately.
What sort of applications do you find the Pi to be lacking? The Pi has been a gateway drug for Ubuntu and CLI applications for me and it hadn't crossed my mind that this could go the other way.
I think the previous poster wasn't writing about applications lacking on the Pi but about having ARM developer devices for Windows 10 developers at a low cost.
I am strongly hoping that the 8GB Rpi + NexDock would be a good low cost laptop for web browsing, text editing and terminal.
I've been turned off from investing in a new laptop for a variety of reasons:
- Apple has made questionable hardware design decisions lately. If Louis Rossmann is to be believed, the T2 chip + secureboot situation is just as bad as the old keyboard situation. It sucks having no say over the hardware.
- Apple's newest Catalina OS has been strangely unstable for many (including me) and is headed in a strange direction. It sucks having no say over the software.
- Intel's chip vulnerabilities are too many and too confusing to keep track of. While their chips are advertised as "Quad core @ 2.4Ghz", I wonder what the actual performance is when all the kernel protections are enabled.
- AMD's Ryzen 4000 products look amazing, but (almost?) none of the laptops are 13" with high quality screens.
A "laptop" composed of 8GB Rpi + NexDock would allow me much greater control of the hardware and software.
My laptop's display panel is almost 10 years old, and has about 96% AdobeRGB coverage. 100% sRGB should be considered barely above shit-tier. Probably fine if you don't do any photo or video editing ever, I guess?
If you're doing photo/video editing for web, sRGB is your target anyway. This is a $530 laptop btw. Which 10 year old laptop had 96% Adobe RGB coverage, out of curiosity?
My screen is the AUO B173HW01 v4, which was shipped with several laptops from that era. The spec [1] was actually finalized in 2009 and that's when my EDID information says the display was manufactured, so "10 years old" was actually a bit conservative it seems.
The spec says it's supposed to have 90% NTSC coverage. I've had it calibrated with a colorimeter, which tells me it has the following properties:
* 89.0% DCI-P3 coverage
* 97.8% AdobeRGB coverage
Here's the graph of its gamut vs sRGB.[2] I edit a lot of photos (though never professionally), and honestly I would never edit on an sRGB screen. Plus, what about HDR video? It's becoming extremely common these days.
That's a 6-bit TN panel, with a decent gamut. Color accuracy will be so-so, most likely. (What's your delta-E?) Viewing angles are not great either. [1] An 8-bit IPS panel with 100% sRGB would likely be superior for most purposes. It's also a 17.3" screen vs. 13" which was what was being discussed.
I did the photography and color grading for a published coffee table book on a monitor with less than 97% AdobeRGB. I would have liked a better monitor, but it worked out....Lightroom and Photoshop have nice tools to help.
If you're doing HDR video editing, get an HDR display. If you're doing professional video/picture editing, get one with a wide-gamut and good color accuracy. For most people, 100% sRGB will be pleasing and accurate for most purposes.
> That's a 6-bit TN panel, with a decent gamut. Color accuracy will be so-so, most likely.
For sure, I grant that there are much much better panels these days. I would get something better if I was purchasing it today. But I would find it difficult to give up the wide gamut with all the work I do with photos (personal and travel photos).
The profile says the white point ΔE 2000 to daylight locus is 21.86. I don't understand exactly what this is, but I know it isn't very good. The viewing angles are indeed quite bad, though actually a bit better than most other TN panels I saw during that era.
My point wasn't that this was the best screen you could get, even at the time, just that an sRGB screen isn't impressive at all - Apple's MacBooks hit 100% sRGB in 2014 at the latest, going by one I profiled with my colorimeter. MacBook Pros have been at ~100% Display-P3 for years now, and HP has a laptop at 100% DCI-P3.
> For most people, 100% sRGB will be pleasing and accurate for most purposes.
I just don't know that this is true. HDR video is becoming a bigger and bigger thing. Image codecs that support 10 bit+ color depth are becoming widespread; the iPhone takes photos with HEIC by default. I assume that eventually consumers are going to want to be able to view movies in HDR on their laptops, if we aren't there already.
ΔE 21.86 isn't just "not very good", it's unusable for color work. (I would call it "shit-tier", in your parlance). Why did you even use a colorimeter and not calibrate? Sorry to say, that's a much bigger issue than being able to show more intense greens. You're just wrong. Someone asking for a "good screen" should be entirely satisfied with a screen with 100% sRGB gamut, a good ΔE (<3) and off-axis viewing angles, much more so than just a screen with a larger gamut.
(And yes, sure go get a laptop with 100% DCI-P3, HDR 600, etc. I'll wait...)
I looked into this for the rpi4 a while ago and unfortunately the PiTop stuff won’t actually fit a rpi4 due to how it’s designed... at least that was what I was able to work out, I would love to be wrong.
Raspberry Pi 4s are powered via USB C PD. You can buy 50 Wh USB C PD battery bricks for $20 (or 100 Wh for $40). RPi4s pull between 2 W and 5 W, so you’re looking at a long, long battery life with a cheap travel battery.
> Apple has made questionable hardware design decisions lately. If Louis Rossmann is to be believed, the T2 chip + secureboot situation is just as bad as the old keyboard situation. It sucks having no say over the hardware.
His claims were significantly overblown. If your T2 dies, yeah, your data is toast, because the disk keys live in the T2. But the same applies for your SSD dying itself, and the T2 is not especially or uniquely unreliable. Even if it dies as frequently as an SSD, the failure rate is still well within the realm of normal/acceptable. He’s blowing it up for clicks.
The keyboard on the other hand was a documented design flaw. The T2 works fine.
It also significantly improves platform speed and security (and they even left a platform security escape hatch if you want to disable the boot security and run Linux, something they didn’t have to do).
His post was inflammatory clickbait, nothing more. This sort of Apple-flaming, choir-preaching basher content has been extremely popular regardless of merit since long before the T2 or even the removal of the headphone jack from the phones.
Unlike GP, his life is also about hating Apple in every way possible. Just looking at his YouTube channel for about 4 seconds makes it pretty obvious that he isn't even trying to be fair.
It's just like the keyboard in a sense. Before the keyboard problems were a wildly-acknowledged issue I had to listen to people online argue that the field failure rate was perfectly fine, when they didn't know either the actual failure rate nor know anything about reliability engineering.
Maybe the failure rate is fine, maybe it's not. All we know right now is that they added an additional single point of failure. The _very best case scneario_ for Apple is that they understood they were adding a single point of failure, ran the math to find what the increased expected failure rate would be (basic statistics say they definitionally raised it), and decided they didn't give enough fucks to do anything differently.
I'm perfectly fine with with a person who frequently deals with the people experiencing that failure doing some old fashioned New York shittalking towards that design decision.
Maybe you're right, maybe you're wrong. I can just pick the second option. This is like digging up an old comment and saying "it's like when you were wrong that time".
> All we know right now is that they added an additional single point of failure.
Do we know that? TPM and encryption are also single points of failure, for your data at least. Apple didn't add that to increase the failure rate, they added it to have more control and to offer features that differentiate them from most other OEMs using COTS components for this.
They did the same with phones, TouchID and FaceID, the secure enclave, things that gave Apple enough control to turn the iPhone into the gold standard for keeping your data secure on a phone. It stands to reason that they want to do the same on laptops.
Sure. I prefer to live in the possibility space of uncertainty. Until something comes along to reduce that uncertainty.
>TPM and encryption are also single points of failure, for your data at least
Indeed they are.
>Apple didn't add that to increase the failure rate,
That's never the goal, but every design decision shifts the likelihood of certain failure types up and down. As I said before, the best case scenario is that Apple ran the math/experimentation to quantity it and was happy with their balance of likelihood of various failures.
You can argue until you're blue in the face Apple chose well. That's fine, doesn't mean someone who is close to the pain of end users experiencing failure is acting in bad faith by disagreeing with you.
Your last paragraph is just an ends justify the engineering-means argument. I don't find those interesting anymore.
> You can argue until you're blue in the face [...] doesn't mean someone who is close to the pain of end users experiencing failure is acting in bad faith by disagreeing with you
Since you didn't actually provide any evidence either way (and nobody did to my knowledge) it stands to reason that indeed your only point is to argue.
There's no reason to assume Apple introduced something like the T2 chip for any other reason than to offer genuinely useful features to users and profit from it. Genuine criticism should be based on more than "maybe I'm right, maybe I'm wrong". Otherwise it's just shilling. ...And losing an argument, whether you found it interesting or not.
Care to see how others read your "brilliant" maybe this, maybe that "argument"? All manufacturers should offer no features to have no point of failure. Any single addition to any product can only be another point that can fail. Thus it should be assumed by default that this must be the intention of every manufacturer. It even begs the conclusion that all of them are the same so singling one out is purely grudge based.
Yeah, it's weird to see the comment criticizing the impact of Intel's CPU patches and Ryzen 4000 systems and then talking about using a RPi4 as an alternative. I don't have experience with the RPi4, but in previous versions the CPU and IO performance was terrible compared to even a mid-range laptop, so I assume it still is. If you use the system for any kind of work (even if it's just web browsing), an RPi system is going to have an awful ROI compared to a 'regular' laptop.
Any relatively cheap (like $300) laptop is going to kick any RPi to the curb. And that gets you the display, the keyboard, the touchpad, and storage, all in the base price.
Beacuse if I spend $1000 (even $2000) I want to get that much value for that purchase. Or I can spend significantly less and be that much less married to that purchase.
Between Apple's missteps, Intel's not ideal products/roadmap, and AMD's lack of uptake, the current products do not seem to offer the value for their costs.
The 2015 13" rMBP was by far my most favorite laptop I have ever owned. I have never regretted spending what I did on that machine.
If a Raspi + Nexdoc costs $300, I'm not going to be very broken up over replacing it in a year or two. I cannot say the same for a $1000-2000 laptop.
Why not just continue to use the 2015 13" rMBP? It outclasses a Raspi + Nexdoc in every way I can think of, and if you still have it it's free :D
Since 2015s doesn't have the T2 nonsense and you were fine with linux on the rpi and there's no T2 stopping you, then you can just wipe it and install what you want.
I really have a hard time understanding the appeal of the NexDock.
With a phone processor it will still be ARM, you won't get a great laptop. But it already costs $250, for that price you can get an entry-level laptop that will probably perform in a similar way while still leaving your phone independent.
(1) Phones have your data, settings, apps, etc. on them. Syncing things sucks.
(2) The ARM in your phone is probably faster than a $250 laptop CPU. It can be plenty fine for most tasks.
(3) Many of us aren't CPU-bound in the first place. The only thing holding our phones back are software, keyboard, and display.
Done well, it'd be awesome. I definitely want something like the NexDock. However, I'll probably wait for V2 or V3. I think V1 will be good for people who would use it like a Chromebook.
My V5 NexDock would look like an actual dock, and plug my phone into:
* a pair of 4k monitors
* a keyboard, a mouse
* a pair of external HDDs (keeping my data backed up in RAID, and holding larger files)
I've been thinking the same thing lately, though I didn't know about the NexDock project until I read your post. My dream would be something like a modern-day ThinkPad X61 or ThinkPad X220 for its durability and its keyboard. However, it would have an upgraded monitor (I would like something better than 1024x768 or even 1333x768), and it would be connected to a Raspberry Pi 4. Any time I want to upgrade, I just upgrade the Raspberry Pi while I keep my laptop chassis.
Just about to grab 2 Pi 4 GB models, will put on hold for price adjustments..
This is really exciting, Pi 4 8GB with support for booting from USB SSD makes a cheap workstation option, many other possibilities as well as server in the garage to replace some VPS in cloud? (for general purpose workloads, can the CPU keep up if 8GB memory is saturated, consider general purpose use cases, memory-bound workloads may work well).
Side note: My original Pi 1 is still running Pi-Hole 5.0 fronted by nginx + php-fpm on Raspbian 10, working well for the home LAN ;-)
> The BCM2711 chip that we use on Raspberry Pi 4 can address up to 16GB of LPDDR4 SDRAM, so the real barrier to our offering a larger-memory variant was the lack of an 8GB LPDDR4 package. These didn’t exist (at least in a form that we could address) in 2019, but happily our partners at Micron stepped up earlier this year with a suitable part.
All other SBCs in this market segments have also been capped at 4GB - I'm hopeful about this leading to larger memory options from other vendors as well before long.
Looking at Micron (who probably tickd the box on price to make this happen), they can do 16Gb modules. So that may well happen. However the power aspect as already indicated from the 4Gb to 8Gb, considerations will prove more interesting.
Biggest gain though is more focus upon 64bit, which goes hand in hand with that extra memory.
This is a case where you should read with care, and notice the difference between Gb and GB. The chip from 2013 that you mention, has a capacity of 8Gb = 1GB. Chips with a capacity of 16Gb are indeed very common, because manufactures put 8 of them on a RAM-module with a capacity of 16GB.
RAM chips are and have always been measured in kibibits and gibibits since long before those two words existed. RAM modules are measured in gibibytes. In a desktop computer you typically use 8 XGb chips to make a single XGB module.
The confusion comes when you blur the distinction between chips and modules as they do in mobile. They either use single package modules created from multiple chips via 3D die stacking or they use single chip modules.
Thanks for the info. Clearly all my experience in this area is from the perspective of a consumer buying entire modules. I can't say I've ever dived down to the individual chip level.
So I just got a fun project working with a Raspberry Pi 4. I have it generating a slideshow of fake images using the impressive BigGAN model, and displaying those on a 7" touch screen.
It's just for kicks; I wanted a little AI daydreaming on my desk :) Also to see if it was even possible, given how large the BigGAN model is. But lo and behold the 4GB Pi is _just_ enough to run it. I'm half tempted to buy this new 8GB version so I can run at higher resolutions ... but I think the precompiled Tensorflow package for RasPi is 32-bit anyway so might not matter.
I've been reading a lot about tiny websites on HN (e.g. https://bearblog.dev). Can we use these Raspberry Pis to host them? Are they performant enough? Is setting things up newbie-friendly?
Having your own little corner on the internet that no one can shut down sounds pretty cool.
When our dev team was just starting out, we ran the business gitlab instance on a raspberry pi 2 model B. It took a few kernel Params to work with only 1GB of RAM, but it got there.
His name was Boris and he was the company’s MVP for several years running :)
Boris was quite reliable, the only issues we had was when we occasionally ran out of disk space and had to clean up some files... Other than that, he kept quite and solemnly did his work. We actually had a separate CI/CD machine to run gitlab runner jobs.. So, boris just needed to run the gitlab web server. Eventually, once we we had more than a handful of staff, we moved to more substantial hardware. We're now running a multi-node on-premises kubernetes cluster (Boris's big brother is putting in work!). Boris is still functional, but he's retired now. He's now part of the business history to inspire the next generation.
Yes, they are definitely fast enough for simple pages to a few thousand daily visitors. I've hosted dynamic websites from home using much more modest hardware and eventually used a first model Raspberry Pi for it.
Eventually I moved my sites to a VPS out of convenience. My ISP can't guarantee a static address with my current plan (although I've found it is static in practice so far), and the additional cost of a plan where they can exceeds the cost of renting a VPS.
If you have an ever changing dynamic IP allocation, you can still use Dynamic DNS services to update DNS records. I use GoDaddy for my names and AFAIR they provide this service, but there are also free providers that'll give you a subdomain, like https://dyndns.org. This requires some additional setup, because you need to update your IP with the dynamic DNS API regularly, for example via a cron job.
As for setup, if all you need is static pages you can use Raspbian and just install Apache or Nginx, put your files under /var/www/html and on your router forward some port to port 80 on your Raspberry Pi.
Back when the entire web was like this, simple websites would be hosted on 50-200 MHz single core machines with RAM in the megabytes. So I don't mean to come off as confrontative but if you can't host a text-based website on a quad-core 1.5 GHz machine with 8 GB of RAM you're doing something seriously wrong.
As for setting things up, if you can apt-get install apache on any generic Linux box, you can do it on a Raspberry Pi.
I host a static website on a Raspberry Pi 3 and it's been running very well for about 3 years. It's static HTML and it also serves files. It's hosted by a provider with a "Pi Cloud". https://www.mythic-beasts.com/order/rpi
The website is https://www.piwheels.org - it's a Python package repository for Raspberry Pi users
It's served over 20M downloads, which is 70TB. Currently serving 1.2M downloads a month (4.4TB).
A cheap VPS would be easier, better and about the same price. But piwheels is a bit of a dogfooding project. But it works!
They are completely performant enough to serve small static websites, if you have low traffic (it should be confortable with a few connections per second).
I think another limiting factor to look at is your upload bandwith. If it's too low, then it will saturate and your conection will be very slow. Some ISP provide asymetric bandwith (much higher download than upload), so that's not the best case. If you have fiber, then you're absolutely good to go.
If your blog is featured on HN first-page, then it certainly won't handle the load with a RPI. But for most blogs, it should be enough.
It should be comfortable with a lot more than a few connections per second. I've just benchmarked my Pi4 running Nginx on Raspbian from my laptop over wifi, running Apachebench and obtained 322 requests per second with a 34kB html file (the HN homepage).
Running ab on the Pi itself, the CPUs can deliver over 3k rps though of course the network subsystem is an important part of the story.
At this point, I would much rather use any x86 based laptop instead.
Their price-to-performance ratio's have significantly increased since the first RPI was introduced and for a 100$, you get a much better package.
So, unless the ARM ecosystem grows to allow for pluggable, extensible and better performing solutions than the latest entry-level celeron or ryzen based systems, I am no longer investing into it.
Can you recommend any specific laptops at around that price point that have superior performance? Would these be new or used? And any idea how their power consumption compares to a Raspberry Pi?
Ya, I'm really confused. Does the poster want a $75 laptop? That's called a smartphone, and last I looked the operating system and specs are pretty terrible.
You can definitely get OK used laptops for $75, that's why I asked that particular clarifying question. I'm wondering if they have any specific models or deals in mind, ideally with a link to an online retailer.
Think that'd make a better server than a Raspberry Pi 4 for the price? It's nice that it includes a built-in battery backup, plus screen/mouse/keyboard for initial setup and later debugging. On the down side, it takes up more space.
I do find that RPI life comes with a lot of add-on costs. Battery pack, input devices, display, SSD, etc. It does start to add up if you aren't using the RPI in headless mode anchored to a power cable.
However, if price is the #1 concern, it's tough to beat the RPI. I checked Amazon and the cheapest 8GB RAM laptop was $239. Or maybe a used one on eBay for ~$175.
But overall this announcement is very impressive. I foresee the SBC "arms race" heating up. I bought a Jetson Nano thinking it was overkill, but it seems the bar keeps going up...
For certain applications, the form factor and power usage are both relevant. So if you really need a small low power computer at that price point, you don't have a ton of options (although there's always the RockPro64 [0]).
Once any of those requirements is loosened the other options start to look way better. So if you want a low power ARM computer and aren't concerned with the size as much, you can always get an old used Chromebook instead - which is well worth the money, but only IF you need any of the extra stuff it comes with (monitor/keyboard/mouse/battery/storage/camera/etc etc...). If you just want raw compute, then you can just buy an old x86 box like you say, but that will cost you in power and size.
I think some people just like to tinker on these little gizmos specifically, though. You read blog posts about running Kubernetes on Pis, and the immediate question is "why would you do that rather than just virtualizing stuff on an old x86 box?" The answer is really just "well, it's fun to work with little computers" which is presumably a factor for a lot of people.
I was recently wondering if there's any kind of single board computer that would be suitable for just shoveling data on the network at close to 10gbps. If you have something that basically just sendfile()s data, an average desktop PC from 5 years won't pull a sweat saturating 10gbits, barely reaching two digit CPU%.
Are there are any SBCs that would have suitable IO for this? The pi4 has USB 3 which internally is connected via PCIe, but the question is whether it could handle eg SATA and Ethernet on it. Not that I actually need this for anything, it would just be cool to have a small case with 1TB SSD that you can plug into a switch and deliver hundreds of MB/s. With 8GB RAM you could even make good use of the fs cache, depending on usage pattern.
My experience is that these small computers don't even do well at 1Gbps. When I worked on Google Fiber, we really wanted to implement a speed test between our routers and some trusted server, but the chipset we used couldn't generate data fast enough. (We could route at 1Gbps because it took a hardware fast path, but the software-only path wasn't good enough to provide useful speed tests.) I have subsequently tested RPi4s at another ISP, and again, we couldn't get a gigabit out of them.
In fact, I have never gotten 10Gbps out of any software-only path. Routers can do it, but it's hardware accelerated and won't generate packets at that speed. I did recently get 7Gbps out of normal Internet usage, though, with a consumer-grade 10Gbps NIC on an Intel system. Pretty decent, but not 10Gbps.
Much optimization is necessary to make 10Gbps a thing. It's not a focus for $35 single-board computers. Those things are just obsolete phone chipsets in a desktop form factor; don't expect much out of them.
But it should be about the buses these boards expose shouldn't it? Even the pi4 still has Ethernet over USB which is unnecessary overhead. Getting access to the PCIe lanes directly would be interesting for this. I just researched a bit and it looks like there are boards with SATA port from hardkernel that completely saturate the 1gbit port, so the question is how much faster they could go, and whether it uses a USB bridge too.
In general, saturating 10gbit on any x86 from the past couple years isn't a problem. Doing it with a single TCP connection is a bit tricky and requires some tweaking, and jumbo frames help too, but a simple threaded server pushing out data via sendfile to multiple clients easily gets the job done.
But yes, I agree that this is not a useful use case for SBCs, one of these things I'd just want to build because I can, not because I should.
230 comments
[ 3.7 ms ] story [ 278 ms ] threadThis plus the pinebook pro, I really hope that it will help the software availability on armv8 to skyrocket!
Also I really hope that someone will come with a new case to use it as tablet and laptop. Like a mix between the old pi-top and a surface book (or a PineTab).
If anyone has experience developing on the raspberry pi 4 (4GB), I really want to hear from you, especially regarding the processor performance.
I know that it may improve in the future, but as of today, 2GB of Ram for the PineTab...
Moreover, I disagree with your suggestion that 2GB is useless in a tablet. It really depends on the use-case, and I have rarely seen a capability range similar to the pineTab in a tablet from-factor.
Have a look at the RockPro64, for instance: https://www.pine64.org/rockpro64/
The 4GB version is about $80, but offers PCIe 4x, 128Mbit Flash, Gigabit Ethernet, USB-C Host With Video Out, and an eDP Port, among others.
It's true that the A64 family only goes up to 2GB, but the rock64 family goes to 4GB for $45. All offer valid tradeoffs to the Pi family :)
I think they technically go up to 3GB, but making a board with that is messy.
EDIT: I haste to add that having the full 64-bit Ubuntu 20.04 was what made this interesting to me.
Talking about the IO, did you experience failure (lose of work) due to the unreliability of microSD?
I've been developing on the platform for a few months now. Though it's difficult to compare performance between architectures, the new BCM2711 compares favorably in CPU bound tasks to Intel's Cherry Trail Atoms from early 2016, primarily the Atom x5-z8350.
The Pi 4 also overclocks easily from 1.5 GHz to 2.0 - 2.1 GHz, where it exceeds Atom performance in some cases. The A72 cores are a huge leap in performance compared to the A53 cores of the RPi 3. These are architecturally "big" cores, as compared to the more power efficient and lower throughput "little" A53 cores of the previous model.
If you have an application you'd like to ship on a Raspberry Pi image, especially if you want to address all 8 GiB from a single process, check out Buildroot.
With the right PiHAT this makes raid FS much nicer.
Generally though it is like a "full computer" so I don't know what I would buy one of these for, I currently have 3 Pis that have been on for a few years now continuously they're running web servers/automated tasks mostly Pi Zeros and a 3B+.
I think it's crazy 8GB is becoming like the new 4GB... 32GB is almost not enough personally.
Also I would much rather do technical support for Pi, than some old junk in questionable state of disrepair.
And then there's the cooling and power consumption difference between a 10 year old system and a RPi. It's not even close.
Like this kind of stuff: https://www.ebay.co.uk/itm/HP-Lenovo-Intel-i3-Dual-Core-i5-Q...
Is plentiful and easy to buy, again - ships anywhere in the EU.
As for the power consumption - I have to disagree again, I have an SFF like this and in idle it uses about 10W of power, 40W under load.
Rpi4 4GB uses 1-3W in idle, and up to 15W under full load. Yes, the numbers for the Pi are lower, but I wouldn't say "it's not even close". It's close enough that I would absolutely consider getting the larger PC instead.
Our used cars go for peanuts compared to used cars in the mainland for example.
I am often trawling the UK ebay for tech stuff because it's so much cheaper than Blocket (a craigslist-like) here in Sweden.
PS: very very pedantically and technically, UK isn't an EU member any longer. Although there is some limbo period.
I was somewhat intrigued by the use of the term mainland, but it certainly has some debate behind it: https://english.stackexchange.com/questions/3652/how-to-refe...
Yes second hand cars in the UK are cheaper, factors being that there has been a consumer rise for a while, seeing more second hand cars. This with the island effect makes for dispersal of those second hand goods more curtailed and for example you may be able to get cheaper cars in Germany second hand if the geography curtailed those easily shifting to nearby countries who have had less of a consumer drive.
Which is why my friend in Bulgaria will buy a second hand car in the UK, drive it back, sell it for twice what he paid for it. Then in Bulgaria Walnut wood is about 1/5th of the price of what it demands in the UK. But most opportunities in life are going the extra mile.
Yes UK is still technically an EU member during the transition period. How exporting/importing of cars pans out, well as long as the we don't end up with trade loop-holes - then I'll be happy and by that. Say EU has import tax of 30% on UK cars, new or second hand. Yet no such taxes for imports from say canada. Then the UK has no tax import on cars from UK to canada, so would be possible to export to Europe, via Canada to skirt such import taxes. That kinda thing worries me as environmentally, not good. But could happen.
All I do know though is that the current trade negotiations seem like a Monty Python playbook:- https://www.youtube.com/watch?v=CK0cUv3ba-o
something to do with steering wheel being on wrong site making them useless to all of EU other than for parts.
I believe it stems from a tradition of not disposing of things, and instead selling them on (via "car boot" sales) or handing them down. I believe (and this is pure conjecture) that it's a recent thing for brits to dispose of things with such frivolity, and this is somewhat echoed in the car-boot sales of late being chock-full of brand new mostly chinese cheap crap.
And maybe if you have an overabundance of second-hand goods then maybe the price goes down as all of the friction to sell even things considered "tat" are gone.
And most of the things we would consider "tat" are not "tat" to everyone else, for instance, I bought my first computers for 5gbp from a boot-sale, for such a token amount of money it is likely a person would not have sold it another way. Thus, I get a cheaper computer than I would have had to pay otherwise.
For $85 I can only find core2duo/quad 4gb RAM and spinning rust so there's no way that is a viable alternative for a RPi.
Maybe UK is an outlier or Austria is, I don't know.
The snide comments like the GP's are due to them living in the american bubble.
EDIT: also note the comment you're replying to is using a traditional workstation "slab" as an example. I'd rather use a small RPi that frees up more desk space than buy something that will make my work environment at home even more cramped!
Are you looking for broken screens? If not, $200 sounds reasonable enough to me.
Curious - any thoughts on why?
I am free to buy and sell to and from any country in the world, the issues is not everyone is willing to deal and pay for international shipping which ends up quite high for a cheap SFF tower system.
In principle, shouldn't one of the consequences of the EU project be that intra-EU shipping rates should, for the same distance, be identical between within the same country and between EU countries? (Of course, I realise that outcome may not yet have actually been achieved in practice.)
One would hope that if they are successful in improving those transport links, shipping costs will reduce as a result.
For desktops, it shouldn't really matter. One quite likely already has a spare USB keyboard with the right layout, and if one doesn't, it shouldn't cost much to buy one (whether new or second-hand).
Big markets like US, UK and Germany have low prices on consumer products and in small markets like Austria and BeNeLux countries consumer products cost more.
Same way hard drives a few years back increased in price, supply and demand as a factory went out of action for a while, curtailing market supply. Also why external USB drives became source of HD's as they had a larger supply buffer as less popular at the time and more in warehouses/stores.
The PI kicks ass if you want to play with hardware and actually use it's pins to build robots or whatever, or you want to use it's amazing camera and libraries. For pure compute its not a great deal.
And the used market is great. My favorite for a very small desktop are the Lenovo USFF devices like the M92P, M93P, etc. You have to search a bit, but you can find 8GB models with an i3/i5, small SSD, and Windows 10 for $110 or so on eBay.
Not as small as a Rpi, but pretty compact: https://cdn.shopify.com/s/files/1/0036/4806/1509/products/m0...
Which is a ridiculous waste of energy... :( you notice it too now using Microsoft Teams (Electron) and Google Meet versus Facetime. Eats your battery in no time.
On my Intel NUC, which is a few years old, I have to use an extension[1] which forces YouTube to serve h264. Otherwise it's a loud laggy mess.
[1]: https://addons.mozilla.org/en-US/firefox/addon/h264ify/
There is no Linux like solution for android phone that runs Xserver and OpenJDK reliably. Samsung DEX emulation was the only usable solution, but that was killed off. Termux can not run OpenJDK, some weird issues with crashing.
Also most devices do not have enough cooling capacity, max performance can be only delivered in short bursts before CPU gets throttled down. If you hack it to run at full speed for hours, it freezes.
PI is simply the only option at this size, with enough CPU power and good Linux support.
When you want Xserver, you just do so from Android vnc viewer to 127.0.0.1:5901 or so.
I find that its pretty difficult to use 4gb of ram on a pi practically, and its often bottlenecked by I/O. It might just be my SD card, but executing trivial commands sometimes takes forever and reminds me that I'm in a pi. I've heard ssds can improve speeds a lot, but you still cant boot off usb.
https://www.raspberrypi.org/forums/viewtopic.php?t=274595
https://youtu.be/zVhYvvrGhMU
That's a game changer.
It was not available at launch for the Pi 4, but was released in beta firmware in late September, and is currently part of the critical firmware update.
It works well, allowing separate boot images for each individual Pi on your network.
> It works well, allowing separate boot images for each individual Pi on your network.
could you elaborate a bit more on that?
If you’re netbooting without any sdcard installed, I’d suggest disabling sd polling.
Isn't that a natural capability with PXE implemented at the server keyed off the MAC address?
Something you've always been able to do is to keep just the boot partition on a small SD card and the root on whatever USB device you like. And yes, even with the supposedly best-performing SD cards, they easily become a problem both with performance and failure.
IITC that is still not officially supported (as a beta-test feature?), though many report it works as intended.
In the meantime if people don't want to use a beta feature the boot-from-SD-mount-root-elsewhere option works well too. You could also run it of an NFS share instead of a USB stick, though the performance metrics will differ.
https://www.seeedstudio.com/ICE-Tower-CPU-Cooling-Fan-for-Ra...
https://hackaday.com/2017/05/15/liquid-cooling-overclocked-r...
https://modmymods.com/water-cooling-kits/raspberry-pi-coolin...
Having just spent a bit north of $1k on a new CPU, RAM, and mobo for my desktop, I'm having a bit of sticker shock (in a good way) at how low these RPi prices are.
Doesn't look as flashy as the Flirc case, but cooling performance is supposed to be similar and you can access all the pins.
I have several pi in flirc cases, one is a pihole and one runs CI jobs (slowly) for PhotoStructure, and they don't get too hot even under sustained cpu load.
Flirc cases are great.
https://shop.pimoroni.com/products/aluminium-heatsink-case-f...
I had a passive aluminum block heat spreader on it, and that works if you have good thermal bonding, unless you are really working it for a long duration. The problem is, if you hammer it long enough to throttle, that same mass takes a long time to cool back down.
The fan shim on a naked card just seems to work.
https://shop.pimoroni.com/products/fan-shim
Technically there’s a very interesting “hybrid” abi you can try - 32 bit pointers on a 64 bit ISA. The main benefit is the extra registers that get unlocked in 64-bit mode while keeping memory usage lower. It seems there’s an attempt to do this in Debian [1]. From what I read it’s not a big win for arm though but could be big for x86 - since the ISA is very different between x86 and x86_64. Probably doesn’t matter when ram is so cheap nowadays.
Separately I’ve now got a serious case of buyers remorse. I just bought the 4gb version last week. I would have easily shelled out an extra $20 for double the ram.
[1] https://wiki.debian.org/Arm64ilp32Port
On ARM, there are patches implementing ILP32 with an A64 ISA (AArch64 mode and registers). While there probably is an improvement over the A32/T32 ISA, the benefit is not that great since the 32-bit ARM architecture already had 14 general purpose registers (AArch64 has 30).
The reason the arm64 ILP32 hasn't made it into mainline Linux is not fully justifying the software porting/maintenance cost over the benefit it brings. Full 64-bit software stack is in general a better option unless you run in a very constraint environment but, in that case, an M-class ARM CPU is probably a better option anyway.
At 8 GiB RAM, allowing a single process to "only" use half of the available RAM really doesn't seem too limiting for most workflows. Especially considering that there's 4 CPU cores. If it's a notable hassle-free performance improvement it seems like a no-brainer for most use cases.
https://medium.com/@matteocroce/why-you-should-run-a-64-bit-...
Linux x32 ABI, the free 5-8% performance upgrade!
[1] https://www.youtube.com/watch?v=Mw-ravkg67k&
From René Rebe of T2-SDE pre-Gentoo source-based Linux distro fame.
I occassionally watch it, and the website is deceptive because the latest builds are much newer than the website suggests.
https://www.youtube.com/watch?v=is6qQujk6uU
I wouldn't be surprised if Microsoft was also silently working on it.
The lack of suitable USB driver for the USB A ports is the main holdup to any reasonably usable setup (with varying workarounds which each have their own limitations) followed by the GPU driver - neither are easily fixed unfortunately.
I've been turned off from investing in a new laptop for a variety of reasons:
- Apple has made questionable hardware design decisions lately. If Louis Rossmann is to be believed, the T2 chip + secureboot situation is just as bad as the old keyboard situation. It sucks having no say over the hardware.
- Apple's newest Catalina OS has been strangely unstable for many (including me) and is headed in a strange direction. It sucks having no say over the software.
- Intel's chip vulnerabilities are too many and too confusing to keep track of. While their chips are advertised as "Quad core @ 2.4Ghz", I wonder what the actual performance is when all the kernel protections are enabled.
- AMD's Ryzen 4000 products look amazing, but (almost?) none of the laptops are 13" with high quality screens.
A "laptop" composed of 8GB Rpi + NexDock would allow me much greater control of the hardware and software.
Check out the Xiaomi RedmiBook 13 (100% sRGB coverage).
The spec says it's supposed to have 90% NTSC coverage. I've had it calibrated with a colorimeter, which tells me it has the following properties:
* 89.0% DCI-P3 coverage
* 97.8% AdobeRGB coverage
Here's the graph of its gamut vs sRGB.[2] I edit a lot of photos (though never professionally), and honestly I would never edit on an sRGB screen. Plus, what about HDR video? It's becoming extremely common these days.
[1] http://www.yslcd.com.tw/docs/product/B173HW01%20V.4.pdf
[2] https://cloudflare-ipfs.com/ipfs/Qmaq6553VxMazANuLiFh95aW6fA...
I did the photography and color grading for a published coffee table book on a monitor with less than 97% AdobeRGB. I would have liked a better monitor, but it worked out....Lightroom and Photoshop have nice tools to help.
If you're doing HDR video editing, get an HDR display. If you're doing professional video/picture editing, get one with a wide-gamut and good color accuracy. For most people, 100% sRGB will be pleasing and accurate for most purposes.
[1] http://www.panelook.com/B173HW01%20V4_AUO_17.3_LCM_overview_...
For sure, I grant that there are much much better panels these days. I would get something better if I was purchasing it today. But I would find it difficult to give up the wide gamut with all the work I do with photos (personal and travel photos).
The profile says the white point ΔE 2000 to daylight locus is 21.86. I don't understand exactly what this is, but I know it isn't very good. The viewing angles are indeed quite bad, though actually a bit better than most other TN panels I saw during that era.
My point wasn't that this was the best screen you could get, even at the time, just that an sRGB screen isn't impressive at all - Apple's MacBooks hit 100% sRGB in 2014 at the latest, going by one I profiled with my colorimeter. MacBook Pros have been at ~100% Display-P3 for years now, and HP has a laptop at 100% DCI-P3.
> For most people, 100% sRGB will be pleasing and accurate for most purposes.
I just don't know that this is true. HDR video is becoming a bigger and bigger thing. Image codecs that support 10 bit+ color depth are becoming widespread; the iPhone takes photos with HEIC by default. I assume that eventually consumers are going to want to be able to view movies in HDR on their laptops, if we aren't there already.
(And yes, sure go get a laptop with 100% DCI-P3, HDR 600, etc. I'll wait...)
how do you plan to power the Pi? seems this would more be a 'portable' than a 'laptop'
Also - there is the 'pi top' kit if that suits your fancy.
His claims were significantly overblown. If your T2 dies, yeah, your data is toast, because the disk keys live in the T2. But the same applies for your SSD dying itself, and the T2 is not especially or uniquely unreliable. Even if it dies as frequently as an SSD, the failure rate is still well within the realm of normal/acceptable. He’s blowing it up for clicks.
The keyboard on the other hand was a documented design flaw. The T2 works fine.
It also significantly improves platform speed and security (and they even left a platform security escape hatch if you want to disable the boot security and run Linux, something they didn’t have to do).
His post was inflammatory clickbait, nothing more. This sort of Apple-flaming, choir-preaching basher content has been extremely popular regardless of merit since long before the T2 or even the removal of the headphone jack from the phones.
It does not take a genius to see how the cooling design is supposed to work. Him saying that fan is "for placebo effect" is certainly not fair.
Maybe the failure rate is fine, maybe it's not. All we know right now is that they added an additional single point of failure. The _very best case scneario_ for Apple is that they understood they were adding a single point of failure, ran the math to find what the increased expected failure rate would be (basic statistics say they definitionally raised it), and decided they didn't give enough fucks to do anything differently.
I'm perfectly fine with with a person who frequently deals with the people experiencing that failure doing some old fashioned New York shittalking towards that design decision.
Maybe you're right, maybe you're wrong. I can just pick the second option. This is like digging up an old comment and saying "it's like when you were wrong that time".
> All we know right now is that they added an additional single point of failure.
Do we know that? TPM and encryption are also single points of failure, for your data at least. Apple didn't add that to increase the failure rate, they added it to have more control and to offer features that differentiate them from most other OEMs using COTS components for this.
They did the same with phones, TouchID and FaceID, the secure enclave, things that gave Apple enough control to turn the iPhone into the gold standard for keeping your data secure on a phone. It stands to reason that they want to do the same on laptops.
Sure. I prefer to live in the possibility space of uncertainty. Until something comes along to reduce that uncertainty.
>TPM and encryption are also single points of failure, for your data at least
Indeed they are.
>Apple didn't add that to increase the failure rate,
That's never the goal, but every design decision shifts the likelihood of certain failure types up and down. As I said before, the best case scenario is that Apple ran the math/experimentation to quantity it and was happy with their balance of likelihood of various failures.
You can argue until you're blue in the face Apple chose well. That's fine, doesn't mean someone who is close to the pain of end users experiencing failure is acting in bad faith by disagreeing with you.
Your last paragraph is just an ends justify the engineering-means argument. I don't find those interesting anymore.
Since you didn't actually provide any evidence either way (and nobody did to my knowledge) it stands to reason that indeed your only point is to argue.
There's no reason to assume Apple introduced something like the T2 chip for any other reason than to offer genuinely useful features to users and profit from it. Genuine criticism should be based on more than "maybe I'm right, maybe I'm wrong". Otherwise it's just shilling. ...And losing an argument, whether you found it interesting or not.
Care to see how others read your "brilliant" maybe this, maybe that "argument"? All manufacturers should offer no features to have no point of failure. Any single addition to any product can only be another point that can fail. Thus it should be assumed by default that this must be the intention of every manufacturer. It even begs the conclusion that all of them are the same so singling one out is purely grudge based.
Lenovo will release a x13 with ryzen 4000 later this year.
If that’s all you want try an iPad2. I think it would make a better “laptop” than a RPI. RPI doesn’t have a battery.
Its fantastic automated tool however; and the live interface is great for irregular checks
If the raspberry pi meets your needs, consider a much cheaper laptop
And no screwing around.
Between Apple's missteps, Intel's not ideal products/roadmap, and AMD's lack of uptake, the current products do not seem to offer the value for their costs.
The 2015 13" rMBP was by far my most favorite laptop I have ever owned. I have never regretted spending what I did on that machine.
If a Raspi + Nexdoc costs $300, I'm not going to be very broken up over replacing it in a year or two. I cannot say the same for a $1000-2000 laptop.
Since 2015s doesn't have the T2 nonsense and you were fine with linux on the rpi and there's no T2 stopping you, then you can just wipe it and install what you want.
With a phone processor it will still be ARM, you won't get a great laptop. But it already costs $250, for that price you can get an entry-level laptop that will probably perform in a similar way while still leaving your phone independent.
(2) The ARM in your phone is probably faster than a $250 laptop CPU. It can be plenty fine for most tasks.
(3) Many of us aren't CPU-bound in the first place. The only thing holding our phones back are software, keyboard, and display.
Done well, it'd be awesome. I definitely want something like the NexDock. However, I'll probably wait for V2 or V3. I think V1 will be good for people who would use it like a Chromebook.
My V5 NexDock would look like an actual dock, and plug my phone into:
* a pair of 4k monitors
* a keyboard, a mouse
* a pair of external HDDs (keeping my data backed up in RAID, and holding larger files)
And have good software / application support.
- Rasbian will be renamed to Rasberry Pi OS
- They're adding a 64-bit variant in addition to the existing 32-bit
This is really exciting, Pi 4 8GB with support for booting from USB SSD makes a cheap workstation option, many other possibilities as well as server in the garage to replace some VPS in cloud? (for general purpose workloads, can the CPU keep up if 8GB memory is saturated, consider general purpose use cases, memory-bound workloads may work well).
Side note: My original Pi 1 is still running Pi-Hole 5.0 fronted by nginx + php-fpm on Raspbian 10, working well for the home LAN ;-)
Something like the profile of the original Ubiquiti CloudKey would be ideal (don't need most of the ports on the Pi)
https://www.ui.com/unifi/unifi-cloud-key/
https://rockpi.org/rockpi4
https://wiki.radxa.com/ROCKPI_23W_PoE_HAT
A carrier board for Compute Module with just POE / ethernet, USB ports and and SD card slot would be good enough too
I literally want same form factor as the Ubiquiti board but with open source support, and sufficient power to charge a phone from the USB port
Orange Pi Zero is along the right lines but doesn't have enough RAM for my needs and I'm unsure about the processor
http://www.orangepi.org/orangepizerolts/
But 256MB limitation is killing me. Other was Wifi performance and reliability (for non LTS version).
https://wiki.radxa.com/Rockpis
Basically want something like https://fccid.io/SWX-UCCK/Internal-Photos/Internal-Photos-36...
I network / POE port, 1 or 2 USB sockets to plug a phone into, SD-Card slot, 1GB+ RAM, A53 (or above) processor
All other SBCs in this market segments have also been capped at 4GB - I'm hopeful about this leading to larger memory options from other vendors as well before long.
Great of RPi to keep pushing the envelope.
Interesting as Samsung developed a 8GB LPDDR4 memory chip back in 2013 - https://news.samsung.com/global/samsung-develops-industrys-f...
Looking at Micron (who probably tickd the box on price to make this happen), they can do 16Gb modules. So that may well happen. However the power aspect as already indicated from the 4Gb to 8Gb, considerations will prove more interesting.
Biggest gain though is more focus upon 64bit, which goes hand in hand with that extra memory.
You have to wonder why they deviated from what I thought was a universal custom.
The custom of using bytes only applies when specifying higher-level computer components.
The confusion comes when you blur the distinction between chips and modules as they do in mobile. They either use single package modules created from multiple chips via 3D die stacking or they use single chip modules.
https://en.wikipedia.org/wiki/Intel_1103
It's just for kicks; I wanted a little AI daydreaming on my desk :) Also to see if it was even possible, given how large the BigGAN model is. But lo and behold the 4GB Pi is _just_ enough to run it. I'm half tempted to buy this new 8GB version so I can run at higher resolutions ... but I think the precompiled Tensorflow package for RasPi is 32-bit anyway so might not matter.
Here she is: https://i.imgur.com/i22METZ.jpg
I've been reading a lot about tiny websites on HN (e.g. https://bearblog.dev). Can we use these Raspberry Pis to host them? Are they performant enough? Is setting things up newbie-friendly?
Having your own little corner on the internet that no one can shut down sounds pretty cool.
His name was Boris and he was the company’s MVP for several years running :)
1. What was the highest uptime Boris pulled off?
2. Why did you have to decommission him? Was there an eventual hardware failure?
Thanks!
But if you want to learn about hosting (web servers, static routing, linux), then DIY. Using at least 2 GB RAM makes life easier.
Eventually I moved my sites to a VPS out of convenience. My ISP can't guarantee a static address with my current plan (although I've found it is static in practice so far), and the additional cost of a plan where they can exceeds the cost of renting a VPS.
If you have an ever changing dynamic IP allocation, you can still use Dynamic DNS services to update DNS records. I use GoDaddy for my names and AFAIR they provide this service, but there are also free providers that'll give you a subdomain, like https://dyndns.org. This requires some additional setup, because you need to update your IP with the dynamic DNS API regularly, for example via a cron job.
As for setup, if all you need is static pages you can use Raspbian and just install Apache or Nginx, put your files under /var/www/html and on your router forward some port to port 80 on your Raspberry Pi.
If your DNS provider has a CLI just use that. I use Route53 and a simple cronjob to update it.
Nginx? Apache? Cloudflare?
Would be nice if the Raspberry Pi kit comes with IKEA-level instructions to do this as weekend project.
As for setting things up, if you can apt-get install apache on any generic Linux box, you can do it on a Raspberry Pi.
Sitting behind a free CloudFlare account mitigates a lot of bandwidth usage too.
The website is https://www.piwheels.org - it's a Python package repository for Raspberry Pi users
It's served over 20M downloads, which is 70TB. Currently serving 1.2M downloads a month (4.4TB).
A cheap VPS would be easier, better and about the same price. But piwheels is a bit of a dogfooding project. But it works!
I think another limiting factor to look at is your upload bandwith. If it's too low, then it will saturate and your conection will be very slow. Some ISP provide asymetric bandwith (much higher download than upload), so that's not the best case. If you have fiber, then you're absolutely good to go.
If your blog is featured on HN first-page, then it certainly won't handle the load with a RPI. But for most blogs, it should be enough.
Running ab on the Pi itself, the CPUs can deliver over 3k rps though of course the network subsystem is an important part of the story.
Got myself a RPi4 last month.
I'll give you folks a heads up next time I'm buying a Raspberry, seems like I can turn my luck into others fortunes.
However, if price is the #1 concern, it's tough to beat the RPI. I checked Amazon and the cheapest 8GB RAM laptop was $239. Or maybe a used one on eBay for ~$175.
But overall this announcement is very impressive. I foresee the SBC "arms race" heating up. I bought a Jetson Nano thinking it was overkill, but it seems the bar keeps going up...
Once any of those requirements is loosened the other options start to look way better. So if you want a low power ARM computer and aren't concerned with the size as much, you can always get an old used Chromebook instead - which is well worth the money, but only IF you need any of the extra stuff it comes with (monitor/keyboard/mouse/battery/storage/camera/etc etc...). If you just want raw compute, then you can just buy an old x86 box like you say, but that will cost you in power and size.
I think some people just like to tinker on these little gizmos specifically, though. You read blog posts about running Kubernetes on Pis, and the immediate question is "why would you do that rather than just virtualizing stuff on an old x86 box?" The answer is really just "well, it's fun to work with little computers" which is presumably a factor for a lot of people.
[0] https://store.pine64.org/?product=rockpro64-4gb-single-board...
To give a rough idea, that 8GB chip is the same kind that's used in the 256GB DDR4 modules (2x8 chips per side, 2 sides).
Are there are any SBCs that would have suitable IO for this? The pi4 has USB 3 which internally is connected via PCIe, but the question is whether it could handle eg SATA and Ethernet on it. Not that I actually need this for anything, it would just be cool to have a small case with 1TB SSD that you can plug into a switch and deliver hundreds of MB/s. With 8GB RAM you could even make good use of the fs cache, depending on usage pattern.
In fact, I have never gotten 10Gbps out of any software-only path. Routers can do it, but it's hardware accelerated and won't generate packets at that speed. I did recently get 7Gbps out of normal Internet usage, though, with a consumer-grade 10Gbps NIC on an Intel system. Pretty decent, but not 10Gbps.
Much optimization is necessary to make 10Gbps a thing. It's not a focus for $35 single-board computers. Those things are just obsolete phone chipsets in a desktop form factor; don't expect much out of them.
In general, saturating 10gbit on any x86 from the past couple years isn't a problem. Doing it with a single TCP connection is a bit tricky and requires some tweaking, and jumbo frames help too, but a simple threaded server pushing out data via sendfile to multiple clients easily gets the job done.
But yes, I agree that this is not a useful use case for SBCs, one of these things I'd just want to build because I can, not because I should.
Yeah,
Are there any SBC with an off-CPU networking fibre/module?
What I am building are LTE enabled ethernet bridges.
The closest fits that I have found are router boards but most of them don't come with mini PCIe ports that I can connect LTE modems to.
There are a few proprietary router boards but I was thinking there could be other options - you would likely know!
https://www.tomshardware.com/how-to/boot-raspberry-pi-4-usb
I am also interested so here's what I have found:
We want a SBC with an off-CPU networking module.
You don't want to tie up the CPU shuffling data bits for networking - there are now cheap and dedicated modules for that.
Most of what I have found are 1gbps Broadcom modules.
My issue is that the closest fits that I have found are router boards.
Please reach out if you can, would love to solve this myself.