I'll go ahead and ask the newbie question - what's the benefit? I can see how more cache and better multiprocessing would be important to a server but how does a Xeon improve a single user system? Do they have some other secret sauce? I'm assuming they're not expecting anyone to build a laptop with more than one of these in it?
A ton of designers or developers are doing work with graphics or compiling on the go - both very CPU bottlenecked operations, and the Xeon is targeting this audience.
Me personally I do a lot of Android development, and waiting 20 seconds to compile each time I make a change is a huge bottleneck.
E3s are basically i7s with different binning, e.g. there are E3s without the IGP ; so if an i7 part has yield issues with the IGP, it can be disabled, binned as a Xeon E3 and sold more cheaply than the equivalent i7.
2P/4P multi-cpu is the other major feature of Xeon which only appears in E5/E7.
I am not an expert, but from what I know, Xeon based chips have a lot more PCI and other IO lanes. So you can for instance have a dedicated lane for each thunderbolt and USB 3 port on the laptop. With i7, you may have 2 thunderbolt ports, for instance, but those 2 ports will share the same 20Gbps lane to the chip. With Xeon you can, at least theoretically, connect each thunderbolt on its own lane.
Edit: According to the Wikipedia the Skylake i7 and i5 suppose to have 20 PCIe lanes. [1] The Xeon E5 v3 Family has 40 PCIe lanes, so I am assuming same is going to be true for the Skylake version of the Xeon.
You also have the possibility of 6 cores! I'll take that :)
What are you going to use 40 pice lanes on a laptop for?
Unless you are running 4 discrete graphic cards or an army of nvme ssd's you won't tap half of that...
There are all sorts of things that "should" be standard on all computers, but until the average consumer (non-technical, not HN reading) starts saying "I got to get me some ECC memory" it's not going to be something Intel cares about.
There really isn't a big price difference between an i7 and an E3. Most consumers don't need ECC memory. So, leave it out and charge less. If you need it, get the Xeon.
I disagree. The sheer volume of data being pushed these days even by consumer computers is enough to be occasionally affected by a bit flip. ECC is used in the cache memory found on the CPU itself, so it's been important at that level for a long time even on consumer products.
my guess is that the ECC aspect is really a hit or miss on the waffer fabrication, that's why they have such a huge pool of the market without it. also that is why intel has 200 models of each CPU. they waste nothing that cames out of the oven there.
Many server vendors are dropping ecc as the performance hit isn't worth the additional reliability in many cases. I don't see the point of ecc today for most applications also...
> Many server vendors are dropping ecc as the performance hit isn't worth the additional reliability in many cases.
Bullshit on both counts. (even the low-end Lenovo, HP and Dell servers I see sometimes on Slickdeals still have ECC, so what brands are you even talking about?)
Hey, how about keeping it civil and polite? It's been pretty good here for the last year, following a relentless drive to call out bad behaviour. It'd be a shame to see it backslide.
I'm not sure if you're familiar with the idiom "calling bullshit"
I am familiar with it and, to me, it's rude. I might say it to a friend if we're just bantering, but I'd never say it to a colleague. Clearly, I'm in the minority here.
Please (re-)read the site rules. Your comment breaks them in two ways: it's a personal attack, and it's unsubstantive. A good form of this comment would be to drop the attack and add new information about how the claim is wrong, or what is true.
Troll? ECC is only going to become more important as memory sizes grow. 256 GB isn't rare in a multi socket server and I cannot phantom that without ECC.
Sure there is - they cut the feature, and people still keep buying their CPUs in preference to the AMD ones which retain it. Since people obviously prefer speed to data integrity, why would Intel do anything different?
Market segmentation is a good thing. The minority of people that need workstation/server -level reliability pay a premium, while the average consumer gets to pay less. If they abandoned this practice, the typical consumer would pay a lot more.
Bit errors have been used to hijack communications. Just register a bunch of domains that differ from popular ones by a single bit and watch the requests come in.[1]
Typically, users don't recognize single bit errors. Something breaks and they think, "Stupid buggy software." With modern RAM capacities, you're pretty likely to get some bits flipped every few months.
> Typically, users don't recognize single bit errors. Something breaks and they think, "Stupid buggy software." With modern RAM capacities, you're pretty likely to get some bits flipped every few months.
Supposedly much higher than that[1]. More like "5 single bit errors in 8 Gigabytes of RAM per hour using the top-end error rate" [2]
Memory bit flips happen all the time—not like daily or weekly per computer, but you can have multiple memory bit flip incidents during the multi-year life of an average computer. Either you detect them, or you get silent data corruption. If you're lucky it's in unused ram; if you're not, it may be in running code (possible crash or software misbehavior) or data (in a text document, one corrupted character; in some other kind of data, it could be much worse). Think about how a bit flip in code or data can affect things like ciphertext generation. It can become a huge deal.
Think of "bad ram" —the kind that fails memtest86— as ram that has an increased tendency (compared to "good ram") to flip certain bits that with a little more luck, or more conservative timings, wouldn't flip. The threshold between good ram and bad ram is not black and white.
Aggressive memory testing before use can weed out memory that's plain bad (one bit always screwing up), and certain borderline cases of memory that will flip bits under certain high stress loads or access patterns (I suppose that's due to thermal entropy?), but it's no guarantee, and it certainly offers no protection against cosmic ray induced bit flips, which happen too.
Ordinary consumer memory is not "quite reliable" if you care about data integrity or stability.
You're spinning the odds a bit. The gross majority of data in memory is never written to persistent storage.
If you're a typical consumer, then of that data that is written, only a tiny fraction is user data, and of that user data, media usually represents most of the volume. Media is tolerant to a random bit flip.
Enterprise needs to worry about these things due to sheer scale and things like automatic propagation, but the odds of getting a bit flip in your tax spreadsheet or research paper and then writing it to disk are incredibly small, and personally I figure versioning/backups protect me from a significant portion of that (small) risk.
Which brings up disk crashes. Not to divert the conversation, but disk crashes are massively more dangerous than bit flips, and so many people haven't taken even the first steps against them...
At work, we see ECC events frequently (our servers are usually at least 128 GB, so we have higher chances). Most of which are correctable. Some servers will register a couple of correctable events a week, which I would count as a tangible benefit. If not for ecc, this servers would have corrupted data, possibly undetected for a long time, or possibly resulting in unexplained poor reliability; once we figured it out, we would need to spend time figuring out which ram stick is malfunctioning, too, instead of using the diagnostic less on the motherboard. Many times the ECC errors come in a huge batch, and the resulting NMIs make the server unusable, but again we have diagnostics, and not unexplained failures.
You really have to read the wikipedia page on ECC. There is the result of a large scale study by Google.
Beside that, if you use a workstation with ECC memory and regularly utilized a good portion of that RAM (I use 128GB for bioinformatics / genomics applications), there is a good chance of witnessing an ECC error / MCE (Machine check exception).
Here's the tl;dr
"(i.e. about 5 single bit errors in 8 Gigabytes of RAM per hour using the top-end error rate), and more than 8% of DIMM memory modules affected by errors per year."
On a small data center we have, running 6 V40z servers with 64gb of memory each, in a controlled environment at 25c, we register roughly one parity event a month per-system.
And that's ~6-7 years old memory, which is less dense than modern chips, in a ideal temperature environment (the racks are water-cooled), down in a basement.
You average laptop is never going to register less than that, considering you're carrying it around with less environmental shielding, overheat it, and so on.
In fact, with a laptop with 64gb of memory that has gbs of paged data in memory for weeks, the chances are greater that your laptop with linux will crash due to a random bitflip that goes unnoticed, as opposed to a kernel bug.
And there's simply no way to tell.
How about that?
The lack of ECC in consumer grade CPUs is irritating, because when AMD was viable, they had ECC support in most of their CPUs (but you had to make sure the motherboard supported it). You could use a random AMD CPU + a motherboard supporting ECC & be sure of your data integrity.
These days I pay the "server" tax on the machines I build, buying Xeons just to get ECC. If they supported ECC, a normal i5 / i7 would probably be plenty, and the build would be less of a PITA as there would be more and cheaper motherboard choices with workstation basics like audio on board.
And yes, I use ZFS, and am otherwise pretty paranoid..
The suitability of i5/i7 for most workloads hasnt gone unnoticed at Intel. I expect future i5/i7 to be gimped with respect to IO connectivity to ensure that people doing real computation continue to pay the Xeon tax. For reference see the bits about pcie lanes at http://semiaccurate.com/2015/08/05/intel-plays-press-skylake... . Another place where this shows up is the AVX 512 fiasco where AVX 512 is missing from consumer chips.
Intel has some of the best engineers in the world and make exceptional products. I wish the beancounters at Intel were a bit less into market segmentation.
The Lenovo Thinkstation C20 is also a good dual Xeon choice if you can live with a 24GB RAM limit (it's also very compact).
There's also specialized resellers like ITSCO.de and ESM-Computer in Germany (they ship EU-wide).
For ~$1000, a dual X5670 workstation with 12 ECC RAM slots, I think it's an excellent deal.
If you want something newer, the Z820/Z620 are a good choice, they should start rapidly dropping in price soon enough.
These computers are designed extremely well (the 800+ Watt PSUs alone are eons ahead of anything on the consumer market), but even in the rare cases they break the parts are cheap (a replacement system board for ~$100, older but still powerful Xeons also drop in price insanely fast).
I don't know why people buy new midrange PCs for the same price anymore, unless they need the latest processors :-)
Thanks very much for this. It's the Z620 I'm using at the moment, and I think it's excellent. Such a productive machine. Very nice internals too, logically laid out and seems like it'd be easy to upgrade in future.
The performance difference between Westmere (X5000 series in the Z600/Z800) and Sandy Bridge (E5 used in the Z620) looks negligible.
The high end X5670 or X5690 should outperform all but the top first gen Xeon E5.
It uses more power at idle, but the upfront cost is almost halved at the moment...
I'd say you should get a unit with the highest end processors you can find (it will end up way cheaper than upgrading them) and preferably with the 1100 Watt PSU, the rest can be upgraded later and you should have plenty of power for anything (dual GTX 9xx for gaming and stuff :-))
That's not correct. Sandy Bridge was a huge performance jump over Westmere, with ~50% performance for most applications on equivalent models between these two generations. A first gen E5 2640 should handily outperform the X5690, and a current gen (v3) E5 2620 would even outperform that.
http://cpubenchmark.net/high_end_cpus.html
Obviously, the upfront cost of the older hardware is still better value when ignoring power consumption, but the v3 will also support much more and faster RAM, PCI-E 3.0, USB 3.0, and other newer technologies which will matter more as time goes on. You'll also have the option of a CPU only upgrade to higher end E5 v3's or even E5 v4's at a later point in time when those CPU models are more affordable.
I totally missed the fact that the HP Z620/820 supports v2 and v3 E5s. That definitely makes it a better choice for future upgrades.
The X5690 is comparable to the first generation E5-2640 (slightly slower), and the X5670 is comparable to the E5-2620.
Even the first gen E5 has the AVX instruction set, 4 memory channels (vs 3 for the Westmere-EP) and a faster QPI, so it might have better performance in certain applications.
Sorry, I didn't meant to imply the Z620/820 supports v3. Just pointing out that the performance jump in Sandy Bridge was significant.
v1 and v2 are socket compatible, but v3 is not socket compatible with the earlier versions, but will likely be compatible with v4 (Intel socket changes usually happen only every other generation). I'd suggest going all new with even a low end E5 2620v3 if you can, which you could for everything other than the GPUs for under 1700 (based on a Supermicro workstation build).
I learned this the hard way when I bought a 2x16GB registered DDR3 kit, and then finding out my i5 won't boot with it. It was next to impossible to find unbuffered memory with 16GB per module last time I checked.
He said registered. I don't think you can buy registered RAM that's not ECC.
It's the type of ECC RAM you need for E5 & E7 Xeons - multiple physical CPUs. Many of these systems won't even boot with normal DDR3 sticks, ECC or otherwise. Must have registered DDR3 (and I suppose DDR4 now).
I vaguely remember you could buy non-ECC registered memory back in the SDRAM days for servers. It was before we really understood trans-gigabyte error rates.
Oddly enough, one of the biggest benefits is going to be in the gpu. If you're doing 3d design, you're not going to be able to get support for Autocad, Solidworks, etc unless you have a supported platform. One of the benefits of Xeon graphics is the extensive testing done on the chips to the point where you can use the integrated graphics in CAD software. Up until now, if you wanted a mobile workstation to do cad, you had to get one with a mobile firepro or quadro, even if your needs were relatively modest.
A very good question. Impossible to answer until it shows up on Ark, but Ark currently does not list any kind of "v5" Xeon CPU, nor any notebook CPU bearing the Xeon branding, so all we have is this blog.
Unfortunately, I'm kinda really doubting it. Gut instinct is saying that the AVX-512 thing is Intel trying to keep companies doing high end data crunching buying their many-cored xeons. Given the performance boost that the features are going to give, it's not inconceivable that Intel doesn't want people buying their cheap quad core chips instead of a pair of their more expensive 8+ core Xeons.
They'll also need to drop some to developer systems. Then again, I'll probably get a brand new workstation when new AVX-512 Skylake Xeons arrive. Intel SDE emulator is goes only so far...
They might even support Coreboot. This is good. As a developer, I hate bad quality hardware. Apple is good quality, but some stuff is too consumer oriented: glossy screens and fragile alu+glass. And since Tiger I dislike their software.
At least there's huge pressure from the community. Who knows. Thousands of people requested it in the survey. It's hard to tell, but I think Lenovo might be steering in the right direction after realising techie users are a very good niche.
For me it finally feels great to write -- low profile keys with great separation. I personally feel clumsy on most desktop keyboard because of their sloped keys.
Apple has this weird history of using laptop CPUs to make their machines smaller or thinner. I'm hoping they update the Mac Pro soon, but I can see them letting it continue to languish (no updates since December 19, 2013) or coming up with a smaller form factor (thus using this CPU).
I do wonder how these CPUs would do as processors for blades?
For laptop? Xenon has server chips with 20 cores. That seems a bigger upgrade than just two standard 4 core i7s for the really crazy parallel workload. Not awesome for many workloads of course.
I think I read somewhere that the Skylake chips will have support for the new xpoint 3d memory. I would be surprised if it became commercially available this soon, but it would be great to have a MBP with the new xeon and xpoint memory instead of the SSD. The xpoint memory is supposed to be as much as 1,000x faster than SSD while drawing less power and having a lot more write-cycles.
It would be interesting to know if the number of CPU types we have right now is because of hardware issues (binning) or just plain market segmentation. Because many of the features could be enabled in software/firmware, I think they could just do a "custom" CPU where you just license whatever features you use. Do you want ECC and AVX-512 on your i7? Do you need a GPU on your Xeon? Just pay $50 (or whatever) for the key to enable these features.
Any chance of getting > 4 cores in a notebook with one of these chips? Would love a 6-core or 8-core MacBook Pro, even if it had to be run at lower frequency or aggressively throttled while on battery.
If this also means ECC memory in laptops, that's friggin' amazing.
I sometimes dream of a world where consumer equipment has ECC memory. Some say: total overkill. I say: the stats on servers are so that you are probably hit multiple times a year with bit-flips in memory. You'll never know because it's not registered anywhere. But those apps hanging, this strange crash, a corrupted document. All these small things.
Absolutely. Memory errors are insidious because they can look like just about anything - driver bugs, software bugs, disk errors, power issues, media blips, or even nothing at all, just some silent bit of data corruption you don't even notice. It's easy to dismiss the risk when you have no idea how big it is.
It's eye-opening once you've managed a few machines with ECC and seen first hand how common correction events are.
About time! Workstations equipped with ECC just don't crash or do occasional weird things. ECC helps against all kind of unquantifiable weirdness and unexplainable bugs.
I'll be first in line to buy a laptop with ECC. Extra stability is worth it.
OMG I desperately want a powerful portable PC!!! I'd even buy a big ugly "notebook" without battery. Who needs a battery? There's our phone for the go.
These days I'm playing with the idea of syncing a VM between my notebook and a powerful desktop PC for at desk use. Anyone tried it? Would it work? Virtualbox, maybe Docker?
I actually have done a lot of that with one of the gigantic gaming laptops. an Alienware M18X r2 in my case. I'd love to have gotten ECC ram in it but that's the only real issue I have with it. Weighs about 20ish lbs (9kg for you non-yanks). But I've run database clusters with qemu/libvirt to learn how to set them up, but my main use of that kind of setup has been to test builds of software on other operating systems.
It's been a really interesting laptop for a lot of stuff. 3 internal 2.5" bays, you can convert the optical bay into a 4th. An mSATA bay for flash, and an express card slot that can take any pci-e 1x compatible card (pretty much everything) with the right enclosure. 2x graphics card slots with adequate cooling for everything. it's definitely a beast but i bought it entirely for that purpose. it's been great for doing CAD (though i plan on making it better) for 3d printing and also for being able to keep it up to date (aside from cpu and ram, those are already maxed out for its generation).
121 comments
[ 3.2 ms ] story [ 197 ms ] threadMe personally I do a lot of Android development, and waiting 20 seconds to compile each time I make a change is a huge bottleneck.
2P/4P multi-cpu is the other major feature of Xeon which only appears in E5/E7.
Edit: According to the Wikipedia the Skylake i7 and i5 suppose to have 20 PCIe lanes. [1] The Xeon E5 v3 Family has 40 PCIe lanes, so I am assuming same is going to be true for the Skylake version of the Xeon.
You also have the possibility of 6 cores! I'll take that :)
[1] https://en.wikipedia.org/wiki/Skylake_(microarchitecture)
They're marketed as "high end desktop" chips, and they're socket-compatible with the Xeon.
From a marketing perspective it might not be a good idea to introduce a feature that would cause performance regressions in a new product.
Bullshit on both counts. (even the low-end Lenovo, HP and Dell servers I see sometimes on Slickdeals still have ECC, so what brands are you even talking about?)
Hey, how about keeping it civil and polite? It's been pretty good here for the last year, following a relentless drive to call out bad behaviour. It'd be a shame to see it backslide.
Call a spade a spade, etc.
It is an old enough expression that an urban dictionary reference shouldn't be necessary, but here you go:
http://www.urbandictionary.com/define.php?term=I+call+bullsh...
I am familiar with it and, to me, it's rude. I might say it to a friend if we're just bantering, but I'd never say it to a colleague. Clearly, I'm in the minority here.
Please (re-)read the site rules. Your comment breaks them in two ways: it's a personal attack, and it's unsubstantive. A good form of this comment would be to drop the attack and add new information about how the claim is wrong, or what is true.
https://news.ycombinator.com/newsguidelines.html
https://news.ycombinator.com/newswelcome.html
The last time I really looked into this stuff there was non-ECC, ECC and Rambus for servers.
Typically, users don't recognize single bit errors. Something breaks and they think, "Stupid buggy software." With modern RAM capacities, you're pretty likely to get some bits flipped every few months.
1. http://dinaburg.org/bitsquatting.html
Supposedly much higher than that[1]. More like "5 single bit errors in 8 Gigabytes of RAM per hour using the top-end error rate" [2]
1. http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf
2. https://en.wikipedia.org/wiki/ECC_memory
Think of "bad ram" —the kind that fails memtest86— as ram that has an increased tendency (compared to "good ram") to flip certain bits that with a little more luck, or more conservative timings, wouldn't flip. The threshold between good ram and bad ram is not black and white.
Also see "rowhammer" — https://news.ycombinator.com/item?id=9171722 — ECC mitigates that to a significant degree.
Aggressive memory testing before use can weed out memory that's plain bad (one bit always screwing up), and certain borderline cases of memory that will flip bits under certain high stress loads or access patterns (I suppose that's due to thermal entropy?), but it's no guarantee, and it certainly offers no protection against cosmic ray induced bit flips, which happen too.
Ordinary consumer memory is not "quite reliable" if you care about data integrity or stability.
Google's study: http://research.google.com/pubs/pub35162.html (direct pdf link: http://static.googleusercontent.com/media/research.google.co... )
If you're a typical consumer, then of that data that is written, only a tiny fraction is user data, and of that user data, media usually represents most of the volume. Media is tolerant to a random bit flip.
Enterprise needs to worry about these things due to sheer scale and things like automatic propagation, but the odds of getting a bit flip in your tax spreadsheet or research paper and then writing it to disk are incredibly small, and personally I figure versioning/backups protect me from a significant portion of that (small) risk.
Which brings up disk crashes. Not to divert the conversation, but disk crashes are massively more dangerous than bit flips, and so many people haven't taken even the first steps against them...
Here's the tl;dr "(i.e. about 5 single bit errors in 8 Gigabytes of RAM per hour using the top-end error rate), and more than 8% of DIMM memory modules affected by errors per year."
And that's ~6-7 years old memory, which is less dense than modern chips, in a ideal temperature environment (the racks are water-cooled), down in a basement.
You average laptop is never going to register less than that, considering you're carrying it around with less environmental shielding, overheat it, and so on.
In fact, with a laptop with 64gb of memory that has gbs of paged data in memory for weeks, the chances are greater that your laptop with linux will crash due to a random bitflip that goes unnoticed, as opposed to a kernel bug.
And there's simply no way to tell. How about that?
These days I pay the "server" tax on the machines I build, buying Xeons just to get ECC. If they supported ECC, a normal i5 / i7 would probably be plenty, and the build would be less of a PITA as there would be more and cheaper motherboard choices with workstation basics like audio on board.
And yes, I use ZFS, and am otherwise pretty paranoid..
Intel has some of the best engineers in the world and make exceptional products. I wish the beancounters at Intel were a bit less into market segmentation.
24GB DDR2 ECC RAM - £16 (I was VERY lucky to get this one) 2x Xeon X5450 - £22 Intel S5000PALR Motherboard - £30
These specs are roughly equivalent to a 2008 mac pro, which costs around £500.
http://www.ebay.de/itm/HP-Z800-Workstation-2-x-Xeon-6-Core-X...
The Lenovo Thinkstation C20 is also a good dual Xeon choice if you can live with a 24GB RAM limit (it's also very compact).
There's also specialized resellers like ITSCO.de and ESM-Computer in Germany (they ship EU-wide).
For ~$1000, a dual X5670 workstation with 12 ECC RAM slots, I think it's an excellent deal.
If you want something newer, the Z820/Z620 are a good choice, they should start rapidly dropping in price soon enough.
These computers are designed extremely well (the 800+ Watt PSUs alone are eons ahead of anything on the consumer market), but even in the rare cases they break the parts are cheap (a replacement system board for ~$100, older but still powerful Xeons also drop in price insanely fast).
I don't know why people buy new midrange PCs for the same price anymore, unless they need the latest processors :-)
The high end X5670 or X5690 should outperform all but the top first gen Xeon E5.
It uses more power at idle, but the upfront cost is almost halved at the moment...
I'd say you should get a unit with the highest end processors you can find (it will end up way cheaper than upgrading them) and preferably with the 1100 Watt PSU, the rest can be upgraded later and you should have plenty of power for anything (dual GTX 9xx for gaming and stuff :-))
Obviously, the upfront cost of the older hardware is still better value when ignoring power consumption, but the v3 will also support much more and faster RAM, PCI-E 3.0, USB 3.0, and other newer technologies which will matter more as time goes on. You'll also have the option of a CPU only upgrade to higher end E5 v3's or even E5 v4's at a later point in time when those CPU models are more affordable.
The X5690 is comparable to the first generation E5-2640 (slightly slower), and the X5670 is comparable to the E5-2620.
Even the first gen E5 has the AVX instruction set, 4 memory channels (vs 3 for the Westmere-EP) and a faster QPI, so it might have better performance in certain applications.
v1 and v2 are socket compatible, but v3 is not socket compatible with the earlier versions, but will likely be compatible with v4 (Intel socket changes usually happen only every other generation). I'd suggest going all new with even a low end E5 2620v3 if you can, which you could for everything other than the GPUs for under 1700 (based on a Supermicro workstation build).
https://docs.google.com/spreadsheets/d/16x6xeENVLNySuVxpjfLx...
Notice the two tabs for CPUs and Motherboards.
Who said consumer use cases are OK with those?
I learned this the hard way when I bought a 2x16GB registered DDR3 kit, and then finding out my i5 won't boot with it. It was next to impossible to find unbuffered memory with 16GB per module last time I checked.
It's the type of ECC RAM you need for E5 & E7 Xeons - multiple physical CPUs. Many of these systems won't even boot with normal DDR3 sticks, ECC or otherwise. Must have registered DDR3 (and I suppose DDR4 now).
They do support unbuffered ECC, which is pretty great, but at least the v2 units I've got are limited to 32gb ram, which stinks.
http://i.imgur.com/7ZorPrq.jpg http://i.imgur.com/AnP4eVU.jpg
Aside, we started emailing Lenovo 2 years ago, when they ruined the touchpad & keyboard, to express our discontent. Seems that they will finally get back on track: http://blog.lenovo.com/en/blog/retro-thinkpad-survey-4-misce...
They might even support Coreboot. This is good. As a developer, I hate bad quality hardware. Apple is good quality, but some stuff is too consumer oriented: glossy screens and fragile alu+glass. And since Tiger I dislike their software.
For me it finally feels great to write -- low profile keys with great separation. I personally feel clumsy on most desktop keyboard because of their sloped keys.
I do wonder how these CPUs would do as processors for blades?
My bad.
Earlier HN discussion: https://news.ycombinator.com/item?id=8565119
So you might be right.
I sometimes dream of a world where consumer equipment has ECC memory. Some say: total overkill. I say: the stats on servers are so that you are probably hit multiple times a year with bit-flips in memory. You'll never know because it's not registered anywhere. But those apps hanging, this strange crash, a corrupted document. All these small things.
Servers aren't desktops/laptops, but I do wonder.
It's eye-opening once you've managed a few machines with ECC and seen first hand how common correction events are.
I'll be first in line to buy a laptop with ECC. Extra stability is worth it.
These days I'm playing with the idea of syncing a VM between my notebook and a powerful desktop PC for at desk use. Anyone tried it? Would it work? Virtualbox, maybe Docker?
Interesting, thanks for the input!