ASRock Brings Supercomputing to X58 Mobos

Status
Not open for further replies.

A Stoner

Distinguished
Jan 19, 2009
325
100
18,960
Wonder why they cannot get more PCIe pathways for full X16 on all 4 slots, if it is supposed to be a super computer, you would think it would need more bandwidth.
 

dacman61

Distinguished
Nov 5, 2007
62
0
18,630
Call me crazy, but what is so special about this board? It has all of the same X58 Chipset Specs as most of the other models out there today. The X58 chip basically determines the features of a motherboard regardless of the manufacturer. Maybe just a few different slot options and how they chop up the different x1 PCI-Express lanes and such.

Also, when are motherboard manufacturers going to stop putting in old PCI Slots? It's exactly like when they kept around those really old ISA slots back in the day.
 
G

Guest

Guest
[citation][nom]A Stoner[/nom]Wonder why they cannot get more PCIe pathways for full X16 on all 4 slots, if it is supposed to be a super computer, you would think it would need more bandwidth.[/citation]
Because there's just no chipset out there that supports this!
The speed is subject to the North/southbridge of a board.

I'm interested in just how much this setup will cost!
Seems to support upto 6x4GB DDR3!

So what else can you do with it, besides playing crysis and running the 'folding@home' project?
 
G

Guest

Guest
Old news are old. Tesla never needed any SLI enabled to work on X38. You could already build this last year with almost any X38 motherboard.
 

bustapr

Distinguished
Jan 23, 2009
1,613
0
19,780
Who would actually use 4 pci-e2.0 slots and for what. All I know is that for gaming people prefer putting 2 cards on sli and not 3. Plus a super computer would benefit more from more pci slots instead of an extra pci-e.
 
G

Guest

Guest
[citation][nom]dacman61[/nom]Call me crazy, but what is so special about this board? It has all of the same X58 Chipset Specs as most of the other models out there today. The X58 chip basically determines the features of a motherboard regardless of the manufacturer. Maybe just a few different slot options and how they chop up the different x1 PCI-Express lanes and such.Also, when are motherboard manufacturers going to stop putting in old PCI Slots? It's exactly like when they kept around those really old ISA slots back in the day.[/citation]
Because lots of users still use extension board audiocards like Soundblaster Audigy, network cards, etc... Most of them still use PCI.

I wondered if someone knew if the pci slots on a board use different lanes compared to PCIE slots?
I know one thing, that's PCIE reduces speed depending on the amount of slots that are used.
 
G

Guest

Guest
[citation][nom]Anonymous Coward[/nom]Old news are old. Tesla never needed any SLI enabled to work on X38. You could already build this last year with almost any X38 motherboard.[/citation]
Perhaps the extra memory bandwidth of this board (2000 oc, compared to 1600 on the X38) made the computer topple 1Teraflop of data, which technically put it into the supercomputer category?
 
G

Guest

Guest
[citation][nom]bustapr[/nom]Who would actually use 4 pci-e2.0 slots and for what. All I know is that for gaming people prefer putting 2 cards on sli and not 3. Plus a super computer would benefit more from more pci slots instead of an extra pci-e.[/citation]
Could be on a dual/quadcore system where the memory controller used part of the FSB speed.
There is a slight chance that the improved design of the corei7 frees up some bandwidth, and that using 3 or 4 cards could improve performance, something we could not see with tom's benchmark results on a Core2Quad Extreme.
 

dacman61

Distinguished
Nov 5, 2007
62
0
18,630
[citation][nom]ProDigit80[/nom]Because lots of users still use extension board audiocards like Soundblaster Audigy, network cards, etc... Most of them still use PCI.I wondered if someone knew if the pci slots on a board use different lanes compared to PCIE slots?I know one thing, that's PCIE reduces speed depending on the amount of slots that are used.[/citation]

No kidding! But I think it's time to move on.
 

gwolfman

Distinguished
Jan 31, 2007
782
0
18,980
[citation][nom]dacman61[/nom]Call me crazy, but what is so special about this board? It has all of the same X58 Chipset Specs as most of the other models out there today. The X58 chip basically determines the features of a motherboard regardless of the manufacturer. Maybe just a few different slot options and how they chop up the different x1 PCI-Express lanes and such.Also, when are motherboard manufacturers going to stop putting in old PCI Slots? It's exactly like when they kept around those really old ISA slots back in the day.[/citation]
It can use Xeon CPUs and ECC RAM, therefore a truer to the definition workstation/supercomputer.
 

gwolfman

Distinguished
Jan 31, 2007
782
0
18,980
[citation][nom]kittle[/nom]"supercomputer" with only 1 CPU slot?mabye im old-fashioned, but dont all supercomputers nowdays have more than 1 cpu socket in them?[/citation]
It will primarily use nVidia's CUDA for processing, not the CPU.
 

hellwig

Distinguished
May 29, 2008
1,743
0
19,860
This is NOT a gaming motherboard (at least, not as ASRock is marketing it).

Do people know what Tesla is? The "supercomputer" part comes from the fact that you can pack a bunch of Nvidia GPUs in this sucker, and use them to do the computation (not the CPU). The frequency (600+MHz) and sheer number of computational cores (128-256) mean a single GPU can perform many more floating-point operations per second than even a modern 6-core Xeon. Throw 4 such GPUs in a single case, and you have one powerful floating point machine. I don't think any motherboard by any manufacturer supports enough CPU sockets to get that kind of performance.
 
The GPU architecture is going to make this NOT a true "Supercomputer". It's more like a HPC.

The first is that of memory latency; GPUs operate with a very high degree of latency on the memory; since they're handling relatively linear tasks, and when dealing with textures and shaders, always call up very large, sequential blocks of memory at a time, having a CAS latency of 30+, 40+, or more clock cycles doesn't really matter, since the GPU will know much farther in advance what it'll be needing next around 99% of the time. The same benefit can be applied to decoding media; being a streaming application, latency doesn't hurt it. However, when it comes to scientific applications, that really can be harmful, as in those cases the predominant bottleneck invariably winds up being data and instruction latency, something that's also hurt heavily by how GPUs have an extremely skewed processing unit-to-cache ratio, a ratio that's vastly different than what's found in general-purpose CPUs.

The second reason that occurred to me is the lack of a standard multi-GPU architecture that would be able to support a large quantity of GPUs even just for mathematic operations; the current limit for ANY design appears to be 4 GPUs, from either nVidia or ATi/AMD. So, while yes, while in theory you could produce the same floating-point capacity using only 1/7.5th the number of RV770s compared to what Sequoia uses (i.e, 13.3% the number) as of yet, there is no way to actually build that assembly, so in practice, it's a moot point.

The final reason is actually that of power and heat; GPUs may have a very high degree of performance-per-watt efficiency when it comes to math, but they STILL have a very high TDP per chip. The cost of the actual chips are usually one of the minor parts of a supercomputer, as a lot more care has to be given to providing enough power to stably run thousands upon thousands of nodes, with not just multiple CPUs per node, but all the other components as well, all of which must be powered and cooled. With GPUs, you're going to have your heat production focused on a far smaller number of chips, so you'll need to actually have more intensive cooling, and likely greater spacing between GPUs, since you can't just blow hot air out the back of the case, since there will be more nodes in every direction. There's a good chance that one would actually have to construct a LARGER facility to house an equally-powerful supercomputer built from GPUs than one built from multi-core general-purpose CPUs.
nottheking
http://www.tomshardware.com/news/IBM-Sequoia-Supercomputer,6955.html
 

Tindytim

Distinguished
Sep 16, 2008
1,179
0
19,280
[citation][nom]bustapr[/nom]Who would actually use 4 pci-e2.0 slots and for what. All I know is that for gaming people prefer putting 2 cards on sli and not 3. Plus a super computer would benefit more from more pci slots instead of an extra pci-e.[/citation]
If you really had money to burn, you could do a Tri-SLI setup, then get a forth card for Physics.

However, it is for number crunching.
 

random1283

Distinguished
Oct 26, 2007
222
0
18,680
Hmmm soz if you read my other post but the ASUS P6T WS 6 x pcie
and an high level onboard raid controller. and these boards are mainly for CUDA and Such As
 

nottheking

Distinguished
Jan 5, 2006
1,456
0
19,310
Well, it looks like some of my words made it here before I did... So I guess there might not really be all that much for me to say.

At any rate, if I remember correctly, the GTX 280, of which it appears the Tesla C1060 card is effectively the same as, has a peak theoretical floating-point capability of 936 gigaFLOps. Since as I mentioned prior, (in what Shadow quoted) the limit for a single system is 4 GPUs, (for a total of 3.888 TFLOps) and hence it's no surprise that the bulk of units that are based on these Tesla chips, it seems, come with four of them, with none having any more. Most of the units I see are packed into tower-style PC cases, and simply equip 4 Tesla cards. The exception is nVidia's own Tesla S1070, which comes as a 1U rackmount piece, which suggests that mounts the GPUs in a different PCB format.

I'll admit, this sort of board release really isn't all that unique; as the third-party Tesla units have shown, you can already buy an OEM version of what you could assemble using this motherboard. It's a fairly familiar case of a previously workstation-only part making its way to availability for the home market; I'm reminded of AMD's Quad FX and Intel's Skulltrail platforms, which weren't exactly novel on the workstation market, which have had dual-CPU motherboards for decades.

No, this really isn't a "supercomputer," and nVidia appears to make to implication that such a platform is. (though they have backed the "personal supercomputer" made by third parties) Again, I don't believe that there really is no existing architecture that would allow you to link these units up as individual nodes in a single supercomputer, so this would be chiefly for HPC applications, which granted, is way above the level of what home users (and even the vast bulk of enthusiasts) are doing. The only thing that slightly irks me is that nVidia has opted, on their own page for Tesla, to call it "the world's first Teraflop processor," a claim it really can't back up; if you go by outright chip, RV770 hit shelves well before GTX 200, and I believe even the FireStream 9250 version did as well, giving it a far more defendable claim to being the "first teraflop processor."
 
Status
Not open for further replies.