The situation has improved a lot on the software front, but I agree it's pretty hopeless / useless for gaming. This is for people with more money than sense.
This would _barely_ make sense if they limited it to 6-8-core high clock speed CPUs (which ... well, Xeons aren't the latter, at least) and had software in place that automatically limited, say, CPU1 to games only (with an exhaustive list of games and their process and executable names, I suppose) and left everything else to CPU0. Of course you'd also (likely) need to make sure that the GPU(s) is(/are) seated in PCIe slots linked to the correct CPU at that point, making everything even more complicated. No thanks. Wouldn't buy this even if I could afford it.
I'd really like to know how one would move from MT to MPI considering that for situations that are highly MT one can get way more than 64 threads = number of CPU cores. Would also be terrible to scale as you'd work for a specific configuration each time.
Current solution is fine, the problem is more that tasks can only scale that much, and gaming currently still needs that one thread that deals with the game logic which requires a really fast single core.
The point is probably to have name and brand recognition from building and selling a few silly, but interesting on paper halo product systems to people who don't understand what they're purchasing. There's the secondary goal of selling said halo product with an absurd markup so the profit margin looks juicy-delicious to the corporate bobbleheads that approved it.
If they're smart about it, they can actually try to compete against Supermicro workstations in this space.
It's too bad that AMD Epyc hasn't caught up to the 2nd gen TR processors yet, because then it could be interesting to have a quad dual-socket 32-core AMD Epyc (256-cores/512-threads total) in a 2U rackmount because I can use up all of those cores doing what I do.
(My main central server has 80 TB raw storage, 70 TB RAID5.)
"Consumer-grade" and "Enterprise-grade" is somewhat a misnomer.
An i7 is not significantly more likely to die than it's Xeon counterpart. The motherboard VRM's might, but this is a case-specific scenario that is determined by the "workstation" manufacturer (For example Dell Precision's use the same VRM's whether you get a Xeon or i7/i9).
No idea what you are talking about them being able to handle larger datasets, unless you are talking about Skylake-SP, which is basically Server only.
This is 2019. I think , that though so called enterprise-grade hardware may be in general better than consumer-grade hardware., there's much hype in so called "professional" applications and machines. I am still rocking a core i7 920 and would still be using an ATI-Radeon 4870 if AMD didn't render it obsolete by messing with the supporting software. The hard disk works flawlessly from that day, and so do the ram sticks. My ancient equipment is definitely consumer-grade and with low-credentials assembled in a lowly local business. Still it is chugging along for almost a decade like nothing happened. Conclusion is, there's too much hype in professional hardware; which is understandable , every business wants to make the bigger possible profits with a given product; but not acceptable when concerning my point of view and my purchasing decisions.
That case looks suspiciously similar to a $50 Corsair case I've got...
I'm also *really* curious who buys these workstations with gamer aesthetics. I don't see much of a point of the aesthetic on gaming computers, but especially not workstations with professional ambitions.
I suspect they only sell 1 or 2 hand built boxes to the same sort of idiots who'll buy a Bugatti Veyron to go grocery shopping (as opposed to those who buy it for driving on a race track and dreaming of working for nascar/f1/etc). The actual cost of design/testing is probably billed to marketing since these exist just to generate attention by the press.
There are ppl with a lot more money than knowledge of PC's. For instance, I paid $1k to by a Core 2 QX9650 (Extreme) back in 2008 instead of a Q9650 for $300 b/c I thought (mistakenly) that the QX9650 would perform better than a Q9650 in gaming. Bottomline, this is a ripoff and they may very well sell it to ppl who have so much money that they don't know what to do with or ppl who just don't understand PC's.
This system is one of the few where it is likely to be a big issue, as most consumers run Windows. But if anything, the presence of such systems might kick Microsoft (and perhaps even game engine devs) into improving their NUMA handling by the time Cascade Lake-SP comes out.
"they really need to have some software wrapper in place to enumerate cores and put Core Affinity in place"
Process Lasso can already do this on the fly. However, it's a bit cumbersome to find and set optimal configurations for different programs (i.e. steam library).
There are some other uses for these systems besides just ray-tracing. Dual CPU systems are pretty standard issue for programmers in game development for example, but that may no longer be necessary in the future with more high core count, high-clockspeed processors available.
I'm interested to see how high Intel can push the frequency on their 28-core i9, and the Zen 2 Epycs should be interesting as well next year.
The problem with the 32-core TR2 right now is that the indirect access to memory is killing performance in a lot of tasks - including code compiling on Windows - and that is a deal-breaker for my use-case.
Well, in gaming (or generally GPU coding) you rather need dual gpu... when you start debugging shaders, your card can hardly render visual studio (at least it was the case a while back)... one of few places where iGPU comes handy. But dual CPU... what for?
Assuming to allow hot loading or something like that? Keep game on one CPU, run the rest of it on the other including compiling, with e.g. unreal engine detecting the newly compiled dll.
Content creators largely avoid Windows for the reasons you described. Games largely avoid Linux because it has few native AAA games and the rest need to be run via Wine, its recent Proton fork by Valve or a virtual machine, with a performance loss.
Windows has crappy NUMA support (which is what you imply but do not directly state here), and of course I am talking about its more premium versions, including Windows Server, not... Home. Linux and the various flavors of BSD have great NUMA support in all their versions*.
So a machine like that from Acer cannot have it both ways. They will either lose gamers or content creators, but cannot satisfy both with a single machine and a single OS.
*Michael Larabel at Phoronix ran a benchmark a few days ago comparing Threadripper 2990WX under Linux (Ubuntu 18.4., i.e. a "Windows Home" Linux equivalent) versus the latest Windows Server. The benchmark used many CCX variations and active core counts of 2900WX, from 4 cores to 32 cores. Windows Server was decimated in all of them.
Nah. They don't look like they want to sell it. Just look at the PC case, it looks like an entry level desktop gaming box. It has no ventilation at all except a restrictive 120mm mesh at the back.
That is the reason I'd like them to scale GPGPU or APU clusters rather than thowing CPU cores and GPU cores into a big, big compute lake that just cooks into steam with evaporating returns.
If you take the type of SoCs AMD produced for the Chinese, perhaps use HBM2 for the HSA parts and add a bit of DDRx for "storage", you'd make the job of the programmer much, much easier to scale the game-engine or the HPC.
Hardware independent coding is dead, because More's law is dead, so if close to the metal coding is required anyway, at least make it easier by using "Lego block" hardware.
Lots of people have use cases for dual-socket systems and dual-socket workstations aren't uncommon.
> "without a NUMA-aware operating system or software in place, memory for one process on one CPU can be allocated on the memory of the other CPU, causing additional latency."
CPU affinity and process groups is completely configurable. There doesn't have to be any additional latency if you don't want it but if you want your application to access all your memory you have to accept the physical fact of life that memory further away will be slower than memory closer. This is a perfectly acceptable trade-off for many applications.
> "We tested this way back in 2013, and the situation has not improved since."
Pretty sure it has improved. Interconnects are faster and memory is lower latency now than it was in 2013 and software has come a long way as well largely thanks to Threadripper 1 which was the first high end desktop chip to bring this server focus architecture to the desktop in a single socket.
> "Most software assumes all the cores and memory are identical, so adding additional latency causes performance to tank. Tank hard"
Define "most software". Do you mean games? Or do you mean operating system kernels, web servers, rendering applications, databases, physics simulations, AI/deep learning training? Most machines in the data center are dual-socket and if this caused any significant performance degradation that would not be the case.
Symmetric Multiprocessing goes back to 1962 and developers are pretty damn good and making software work over multiple CPUs (when they have to). If an application doesn't scale over dozens of cores it's not the fault of the hardware.
> "even scientific software was not built for multi-CPU hardware, and often performed worse than a single CPU with fewer cores"
File a bug report and in the meantime limit that application to a single CPU.
> " the 32-core Threadripper 2 show performance deficits against monolithic solutions"
The 2990WX is a weird case. It's a server CPU put on a consumer board and is starved of memory bandwidth in some cases. That said you can configure it to act like a single die. Of course this system isn't using a 2990WS it's using Xeon 8180s which has a two more memory channels.
I agree this is insane. I have owned two dual socket machines and that added latency for memory synchronization really kills gaming performance. Multi-socket only makes sense for virtualization, where you can put one virtual machine on each physical CPU chip.
The title of this article does the math beautifully:
"Dual Xeon Processor" (expensive as hell with limited server applications) + "Predator X System" (high-end gaming consumer targeted product name) = "Please no" (because we know from Skull Trail this will end badly)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
37 Comments
Back to Article
TheDoctor46 - Wednesday, August 29, 2018 - link
Lol! Loved the way u described it, esp the leather thing;) what were they thinking!boozed - Wednesday, August 29, 2018 - link
Just the perfect amount of sarcasmxchaotic - Wednesday, August 29, 2018 - link
The situation has improved a lot on the software front, but I agree it's pretty hopeless / useless for gaming. This is for people with more money than sense.Valantar - Wednesday, August 29, 2018 - link
This would _barely_ make sense if they limited it to 6-8-core high clock speed CPUs (which ... well, Xeons aren't the latter, at least) and had software in place that automatically limited, say, CPU1 to games only (with an exhaustive list of games and their process and executable names, I suppose) and left everything else to CPU0. Of course you'd also (likely) need to make sure that the GPU(s) is(/are) seated in PCIe slots linked to the correct CPU at that point, making everything even more complicated. No thanks. Wouldn't buy this even if I could afford it.alpha754293 - Wednesday, August 29, 2018 - link
Depends on the motherboard.Supermicro boards are super robust (plus the built-in IPMI is nice so that you don't need a separate, external KVMoIP).
It's too bad that most programs are actually written for multi-threading (e.g. OpenMP/SMP) rather than truly multi-processing (MPI).
RSAUser - Friday, August 31, 2018 - link
I'd really like to know how one would move from MT to MPI considering that for situations that are highly MT one can get way more than 64 threads = number of CPU cores. Would also be terrible to scale as you'd work for a specific configuration each time.Current solution is fine, the problem is more that tasks can only scale that much, and gaming currently still needs that one thread that deals with the game logic which requires a really fast single core.
duploxxx - Wednesday, August 29, 2018 - link
just buy a HPinc z6 -z8 workstation and have the same....they should rather bring something new to the market, Threadripper based station for half the price i.s.o the Xeon Bullshit
alpha754293 - Wednesday, August 29, 2018 - link
Consumer-grade hardware have a SIGNIFICANTLY higher rate of failure than server/enterprise-grade hardware.And AMD hasn't released the 2nd gen TR in their Epyc line. (Their Epyc line is still based on first gen TR.)
eek2121 - Sunday, September 2, 2018 - link
Threadripper != EPYC. EPYC is a completely different stepping from Threadripper, features more PCIE lanes and more memory channels.piroroadkill - Wednesday, August 29, 2018 - link
What's the point?Just use Threadripper 2950X and call it a day...
PeachNCream - Wednesday, August 29, 2018 - link
The point is probably to have name and brand recognition from building and selling a few silly, but interesting on paper halo product systems to people who don't understand what they're purchasing. There's the secondary goal of selling said halo product with an absurd markup so the profit margin looks juicy-delicious to the corporate bobbleheads that approved it.alpha754293 - Wednesday, August 29, 2018 - link
Again, it depends.If they're smart about it, they can actually try to compete against Supermicro workstations in this space.
It's too bad that AMD Epyc hasn't caught up to the 2nd gen TR processors yet, because then it could be interesting to have a quad dual-socket 32-core AMD Epyc (256-cores/512-threads total) in a 2U rackmount because I can use up all of those cores doing what I do.
(My main central server has 80 TB raw storage, 70 TB RAID5.)
Diji1 - Wednesday, August 29, 2018 - link
>silly, but interesting on paper halo product systems to people who don't understand what they're purchasing.That just seems silly.
No one is spending this amount of money and not knowing what they're purchasing.
And somehow you already know their requirements before they do which is also silly.
alpha754293 - Wednesday, August 29, 2018 - link
The point is that consumer-grade hardware shouldn't even compete in the space that is dominated by enterprise-grade hardware.My decade-and-a-half old AMD Opterons are still running. My Core i7-3930K died and had to be replaced.
I'm slowly in the process of changing out all of my consumer-grade hardware based workstations over from Core i7s to Xeons.
Costs more, but they also are SIGNIFICANTLY less likely to die and can handle MUCH, MUCH, MUCH larger datasets.
This is the point.
diehardmacfan - Wednesday, August 29, 2018 - link
"Consumer-grade" and "Enterprise-grade" is somewhat a misnomer.An i7 is not significantly more likely to die than it's Xeon counterpart. The motherboard VRM's might, but this is a case-specific scenario that is determined by the "workstation" manufacturer (For example Dell Precision's use the same VRM's whether you get a Xeon or i7/i9).
No idea what you are talking about them being able to handle larger datasets, unless you are talking about Skylake-SP, which is basically Server only.
IUU - Saturday, June 8, 2019 - link
This is 2019. I think , that though so called enterprise-grade hardware may be in general better than consumer-grade hardware., there's much hype in so called "professional" applications and machines. I am still rocking a core i7 920 and would still be using an ATI-Radeon 4870 if AMD didn't render it obsolete by messing with the supporting software. The hard disk works flawlessly from that day, and so do the ram sticks. My ancient equipment is definitely consumer-grade and with low-credentials assembled in a lowly local business. Still it is chugging along for almost a decade like nothing happened. Conclusion is, there's too much hype in professional hardware; which is understandable , every business wants to make the bigger possible profits with a given product; but not acceptable when concerning my point of view and my purchasing decisions.Inteli - Wednesday, August 29, 2018 - link
That case looks suspiciously similar to a $50 Corsair case I've got...I'm also *really* curious who buys these workstations with gamer aesthetics. I don't see much of a point of the aesthetic on gaming computers, but especially not workstations with professional ambitions.
DanNeely - Wednesday, August 29, 2018 - link
I suspect they only sell 1 or 2 hand built boxes to the same sort of idiots who'll buy a Bugatti Veyron to go grocery shopping (as opposed to those who buy it for driving on a race track and dreaming of working for nascar/f1/etc). The actual cost of design/testing is probably billed to marketing since these exist just to generate attention by the press.matfra - Wednesday, August 29, 2018 - link
Thanks for trashing Acer, I don't feel alone in this world anymore.Achaios - Wednesday, August 29, 2018 - link
There are ppl with a lot more money than knowledge of PC's. For instance, I paid $1k to by a Core 2 QX9650 (Extreme) back in 2008 instead of a Q9650 for $300 b/c I thought (mistakenly) that the QX9650 would perform better than a Q9650 in gaming. Bottomline, this is a ripoff and they may very well sell it to ppl who have so much money that they don't know what to do with or ppl who just don't understand PC's.darcotech - Wednesday, August 29, 2018 - link
Then you do not know the power of brainless with deeeep pockets.That is their target. Leather is the best proof.
ImSpartacus - Wednesday, August 29, 2018 - link
Aren't the RTX 8000s $10k each?https://www.anandtech.com/show/13217/nvidia-announ...
Dug - Wednesday, August 29, 2018 - link
That case!!! That's what dreams are made of.MrSpadge - Wednesday, August 29, 2018 - link
You mean nightmares?GreenReaper - Wednesday, August 29, 2018 - link
NUMA can be a real issue, but the level of that impact depends in part on the OS and app. For example, AnandTech saw more of an impact than Phoronix as it tests on Windows, not Linux:https://www.phoronix.com/vr.php?view=26716
https://www.phoronix.com/vr.php?view=26705
https://www.phoronix.com/scan.php?page=article&...
This system is one of the few where it is likely to be a big issue, as most consumers run Windows. But if anything, the presence of such systems might kick Microsoft (and perhaps even game engine devs) into improving their NUMA handling by the time Cascade Lake-SP comes out.
MrSpadge - Wednesday, August 29, 2018 - link
"they really need to have some software wrapper in place to enumerate cores and put Core Affinity in place"Process Lasso can already do this on the fly. However, it's a bit cumbersome to find and set optimal configurations for different programs (i.e. steam library).
twtech - Wednesday, August 29, 2018 - link
There are some other uses for these systems besides just ray-tracing. Dual CPU systems are pretty standard issue for programmers in game development for example, but that may no longer be necessary in the future with more high core count, high-clockspeed processors available.I'm interested to see how high Intel can push the frequency on their 28-core i9, and the Zen 2 Epycs should be interesting as well next year.
The problem with the 32-core TR2 right now is that the indirect access to memory is killing performance in a lot of tasks - including code compiling on Windows - and that is a deal-breaker for my use-case.
HollyDOL - Thursday, August 30, 2018 - link
Well, in gaming (or generally GPU coding) you rather need dual gpu... when you start debugging shaders, your card can hardly render visual studio (at least it was the case a while back)... one of few places where iGPU comes handy.But dual CPU... what for?
RSAUser - Friday, August 31, 2018 - link
Assuming to allow hot loading or something like that? Keep game on one CPU, run the rest of it on the other including compiling, with e.g. unreal engine detecting the newly compiled dll.Santoval - Wednesday, August 29, 2018 - link
Content creators largely avoid Windows for the reasons you described. Games largely avoid Linux because it has few native AAA games and the rest need to be run via Wine, its recent Proton fork by Valve or a virtual machine, with a performance loss.Windows has crappy NUMA support (which is what you imply but do not directly state here), and of course I am talking about its more premium versions, including Windows Server, not... Home. Linux and the various flavors of BSD have great NUMA support in all their versions*.
So a machine like that from Acer cannot have it both ways. They will either lose gamers or content creators, but cannot satisfy both with a single machine and a single OS.
*Michael Larabel at Phoronix ran a benchmark a few days ago comparing Threadripper 2990WX under Linux (Ubuntu 18.4., i.e. a "Windows Home" Linux equivalent) versus the latest Windows Server. The benchmark used many CCX variations and active core counts of 2900WX, from 4 cores to 32 cores. Windows Server was decimated in all of them.
zodiacfml - Thursday, August 30, 2018 - link
Nah. They don't look like they want to sell it.Just look at the PC case, it looks like an entry level desktop gaming box. It has no ventilation at all except a restrictive 120mm mesh at the back.
s.yu - Friday, August 31, 2018 - link
That thing looks impossibly tacky and cheap for all the high end components.abufrejoval - Friday, August 31, 2018 - link
That is the reason I'd like them to scale GPGPU or APU clusters rather than thowing CPU cores and GPU cores into a big, big compute lake that just cooks into steam with evaporating returns.If you take the type of SoCs AMD produced for the Chinese, perhaps use HBM2 for the HSA parts and add a bit of DDRx for "storage", you'd make the job of the programmer much, much easier to scale the game-engine or the HPC.
Hardware independent coding is dead, because More's law is dead, so if close to the metal coding is required anyway, at least make it easier by using "Lego block" hardware.
abufrejoval - Friday, August 31, 2018 - link
Or quite simply imagine a Threadripper made up of 2700G dies with HBM2.kneelbeforezod - Saturday, September 1, 2018 - link
Lots of people have use cases for dual-socket systems and dual-socket workstations aren't uncommon.> "without a NUMA-aware operating system or software in place, memory for one process on one CPU can be allocated on the memory of the other CPU, causing additional latency."
CPU affinity and process groups is completely configurable. There doesn't have to be any additional latency if you don't want it but if you want your application to access all your memory you have to accept the physical fact of life that memory further away will be slower than memory closer. This is a perfectly acceptable trade-off for many applications.
> "We tested this way back in 2013, and the situation has not improved since."
Pretty sure it has improved. Interconnects are faster and memory is lower latency now than it was in 2013 and software has come a long way as well largely thanks to Threadripper 1 which was the first high end desktop chip to bring this server focus architecture to the desktop in a single socket.
> "Most software assumes all the cores and memory are identical, so adding additional latency causes performance to tank. Tank hard"
Define "most software". Do you mean games? Or do you mean operating system kernels, web servers, rendering applications, databases, physics simulations, AI/deep learning training? Most machines in the data center are dual-socket and if this caused any significant performance degradation that would not be the case.
Symmetric Multiprocessing goes back to 1962 and developers are pretty damn good and making software work over multiple CPUs (when they have to). If an application doesn't scale over dozens of cores it's not the fault of the hardware.
> "even scientific software was not built for multi-CPU hardware, and often performed worse than a single CPU with fewer cores"
File a bug report and in the meantime limit that application to a single CPU.
> " the 32-core Threadripper 2 show performance deficits against monolithic solutions"
The 2990WX is a weird case. It's a server CPU put on a consumer board and is starved of memory bandwidth in some cases. That said you can configure it to act like a single die. Of course this system isn't using a 2990WS it's using Xeon 8180s which has a two more memory channels.
SvenNilsson - Sunday, September 16, 2018 - link
I agree this is insane. I have owned two dual socket machines and that added latency for memory synchronization really kills gaming performance. Multi-socket only makes sense for virtualization, where you can put one virtual machine on each physical CPU chip.woogitboogity - Saturday, June 13, 2020 - link
The title of this article does the math beautifully:"Dual Xeon Processor" (expensive as hell with limited server applications)
+
"Predator X System" (high-end gaming consumer targeted product name)
=
"Please no" (because we know from Skull Trail this will end badly)