Tuesday, May 17th 2011

New NVIDIA Tesla GPU Smashes World Record in Scientific Computation

NVIDIA today unveiled the Tesla M2090 GPU, the world's fastest parallel processor for high performance computing. In addition, the Tesla M2090 GPU achieved the fastest-ever performance in a key measure of scientific computation. Equipped with 512 CUDA parallel processing cores, the Tesla M2090 GPU delivers 665 gigaflops of peak double-precision performance, enabling application acceleration by up to 10x compared to using a CPU alone.

In the latest version of AMBER 11, one of the most widely used applications for simulating behaviors of biomolecules, four Tesla M2090 GPUs coupled with four CPUs delivered record performance of 69 nanoseconds of simulation per day. The fastest AMBER performance recorded on a CPU-only supercomputer is 46 ns/day.

"This is the fastest result ever reported. With Tesla M2090 GPUs, AMBER users in university departments can obtain application performance that outstrips what is possible even with extensive supercomputer access," said Ross Walker, assistant research professor at the San Diego Computer Center, and principle contributor to the AMBER code.

The Tesla M2090 GPU will be available in servers such as the new HP ProLiant SL390 G7 4U server. As part of the SL6500 Scalable System of HP server solutions, optimized for scale-out and HPC market segments, the SL390 family was purpose built for hybrid computing environments that combine GPUs and CPUs.

The SL390 G7 4U server incorporates up to eight Tesla M2090 GPUs in a half-width 4U chassis and, with a configuration of eight GPUs to two CPUs, offers the highest GPU-to-CPU density on the market. The system is ideally suited for applications ranging from quantum chemistry and molecular dynamics to seismic processing and data analytics.

"Clients running intensive data center applications require systems that can process massive amounts of complex data quickly and efficiently," said Glenn Keels, director of marketing, Hyperscale business unit, HP. "The decade long collaboration between HP and NVIDIA has created one of the industry's fastest CPU to GPU configurations available, delivering clients the needed processing power and speed to handle the most complex scientific computations."

In addition to AMBER, the Tesla M2090 GPU is ideally suited to a wide range of GPU-accelerated HPC applications. These include molecular dynamics applications, NAMD and GROMACS, computer-aided engineering applications, ANSYS Mechanical, Altair Acusolve and Simulia Abaqus, earth science applications, WRF, HOMME and ASUCA, oil and gas applications, Paradigm Voxelgeo and Schlumberger Petrel as well as other key applications such as MATLAB, GADGET2 and GPU-BLAST.

For more information on the HP ProLiant SL390 G7 4U server, go here, and for more information on the Tesla M2090 and the whole family of M-class products, please go here.
Add your own comment

25 Comments on New NVIDIA Tesla GPU Smashes World Record in Scientific Computation

#1
Sasqui
What is a "ns" as in "46 ns/day" ?

Edit, Nevermind... "nanoseconds of simulation per day"
Posted on Reply
#2
AsRock
TPU addict
I am more interested in that cooler lol.
Posted on Reply
#3
WhiteLotus
Can somebody please tell me again why you can't use these to play games?
Posted on Reply
#4
Sasqui
by: AsRock
I am more interested in that cooler lol.
Was thinking that too.

by: WhiteLotus
Can somebody please tell me again why you can't use these to play games?
Because there is no DVI or HDMI connector :laugh:
Posted on Reply
#5
cheesy999
by: Sasqui
Because there is no DVI or HDMI connector
i was gonna say because at the price of this thing you could probably go dual proccesor and quad SLI
Posted on Reply
#6
btarunr
Editor & Senior Moderator
by: cheesy999
i was gonna say because at the price of this thing you could probably go dual proccesor and quad SLI
At its price you can build an entire gaming rig that can run games faster than it, if it had a display output.
Posted on Reply
#7
cheesy999
by: btarunr
At its price you can build an entire gaming rig that can run games faster than it, if it had a display output.
that's what i mean, a dual lga1366 build with dual 590
Posted on Reply
#8
LAN_deRf_HA
I've gotten the impression that workstations cards have gotten better at playing games compared to their consumer counterparts, at least with the right drivers. However the last time I was able to see that put to the test was back with the G80s. Nobody bothers to test these things like that.
Posted on Reply
#9
Sasqui
by: LAN_deRf_HA
I've gotten the impression that workstations cards have gotten better at playing games compared to their consumer counterparts, at least with the right drivers. However the last time I was able to see that put to the test was back with the G80s. Nobody bothers to test these things like that.
In most cases, the workstation card drivers are optimized for OpenGL, and specialty products, usually not DirectX. It's the reason they cost so darn much because the user base is so much smaller and the economy of scale in writing drivers just isn't there. From what I've seen the ATI and NV offerings are slight one offs from the consumer cards (in some cases, only very slightly). So, depending on the game 3D display means, the workstation cards will be about the same, or better in games.
Posted on Reply
#10
wolf
Performance Enthusiast
by: btarunr
At its price you can build an entire gaming rig that can run games faster than it, if it had a display output.
by: cheesy999
that's what i mean, a dual lga1366 build with dual 590
you wouldn't even need that much, a 2600K and two GTX570's or 6950's would cream a single Tesla M2090 for cheaper.
Posted on Reply
#12
damric
Are these GTX 580's with a different BIOS and drivers?
Posted on Reply
#13
Easy Rhino
Linux Advocate
by: damric
Are these GTX 580's with a different BIOS and drivers?
no, they are completely different. like apples and oranges really.
Posted on Reply
#14
Damn_Smooth
So basically they're saying that they achieved better performance by adding a gpu?

I could have told them that.
Posted on Reply
#15
Kreij
Senior Monkey Moderator
These are not graphics cards ... at all.
The suite of AMBER program calculates molecular force fields and molecular trajectories of interacting molecules within a given fixed field.
The amount of data calculation to see what happens in a fixed multi-molecular field in 1 nanosecond is astronomical. The computations for single molecule interaction in staggering.
Doing 69ns in a 24 hour period is a real achievement, and I applaude Nvidia on their continued advancement into supercomputing.

Don't expect to see this in Duke Nukem Forever.
Posted on Reply
#16
hellrazor
by: AsRock
I am more interested in that cooler lol.
Looks like a wider version of the one on my CPU.
Posted on Reply
#17
damric
by: Easy Rhino
no, they are completely different. like apples and oranges really.
But it is technically a Fermi GF 110, is it not? Just without display outputs?

Edit: I found this. Looks like it is indeed a GF 110 core. I wonder if someone could take a GTX 580 and manipulate it into one of these new Tesla cards.


http://www.nvidia.com/docs/IO/105880/DS_Tesla-M2090_LR.pdf
Posted on Reply
#18
1c3d0g
Is this the new Folding@Home King? Either way, thumbs up, NVIDIA! :cool:
Posted on Reply
#19
bear jesus
Always good to see advancing computer tech help advance science.

But as this has four Tesla M2090 GPU's coupled with four CPU's and pumped out 69 nanoseconds of simulation in 24 hours, assuming perfect scaling how many would be needed to simulate in real time? :laugh: it's 6am so I'm not doing the math.
Posted on Reply
#20
caleb
Don't expect to see this in Duke Nukem Forever.
Why not it can be tessellation 10.0 :)
Posted on Reply
#21
vanyots
by: bear jesus
Always good to see advancing computer tech help advance science.

But as this has four Tesla M2090 GPU's coupled with four CPU's and pumped out 69 nanoseconds of simulation in 24 hours, assuming perfect scaling how many would be needed to simulate in real time? :laugh: it's 6am so I'm not doing the math.
1 252 173 913 043.5 according to my PC calculator:eek::eek::eek:
(that's 1.25 trillion systems like the one in question)
Posted on Reply
#22
bear jesus
by: vanyots
1 252 173 913 043.5 according to my PC calculator:eek::eek::eek:
:eek: OK we are probably quite a few generations away, probably after we hit the wall with current chip tech.
Posted on Reply
#23
MikeX
That came with a 6GB on board. Ten times more expensive than a Gtx 580.
My logical GPGPU solution would be get the gtx 590 and 12GB kit. You can share system's memory anyway.
Posted on Reply
#24
Steevo
I dont know if we will ever be able to compute real time life. With the limitations of size the only possible way would be quantum processing.
Posted on Reply
#25
bear jesus
by: Steevo
I dont know if we will ever be able to compute real time life. With the limitations of size the only possible way would be quantum processing.
I think it greatly depends where we go after current silicone chip tech, imagine if those graphene transistors IBM made that can scale up to 155ghz or the indium phosphide and indium gallium arsenide transistor that was made at the university of Illinois that hit 845ghz could be used with current/future architecture and at those speeds, how much power do you think a fermi card would have if the core could run at like 155ghz to 845ghz.

Yes i know it does not really work like that and the max transistor speed does not mean how fast the transistors can work in a chip but my point is there is a bunch of new tech that could replace current form silicone chips that could give one hell of a speed boost, personally i can't wait for us to hit the wall down around 8 or 10nm or whatever it will be.
Posted on Reply
Add your own comment