It loads for me. but there is no content. There is a header/menu at the top and a long, black page I can scroll through, at the bottom there are links to other concepts but the main content of the page is just missing.
But it's just a concept, and not a great one at that. They want dual PCIe-x16 graphics cards, which requires 32 lanes, and also SIXTEEN Thunderbolt 3 ports? For full speed, those require 4 lanes each.
Not to mention multiple other PCIe lanes allocated to USB3, SATAIII, M.2, TB2, network, and so on? By my count, the specs are pushing 100 lanes of PCIe. Even a big Extreme edition processor only has 40 lanes of PCIe, and a PLX PEX multiplexing chip only goes so far. This is basically calling for dual-socket 135W $4,000 each Xeon Broadwell E5's PLUS a PCIe mux. And thermally, even doubling the current Mac Pro, you're not going to fit 500W of graphics and 300W of processors in a 30-liter case.
It's a pretty graphic, and I like the concept of lots of ports and expandability, but it's not based on anything realistic.
While I really like the aesthetics of this, I think it still suffers the same problems as the current Mac Pro.
It needs more PCI card slots. It needs to support more than two GPUs. To suppprt that it's also going to need one helluva power supply... it needs its workstation credibility back more than it needs a beautiful design.
I still think the previous generation of Mac Pro look great. They'd look even better in Space Gray.
I appreciate it's just an industrial design concept, but the problem with off-the-shelf GPUs is that you need to route the DisplayPort connectors back to the motherboard in order to mux them with the Thunderbolt ports.
The Thunderbolt add-in cards have DisplayPort inputs for this purpose.
The GPU & motherboard vendors should agree on some extra headers to allow you to route these DP signals without ugly jumper cables on the outside of the case.
> but the problem with off-the-shelf GPUs is that you need to route the DisplayPort connectors back to the motherboard in order to mux them with the Thunderbolt ports
If the Intel CPU supports integrated graphics (which I realize a Xeon may not) there is actually support for sending display streams over PCIe, so it wouldn't be necessary to route DisplayPort cables from the card output to the motherboard.
The eGPU guys are doing this to display the output of an external GPU on the internal LCD of the laptop. [0] It does slightly reduce the PCIe bandwidth available as you need to transfer the rendered data back via the PCIe, which is then output from the IGP to the display. [1]
However this is still limiting because it means you're stuck with the display outputs supported by the IGP, so if you want to upgrade the video card to support a newer standard (e.g. HDMI 2.0) you'd be stuck.
Have you tried this? The amount of bandwidth consumed is non-trivial and especially impacts compute workloads or anything that does read-back from the GPU.
I have not used this feature myself because I'm not aware that it's supported in Linux, and I don't use Windows.
> The amount of bandwidth consumed is non-trivial and especially impacts compute workloads or anything that does read-back from the GPU.
Given that the most typical use case of this technique is for people gaming, I'd say that the amount of data being read back from the GPU in a gaming situation is quite small, so this doesn't have a significant performance impact for its intended audience.
Since I don't use my dGPU for anything but gaming, do you have any more information on the performance impact on compute workloads?
A single 4k/60p/24bit display requires 1.5GB/sec, about 10-15% of your PCIe readback bandwidth.
Since it has to be both written to and read back out of main memory, that's 3GB/sec, which is between 5-10% of your CPUs memory bandwidth.
It's OK for 1080p res but falls down at higher res.
Apple don't do this on their laptops either, and still employ a gmux to switch between integrated and discrete graphics for the laptop display (although that's mainly for power-saving).
I'd like to see the industry move away from having connectors on the back of the GPU at all and just provide a standard routing interface of some sort. I hate whenever I buy a new GPU having to figure out what adapters I'll need for my multi-monitor setup because my new GPU has different numbers of HDMI and DP outputs than my old one.
But then you are limited by the number of connectors on your motherboard. I know people who have more cards to support more monitors(not in SLI, just side by side)
This seems unlikely to happen for the foreseeable future - if ever. As pixel counts grow the port standards are evolving quickly (both HDMI and DisplayPort).
A machine where the display interface is on the motherboard will have a hard time keeping up with these developments, whereas keeping the interface on the GPU will allow the user to add support for new standards over time (see: HDR, 8K, etc).
Reading comprehension: "bigger trashcan without a proper..." means that this is a bigger trash can that does not have a proper cooling system. The regular trash can thus has.
It's just ssh. My local library did have a dial-up line for the longest time but I think they finally retired it, that's the last one I might have actually used.
I think the last one I used was in 2003. My university offered it for internet access. It was actually just dial-in shell access, but you could tunnel an internet connection using Slirp. Nice and slow.
Is your car an entry-level bicycle? What kind of servers are you running that cost 30x an actual car? We run some expensive servers but none that cost anywhere near 30x the cost of my car.
I'm including the support contracts in both numbers. I work for an ISP, so our storage systems are both large and very sensitive to downtime. You don't want x0,000 customers calling about their email, HBO Go, and billing logins failing. So the support contract is a bit extreme to match. The exact price is under some kind of NDA but let's say my car was a bit over $30k.
That must be some impressive support contract. For close to a million dollars per server I hope they come with an indentured engineer who lives in your colo ready to do a hot swap at any moment.
But where is all the bandwidth for 16 TB 3 ports supposed to come from? That’s 64 PCIe 3.0 lanes, mind you.
Adding to that, 32 PCIe 3.0 lanes for the graphics cards, 16 PCIe 2.0 lanes for the TB 2 ports, 2 PCIe lanes for the Ethernet ports, 2 PCIe lanes for the USB ports.
Of course not; this isn't even possible with a dual socket system, which would also lack the memory bandwidth to do anything except idling with that many PCIe lanes. (Plus, if your IO bandwidth is pretty much the same as your memory bandwidth ... what exactly are you gonna do in terms of data processing? 1+2?)
But this is just one of many things that just don't make any sense at all about this design. It might look nice, to some, though.
If you want to feed dozens of GPUs with compute tasks then there is more specialized hardware for that with way better density. (That goes into a rack of course).
Absolutely doable. The modern i7's have 40 PCIe lanes. If you built a quad processor system you would have 160 PCIe lanes... Have a look at the obscene number of PCIe lanes in modern server architectures.
The bigger issue would be memory bandwidth, but I think that could also be engineered around.
Edit: SuperMicro sells a system with 8 PCIE 3.0 x16 + 7 PCIE 3.0 x8 ... 184 lanes not counting the other onboard peripherals. So it basically exists minus the Apple form factor.
The point was less about the form factor and more about the number of PCIe lanes that can conceivably be available and the associated memory/bus bandwidth that is possible.
Haha, I see. That's a blade center with 8 "PCs", each supporting one PCIe x16 card and some supporting an additional x16 card (with only 8 lanes connected).
Yeah I noticed that also. Seems strange to make such a mistake with so much attention to detail.
Pretty design but by my count that machine would need to be a quad CPU (not core) system to support all that I/O. Where are those 4 CPUs gonna fit and how are you gonna keep them cool?!
I do understand the reluctance from Apple to built upgradable computers. They make their money on hardware sales, and an upgradable system would hurt those sales. At the same time their "Pro" gear simply isn't iterating fast enough, perhaps because not using standard components slows them down.
It's not Apples style, but it wouldn't hurt if they gave their professional customers a three year roadmap, just so people would know that they plan to move forward, and in which direction.
Are 16x TB3, 4x TB2 AND 4x USB3 actually achievable with current/near-future hardware?
If Apple were to implement this, I'd imagine it would be N x TB3/USB3 USB-C format ports, an ethernet port and maybe HDMI (though a dongle would possibly negate that – if 2.1 can be achieved that way)
But the sacrifice is that you actually only have 80 lines of PCIe, and it all has to be coordinated. Dual-socket computing does not quite double your processing power, and a switch adds latency. Your overall bandwidth will still be limited to what you have coming from the processor.
This kind of design is WAY more likely to come from a PC vendor than Apple. The 16 TB ports was definitely kind of a LOL moment.
I dig it. They are useful, but that can't be done with "standard components".
The whole idea of a flexible and upgradeable PC is kind of against the Apple "tightly coupled" software and hardware story. The reason that their user experience has historically (I haven't used a mac in about a decade) been so good was because they limit the available hardware for their testing purposes. They don't have a lot of choice in hardware, but what they do support works every time.
I don't know...I got off the Apple train a long time ago. I loved my Mac Pro, but it just wasn't for me.
The chipset can add more also, but they will have higher latency. It's common on lower end systems where the CPU only has 16 or so. Those'll go to the graphics card slot (the only 16x on the small board) and all other peripherals will come off the chipset's PCIe lanes.
Definitely. They'd have to develop something entirely new to do it, which I bet would cost more than just doing a quad socketed system to get all the lanes.
Kaby Lake desktop have upped the number of PCIe lanes by 20% if the Xeons do the same then you just about have enough -- your list above is 112 lanes, 48*1.2=57.6 and so 56 is not unreasonable.
Calling him a game designer based on [1] seems too limited. Yes, those renders you can find are about gaming. But he's done more than that, see his resume.
Well, except for the ridiculous magazine design which looks to be aggressively anti-ergonomic and couldn't actually work as rendered without bullets actually passing through one another as they're fed into the chamber.
The FN P90 has a similar angled entrance to the chamber. The rounds are perpendicular to the chamber while in the magazine, but are directed into place. I never bothered to buy a PS90 (the U.S. semi-auto version) to find out whether this design results in higher-than-typical stoppages.
The PC ecosystem shipped something similar years ago (with different aesthetic choices). Combine a silverstone ft03 with a noctua d14 dual tower cooler, and you get a similar convection cooling design.
The main problem is that tool-less cases were in style back then. If you pick it up in the obvious way, you'll end up with two side panels in your hands and a broken heap of computer on the ground. Other than that, it is a totally solid and very quiet case for high tdp systems.
Actually, the main problem is that you can't put 2x full length video cards in it.
Cooling a mini-itx board with a 14mm processor with <90W TDP is not a unsolved problem. But the standard ATX motherboard layout assumes most of the head will be generated by the processor, and that's simply not true today. A dual GPU setup for gaming or ML might generate 5 to 10 times the heat generated by the processor, and there's literally a couple centimeters for airflow between the two cards.
When I see an innovative design that maximizes airflow across the GPU(s), preferably one that involves removing the stock "windtunnel" cooler and replacing it with massive "tower" passive radiators and 3x 120mm temperature controlled case fans blowing cool air across them, then I'll be impressed. (note that this would require either a new motherboard design or using pci ribbon extenders)
But then you're back to the same problem of needing custom coolers which work with the modules AND the case, where we've seen Apple never release an upgrade part since launch.
Actually, that case fits two full length cards, or at least full-length the year it was built. The power supply is on the bottom, on the "CPU side" of the motherboard (the motherboard ports stick out the top), so you get the full motherboard length, plus the power supply height, plus a bit extra. It leads to a pretty dense, but well laid out install. There is a diagonally mounted 120mm fan that does nothing and has to be removed, if I remember right.
Crucially, convection (with help from fans) means the cards are pulling cooler air than in normal cases (at least in theory -- in practice, it is very quiet with one GPU).
I can't recommend it due to the lack of screws / latches on the side panels, but the general layout and airflow is great, and could easily be adapted for a different model with support for longer cards or even three way GPUs.
Current generation AMD cards have stock water coolers that are similar to what you asked for, but apparently third party air coolers are cooler/quieter. (I don't remember if the case I linked has water cooler mounts. I doubt it.)
Eyeballing that case, it looks like it might have the length to accommodate a modern GPU, but it doesn't have the width for a double-slot GPU, much less any room for airflow across it.
> the general layout and airflow is great, and could easily be adapted for a different model with support for longer cards or even three way GPUs.
My problem with current ATX/ITX is that it was designed before PCI cards consumed more power and generated more heat than the CPU. An ideal solution would allow a 160+mm clearance on the obverse side of the GPU card (allowing the "wind tunnel" cooling system to be replaced with a tower cooler with 120mm fans, and a dedicated exhaust out the back of the case for each GPU.
It would require a redesign of motherboards, cases, and GPU cards, so I'm not holding my breath. I think it's more likely that at some point we'll see cases with integrated watercooling loops, and the user would simply plug in the CPU and GPU blocks.
I am quite sure that apple's is more constrained by appearances than the coupling of hardware and software. the mac pro design while very unique is so constraining that they may have simply painted themselves into a corner and had no graceful out.
if they had kept the previous incarnation then updates could have proceeded yearly or even once every two years.
I do agree this proposal lacks a lot, it certainly has no Apple vibe to it
> 50% of the site was black background + scrolling for me, but I think I can get the gist of it. Nice design + expansion capabilities, right?
Apple's Industrial Design group needs to get it through their skulls that folks doing REAL pro work still need traditional expansion capabilities. At the very least, pro users need to:
1) Have the ability to expand RAM
2) Have space for two video cards (ThunderBolt 3 + video card enclosures is not a desirable solution)
3) Have space for at least a couple internal hard drives
For some reason, I don't think this will ever happen because the end result would probably be bigger, noisier, and uglier than what the ID group would allow. But, man.. wouldn't it be nice to be able to purchase a base config Mac Pro 2 with one stick of RAM, shipped with integrated graphics and the user could drop in any graphics card(s) they wish?
Hackintoshes can work fine for some, but oftentimes we just want to be able to run software update without the fear that a patch will break our bread-and-butter making machines.
I wonder what OSX-specific software keeps professionals on machines that they feel are underpowered, even if it's a rather mighty Mac Pro?
All the graphics and video editing suites, AFAICT, are now also available for Windows, too. For audio editing one probably does not need two video cards.
I write software for iOS / Mac as my day job and it's really gotten to a point where it feels like writing C# on Windows used to feel (which is not a bad thing). The big difference is that back then I could build a powerhouse PC for $1500 and have a handful of big monitors and really feel ahead of the power curve.
I don't feel like I have the option to do that these days, and I'm pretty disappointed by it. $4,000 on a laptop is my best attempt and it's still not.. quite. there.
Apple's vendor lock-in strategy geared towards developers appear to really be bearing fruit at the moment.
They've shifted most of their hardware lineup towards consumers and is the weakest it's been for developers in a long time, but developers are still hanging around because they have to if they want to target the Apple ecosystem.
People who've been in the industry for longer than 15 years would remember the last time this happened.
Perhaps this isn't the right place for this, but I'd be interested in a broader discussion about this. What do we do about it?
Even if we see it coming a mile away, or if we wake up and look around and see it happening.. what can we do?
Go back to the web is always an option, but I don't like it. I still genuinely believe that the experience I can deliver writing software for iOS is higher quality than what I can deliver on the web.
As an Android/iOS developer, I personally wouldn't need all of the power. However, I could see where professional video editors and the like working with multiple 4K streams and wish to stay within the familiarity of the ecosystem would want something a bit more expandable/flexible.
The current Mac Pro and iMac do have expandable / user upgradable RAM.
> 2) Have space for two video cards (ThunderBolt 3 + video card enclosures is not a desirable solution)
It's still early days for external GPUs. I suspect they will become quite popular and maybe even the preferred solution for many people. They offer a lot of practical benefits such as interoperability between desktops and laptops.
3) Have space for at least a couple internal hard drives
IMO low cost / high storage density SSDs and affordable NAS devices pretty much solve this problem.
Exactly. The trashcan Mac Pro had the same idea. But if I recall correctly there were some licensing issues that prevented many manufacturers from adopting the concept, and the cost did it in.
That the idea is not commercially feasible today (as in, my client wants their project done in 2 weeks) is precisely why pros need basic expandability.
Looks nice but I don't think that's practical (or realistic).
That's way too many USB ports, SSDs on one machine.
BTW for devs out there that need that much horsepower, what do you do? (I understand needed 32 GB of ram which was not available in the latest MacBook pros, but when do you need that much storage?)
At my last job, we had servers with 64 cores, 1TB ram, and 100+ TB of storage that were taxed to their full extent by one or two people at a time. The job was essentially taking multi terabyte datasets and matching, filtering, and mutating them based on some customer requirements.
Those servers cost a ton of cash, and we had a handful of them. I could easily see my workflow being improved and maybe even money being saved if that much horsepower was given locally to the few who used it most. But the server model also worked fine.
Personally, I have a monstrous box at home in terms of GPU, ram, and processing power, and it's mostly useful for running automated tests in multiple VMs at a time, kicking ass with ocl-hashcat, and doing hardware accelerated ML training. I don't do those things often, but I'm glad to have a beast when I do. Also, it's fun to eyefinity dual 4k on full settings ;)
Just spent £2000 on a PC after 17 years of purely Apple because I wanted CUDA cores for my creative work.
If Apple had something like this as an option I'd have easily gone upwards of £4000 to foolishly stay within their ecosystem. Guess my wallet is better off in the universe where Apple doesn't want my custom.
Yep, when we needed to buy a beefy GPU machine we did consider Apple. But no nVidia (so no CUDA) and no update in years makes it a bad value proposition. So we dropped the money on a Dell (which besides the Tesla processor has far more modern cores and allows more memory).
I feel very much like doing something similar but I'm reticent to move away from macOS because of privacy concerns.
If I moved to Linux I would still need some commercial software available on macOS but not on Linux. The obvious thing to do would be to supplement Linux with Windows 10 but I worry about privacy in Windows 10.
This maybe obvious to long term Windows users but I can only find examples from Microsoft about what they collect/track, not a definitive list[1].
Also if anyone knows, I'd also like to find out if it's possible to stop Microsoft Support from remotely accessing files on a Windows 10 machine and if you need a particular version of Windows 10 to do that. e.g. Is Windows 10 Home more or less secure than Windows 10 Enterprise.
So for the moment baring in mind I don't do CUDA I'm sticking with my slow old Mac. I should probably should build a dual booting hackintosh but I think if would be a lot of trouble to keep going (and also breaks the licensing terms).
Take that privacy statement with a grain of salt. That's for microsoft.com accounts. When you install Windows 10 they allow you to create a local-only account and opt out of any data collection.
Look into Windows 10 LTSB (Long Term Service Branch). It's rock solid, doesn't auto-update (only critical security patches), as part of the installation process you can switch off everything, doesn't have Edge or Cortana or any other guff. It was made for ATM's and single purpose machines. I use one for my Plex Media Server and XProtect / Sighthound Server etc.. all rock solid. The "apparent" disadvantages listed against LTSB I actually believe are advantages. Like, want new Windows 10 features? Sorry you'll have to reinstall the latest full release of Windows 10.1 LTSB.
I smiled at "Standard components ... exceptionally futureproof". The way Apple are going the next mac pro will probably run an ARM CPU and have the RAM soldered on
Real Mac Pro would be an aluminum "cheese grater" large quiet box that sits under the desk that I never see or hear unless I want to upgrade something.
The trashcan is at most Mac Mini Pro. Having thousand cables and external boxes sitting on your desk to expand it is not practical nor elegant, and gets out of date pretty fast. And quite frankly it's ugly as well.
I have a G5 from the last generation of the cheese graters under my desk right now. The case has seen better days, so I might have to source a better preserved one soon, but I can say it runs Linux just fine.
The mechanical design is quite sound (forgetting about the HDD cage for a second), though, and they look fetching, even today. (imho much better than the trash cans)
There are some really nice PC/Hackintosh builds with hacked up G5 cases. You can even get custom-built motherboard trays that are meant to go in those cases:
i still use my 2009 8x xeon mac pro as my primary workstation. upgraded ram, gpu, ssd, video, 4k monitor, and a couple of other things.
i'm trying to get 10 years out of it, which looks do-able, maybe. it refuses to install the latest OS X's though, so it may finish its life as a linux box.
will 2019 finally be the year of linux on the desktop???
I managed to install the latest OS on mine. Basically you have 4 options:
1. Hack Sierra installer to bypass restrictions.
2. Use target disk mode on Mac Pro and another Mac that supports firewire 800 to install Sierra.
3. Clone internal disk to external. Do upgrade to Sierra on it, then clone back to internal.
4. Remove internal disk, plug into external USB case, install Sierra on it. Plug it back to Mac Pro.
In all 4 cases you must hack the /System/Library/CoreServices/PlatformSupport.plist to make it bootable on Mac Pro.
You need to do it for recovery partition as well, if you ever want to boot to recovery partition.
250 comments
[ 2.9 ms ] story [ 234 ms ] threadBut it's just a concept, and not a great one at that. They want dual PCIe-x16 graphics cards, which requires 32 lanes, and also SIXTEEN Thunderbolt 3 ports? For full speed, those require 4 lanes each.
Not to mention multiple other PCIe lanes allocated to USB3, SATAIII, M.2, TB2, network, and so on? By my count, the specs are pushing 100 lanes of PCIe. Even a big Extreme edition processor only has 40 lanes of PCIe, and a PLX PEX multiplexing chip only goes so far. This is basically calling for dual-socket 135W $4,000 each Xeon Broadwell E5's PLUS a PCIe mux. And thermally, even doubling the current Mac Pro, you're not going to fit 500W of graphics and 300W of processors in a 30-liter case.
It's a pretty graphic, and I like the concept of lots of ports and expandability, but it's not based on anything realistic.
It needs more PCI card slots. It needs to support more than two GPUs. To suppprt that it's also going to need one helluva power supply... it needs its workstation credibility back more than it needs a beautiful design.
I still think the previous generation of Mac Pro look great. They'd look even better in Space Gray.
The current Mac Pro officially supports 64GB and 3rd parties offer upgrades to 128GB.
Even the 2006 models can be upgraded to 32GB.
The Thunderbolt add-in cards have DisplayPort inputs for this purpose.
The GPU & motherboard vendors should agree on some extra headers to allow you to route these DP signals without ugly jumper cables on the outside of the case.
If the Intel CPU supports integrated graphics (which I realize a Xeon may not) there is actually support for sending display streams over PCIe, so it wouldn't be necessary to route DisplayPort cables from the card output to the motherboard.
The eGPU guys are doing this to display the output of an external GPU on the internal LCD of the laptop. [0] It does slightly reduce the PCIe bandwidth available as you need to transfer the rendered data back via the PCIe, which is then output from the IGP to the display. [1]
However this is still limiting because it means you're stuck with the display outputs supported by the IGP, so if you want to upgrade the video card to support a newer standard (e.g. HDMI 2.0) you'd be stuck.
[0] https://www.techinferno.com/index.php?/forums/topic/9658-egp...
[1] https://www.asus.com/support/faq/1006857
[2] https://www.techinferno.com/index.php?/forums/forum/83-diy-e...
I have not used this feature myself because I'm not aware that it's supported in Linux, and I don't use Windows.
> The amount of bandwidth consumed is non-trivial and especially impacts compute workloads or anything that does read-back from the GPU.
Given that the most typical use case of this technique is for people gaming, I'd say that the amount of data being read back from the GPU in a gaming situation is quite small, so this doesn't have a significant performance impact for its intended audience.
Since I don't use my dGPU for anything but gaming, do you have any more information on the performance impact on compute workloads?
Since it has to be both written to and read back out of main memory, that's 3GB/sec, which is between 5-10% of your CPUs memory bandwidth.
It's OK for 1080p res but falls down at higher res.
Apple don't do this on their laptops either, and still employ a gmux to switch between integrated and discrete graphics for the laptop display (although that's mainly for power-saving).
A machine where the display interface is on the motherboard will have a hard time keeping up with these developments, whereas keeping the interface on the GPU will allow the user to add support for new standards over time (see: HDR, 8K, etc).
But really, was that a cute way to describe ssh, or are you actually dialing into stuff in 2017? Because that would be weird, but kind of cool.
Creative pros are increasingly reliant on CUDA and the ones who are not just don't know what they're missing.
But where is all the bandwidth for 16 TB 3 ports supposed to come from? That’s 64 PCIe 3.0 lanes, mind you.
Adding to that, 32 PCIe 3.0 lanes for the graphics cards, 16 PCIe 2.0 lanes for the TB 2 ports, 2 PCIe lanes for the Ethernet ports, 2 PCIe lanes for the USB ports.
That’s 64 + 32 + 16 + 2 + 2 = 116 PCIe lanes
Guess not?
But this is just one of many things that just don't make any sense at all about this design. It might look nice, to some, though.
The bigger issue would be memory bandwidth, but I think that could also be engineered around.
Edit: SuperMicro sells a system with 8 PCIE 3.0 x16 + 7 PCIE 3.0 x8 ... 184 lanes not counting the other onboard peripherals. So it basically exists minus the Apple form factor.
Which product is that, specifically?
It’s of course achievable with any CPU using PCIe switches. However, they don’t magically increase bandwidth.
Two CPUs is probably all that can be crammed into a case that small, so 80 lanes (+ ~8 via PCH) is as good as it gets.
The point was less about the form factor and more about the number of PCIe lanes that can conceivably be available and the associated memory/bus bandwidth that is possible.
It's not really comparable to this project.
Pretty design but by my count that machine would need to be a quad CPU (not core) system to support all that I/O. Where are those 4 CPUs gonna fit and how are you gonna keep them cool?!
I do understand the reluctance from Apple to built upgradable computers. They make their money on hardware sales, and an upgradable system would hurt those sales. At the same time their "Pro" gear simply isn't iterating fast enough, perhaps because not using standard components slows them down.
It's not Apples style, but it wouldn't hurt if they gave their professional customers a three year roadmap, just so people would know that they plan to move forward, and in which direction.
Under the memory probably. It looks like the basic idea is saw the current Mac Pro down the middle and shove some graphics card slots there.
If Apple were to implement this, I'd imagine it would be N x TB3/USB3 USB-C format ports, an ethernet port and maybe HDMI (though a dongle would possibly negate that – if 2.1 can be achieved that way)
Otherwise, looks good to me.
If you really wanted to do it, the biggest processors have 40 lanes of PCIe. Intel's Xeon E5's can be dual-socket mounted, for a total of 80 PCIe lanes. Adding PCIe switching, like two Avago/PLX PEX9797s (see http://www.anandtech.com/show/9245/avago-announces-plx-pex97..., particularly the diagram at http://images.anandtech.com/doci/9245/Slide14.JPG) can get you there in the end.
But the sacrifice is that you actually only have 80 lines of PCIe, and it all has to be coordinated. Dual-socket computing does not quite double your processing power, and a switch adds latency. Your overall bandwidth will still be limited to what you have coming from the processor.
Edit: fixed wrong link
I dig it. They are useful, but that can't be done with "standard components".
The whole idea of a flexible and upgradeable PC is kind of against the Apple "tightly coupled" software and hardware story. The reason that their user experience has historically (I haven't used a mac in about a decade) been so good was because they limit the available hardware for their testing purposes. They don't have a lot of choice in hardware, but what they do support works every time.
I don't know...I got off the Apple train a long time ago. I loved my Mac Pro, but it just wasn't for me.
http://www.apple.com/mac-pro/
It's still nowhere near enough bandwidth to do what this concept wants to do.
16x TB3 = 64x PCIe Lanes
2x GPU = 32x PCIe Lanes
2x M.2 = 8x PCIe Lanes
Chipset = 8x PCIe Lanes
Would be too much for even 2x Skylake Xeons. Maybe 10-12x TB3 ports.
The author is a game designer wanting to enter industrial design. It's just a bunch of renders created for fun. Here is another one you might enjoy:
http://pascaleggert.de/ThorA1.html
[1] http://pascaleggert.de/aboutme.html
The FN P90 has a similar angled entrance to the chamber. The rounds are perpendicular to the chamber while in the magazine, but are directed into place. I never bothered to buy a PS90 (the U.S. semi-auto version) to find out whether this design results in higher-than-typical stoppages.
Pics:
http://www.hardwarecanucks.com/forum/hardware-canucks-review...
http://www.silentpcreview.com/article1190-page4.html
The main problem is that tool-less cases were in style back then. If you pick it up in the obvious way, you'll end up with two side panels in your hands and a broken heap of computer on the ground. Other than that, it is a totally solid and very quiet case for high tdp systems.
Cooling a mini-itx board with a 14mm processor with <90W TDP is not a unsolved problem. But the standard ATX motherboard layout assumes most of the head will be generated by the processor, and that's simply not true today. A dual GPU setup for gaming or ML might generate 5 to 10 times the heat generated by the processor, and there's literally a couple centimeters for airflow between the two cards.
When I see an innovative design that maximizes airflow across the GPU(s), preferably one that involves removing the stock "windtunnel" cooler and replacing it with massive "tower" passive radiators and 3x 120mm temperature controlled case fans blowing cool air across them, then I'll be impressed. (note that this would require either a new motherboard design or using pci ribbon extenders)
But then you're back to the same problem of needing custom coolers which work with the modules AND the case, where we've seen Apple never release an upgrade part since launch.
Crucially, convection (with help from fans) means the cards are pulling cooler air than in normal cases (at least in theory -- in practice, it is very quiet with one GPU).
I can't recommend it due to the lack of screws / latches on the side panels, but the general layout and airflow is great, and could easily be adapted for a different model with support for longer cards or even three way GPUs.
Current generation AMD cards have stock water coolers that are similar to what you asked for, but apparently third party air coolers are cooler/quieter. (I don't remember if the case I linked has water cooler mounts. I doubt it.)
Eyeballing that case, it looks like it might have the length to accommodate a modern GPU, but it doesn't have the width for a double-slot GPU, much less any room for airflow across it.
> the general layout and airflow is great, and could easily be adapted for a different model with support for longer cards or even three way GPUs.
My problem with current ATX/ITX is that it was designed before PCI cards consumed more power and generated more heat than the CPU. An ideal solution would allow a 160+mm clearance on the obverse side of the GPU card (allowing the "wind tunnel" cooling system to be replaced with a tower cooler with 120mm fans, and a dedicated exhaust out the back of the case for each GPU.
It would require a redesign of motherboards, cases, and GPU cards, so I'm not holding my breath. I think it's more likely that at some point we'll see cases with integrated watercooling loops, and the user would simply plug in the CPU and GPU blocks.
if they had kept the previous incarnation then updates could have proceeded yearly or even once every two years.
I do agree this proposal lacks a lot, it certainly has no Apple vibe to it
Apple's Industrial Design group needs to get it through their skulls that folks doing REAL pro work still need traditional expansion capabilities. At the very least, pro users need to:
1) Have the ability to expand RAM
2) Have space for two video cards (ThunderBolt 3 + video card enclosures is not a desirable solution)
3) Have space for at least a couple internal hard drives
For some reason, I don't think this will ever happen because the end result would probably be bigger, noisier, and uglier than what the ID group would allow. But, man.. wouldn't it be nice to be able to purchase a base config Mac Pro 2 with one stick of RAM, shipped with integrated graphics and the user could drop in any graphics card(s) they wish?
Hackintoshes can work fine for some, but oftentimes we just want to be able to run software update without the fear that a patch will break our bread-and-butter making machines.
All the graphics and video editing suites, AFAICT, are now also available for Windows, too. For audio editing one probably does not need two video cards.
So, what's your use case? Just curious.
I don't feel like I have the option to do that these days, and I'm pretty disappointed by it. $4,000 on a laptop is my best attempt and it's still not.. quite. there.
They've shifted most of their hardware lineup towards consumers and is the weakest it's been for developers in a long time, but developers are still hanging around because they have to if they want to target the Apple ecosystem.
People who've been in the industry for longer than 15 years would remember the last time this happened.
Even if we see it coming a mile away, or if we wake up and look around and see it happening.. what can we do?
Go back to the web is always an option, but I don't like it. I still genuinely believe that the experience I can deliver writing software for iOS is higher quality than what I can deliver on the web.
Has anyone found a secret third option yet? :)
The current Mac Pro and iMac do have expandable / user upgradable RAM.
> 2) Have space for two video cards (ThunderBolt 3 + video card enclosures is not a desirable solution)
It's still early days for external GPUs. I suspect they will become quite popular and maybe even the preferred solution for many people. They offer a lot of practical benefits such as interoperability between desktops and laptops.
3) Have space for at least a couple internal hard drives
IMO low cost / high storage density SSDs and affordable NAS devices pretty much solve this problem.
Exactly. The trashcan Mac Pro had the same idea. But if I recall correctly there were some licensing issues that prevented many manufacturers from adopting the concept, and the cost did it in.
That the idea is not commercially feasible today (as in, my client wants their project done in 2 weeks) is precisely why pros need basic expandability.
That's way too many USB ports, SSDs on one machine.
BTW for devs out there that need that much horsepower, what do you do? (I understand needed 32 GB of ram which was not available in the latest MacBook pros, but when do you need that much storage?)
Those servers cost a ton of cash, and we had a handful of them. I could easily see my workflow being improved and maybe even money being saved if that much horsepower was given locally to the few who used it most. But the server model also worked fine.
Personally, I have a monstrous box at home in terms of GPU, ram, and processing power, and it's mostly useful for running automated tests in multiple VMs at a time, kicking ass with ocl-hashcat, and doing hardware accelerated ML training. I don't do those things often, but I'm glad to have a beast when I do. Also, it's fun to eyefinity dual 4k on full settings ;)
;)
If Apple had something like this as an option I'd have easily gone upwards of £4000 to foolishly stay within their ecosystem. Guess my wallet is better off in the universe where Apple doesn't want my custom.
Performance is said to be OK with TB2 and AKiTiO Thunder2 enclosures but it's still a bit hacky and unsupported in macOS.
If I moved to Linux I would still need some commercial software available on macOS but not on Linux. The obvious thing to do would be to supplement Linux with Windows 10 but I worry about privacy in Windows 10.
This maybe obvious to long term Windows users but I can only find examples from Microsoft about what they collect/track, not a definitive list[1].
Also if anyone knows, I'd also like to find out if it's possible to stop Microsoft Support from remotely accessing files on a Windows 10 machine and if you need a particular version of Windows 10 to do that. e.g. Is Windows 10 Home more or less secure than Windows 10 Enterprise.
So for the moment baring in mind I don't do CUDA I'm sticking with my slow old Mac. I should probably should build a dual booting hackintosh but I think if would be a lot of trouble to keep going (and also breaks the licensing terms).
[1] https://privacy.microsoft.com/en-us/privacystatement/
http://www.howtogeek.com/273824/windows-10-without-the-cruft...
* https://www.safer-networking.org/spybot-anti-beacon/ * https://www.oo-software.com/en/shutup10
Note - I'm not sure if anyone has deeply tested/verified them though.
Ideally a good Open Source one would exist. Maybe it does, I just don't know of it yet. ;)
Thunderbolt 3 is an extension to the USB 3.1 standard which utilizes the USB-C connector and implements USB Power delivery.
Assuming they ever make another mac pro of course
The trashcan is at most Mac Mini Pro. Having thousand cables and external boxes sitting on your desk to expand it is not practical nor elegant, and gets out of date pretty fast. And quite frankly it's ugly as well.
The mechanical design is quite sound (forgetting about the HDD cage for a second), though, and they look fetching, even today. (imho much better than the trash cans)
http://www.mountainmods.com/mountain-mods-modular-removable-...
i'm trying to get 10 years out of it, which looks do-able, maybe. it refuses to install the latest OS X's though, so it may finish its life as a linux box.
will 2019 finally be the year of linux on the desktop???
1. Hack Sierra installer to bypass restrictions. 2. Use target disk mode on Mac Pro and another Mac that supports firewire 800 to install Sierra. 3. Clone internal disk to external. Do upgrade to Sierra on it, then clone back to internal. 4. Remove internal disk, plug into external USB case, install Sierra on it. Plug it back to Mac Pro.
In all 4 cases you must hack the /System/Library/CoreServices/PlatformSupport.plist to make it bootable on Mac Pro.
You need to do it for recovery partition as well, if you ever want to boot to recovery partition.