Nvidia Betting its CUDA GPU Future With 'Fermi'
This chip is going to be huge for the supercomputing market -- if Nvidia's has its way.
The video card has evolved now to be termed the GPU, thanks to the growing capability of the hardware. Now the GPU is about to take its next big leap to becoming specialized GPGPU (of course, we realize that the term specialized and general purpose are some what contradictory).
Nvidia is betting heavily on GPGPUs becoming a large need in the computing market. While we'll still need our GPUs to push our pixels for our 3D games, Nvidia has just revealed its next-generation CUDA architecture, codenamed "Fermi."
Nvidia bills Fermi as an entirely new ground-up design that will finally realize the potential of GPU computing. Although Nvidia made big steps with its G80 and later the GT200, the graphics maker has made Fermi a much more pleasant and useful tool for programmers.
“The first two generations of the CUDA GPU architecture enabled Nvidia to make real in-roads into the scientific computing space, delivering dramatic performance increases across a broad spectrum of applications,” said Bill Dally, chief scientist at Nvidia.
“It is completely clear that GPUs are now general purpose parallel computing processors with amazing graphics, and not just graphics chips anymore,” said Jen-Hsun Huang, co-founder and CEO of Nvidia. “The Fermi architecture, the integrated tools, libraries and engines are the direct results of the insights we have gained from working with thousands of CUDA developers around the world. We will look back in the coming years and see that Fermi started the new GPU industry.”
At the unveil event, Nvidia did not give anything away in terms of clock speeds or any of the other specifications that hardcore 3D gamers focus on. Instead, it talked about technical features that lend themselves specifically for GPU computing. Such technologies include:
- C++, complementing existing support for C, Fortran, Java, Python, OpenCL and DirectCompute.
- ECC, a critical requirement for datacenters and supercomputing centers deploying GPUs on a large scale
- 512 CUDA Cores featuring the new IEEE 754-2008 floating-point standard, surpassing even the most advanced CPUs
- 8x the peak double precision arithmetic performance over Nvidia’s last generation GPU. Double precision is critical for high-performance computing (HPC) applications such as linear algebra, numerical simulation, and quantum chemistry
- Nvidia Parallel DataCache - the world’s first true cache hierarchy in a GPU that speeds up algorithms such as physics solvers, raytracing, and sparse matrix multiplication where data addresses are not known beforehand
- Nvidia GigaThread Engine with support for concurrent kernel execution, where different kernels of the same application context can execute on the GPU at the same time (eg: PhysX fluid and rigid body solvers)
- Nexus – the world’s first fully integrated heterogeneous computing application development environment within Microsoft Visual Studio
Oak Ridge National Laboratory (ORNL) has already announced plans for a new supercomputer that will use Fermi to research in areas such as energy and climate change. ORNL’s supercomputer is expected to be 10-times more powerful than today’s fastest supercomputer.
“This would be the first co-processing architecture that Oak Ridge has deployed for open science, and we are extremely excited about the opportunities it creates to solve huge scientific challenges,” Jeff Nichols, ORNL associate lab director for Computing and Computational Sciences said. “With the help of Nvidia technology, Oak Ridge proposes to create a computing platform that will deliver exascale computing within ten years.”
Nvidia did reveal that its upcoming Fermi GPU will pack 3 billion transistors, making it one mammoth chip – bigger than anything from ATI. Of course, the aspirations of Nvidia in the GPU space are far more ambitious than that of AMD. It'll be interesting to see if and how the two head-to-head rivals diverge from the focus on 3D gaming technologies to greater GPGPU application.
- Second Life is King of Social Networks?
- Google Employees Stand Trial in Italy for Video
- YouTube and Warner Music Kiss and Make Up
- Modders Make Star Wars: KOTOR 2 Complete
- Sign Up for Star Wars: The Old Republic Beta
- Google Invites Users to Test Wave
- Steve Ballmer, Other MSFT Execs Take a Pay Cut
- No Beta or Demo for Modern Warfare 2
- Symantec Campaigns Against Free Antivirus
- Microsoft: Windows 7 ''XP Mode'' Finalized
- Turn Empty Drive Bay into Locking Safe
- Microsoft Talks Windows 7 Logo Program
- Lenovo Tweaking Windows 7 Boot Time Speed
- Apple Applies for New Multitouch Patent
- Bing Falls for the First Time Since Launch
- Mechwarrior: The Movie Rumored in the Works
- Apple May Ditch Google Maps
- More Warcraft Movie Details Unleashed
Wow, now that is sweet.
If you're going to put a logo on your chart, common sense states you shouldn't cover up what's on the chart...
What does that say under the TomsHardware logo in the picture..?
The computing capability is great and all but I am personally more interested in affordable GPUs. Let's see if NVIDIA can deliver here.
Hi, I would like a programmable CPU/GPU/GPGPU unit that allowed Virtual Instruments and Effects to be processed on it. Otherwise, this is just more of the same CRAP!
That's nice and all, but when are they gonna start selling the darn thing? Besides, even though an evolution of Cuda is nice and everything, proprietary APIs like that are kind of a hard sell. I think it's cool that it will get some C++ support, we'll see how that one goes, but as OpenCL and DirectCompute are more open it will be more important how this chip compares to AMDs in the performance of those rather than CUDA.
if the performance/price fit the same shoes as ATI's latest release(s), i will be sold and Nvidia will again be an option in my future. Not to mean it isn't now, i'm just saying, ATI has some nice stuff for a low-ish price.
Cool how much will it cost? Most likely have to work for a month just to get one at $10 US an hour.
d
The logo'd out part of the chart reads:
L1 Cache: Configurable 16K or 48K
L2 Cache: 768K
ECC: Yes
Concurrent Kernels: Up to 16
...from another source. Gotta love automated processes like logo stamping
http://media.bestofmicro.com/6/Z/2 [...] erview.png
Not until the card is out and it's not coming out til first quarter of 2010. Besides, with 5870x2 at the corner and 5770, 5850xx... ATI should still hold the best price / performance value.
http://www.hardwarecanucks.com/new [...] s-surface/
The Quadros and the Tesla product lines have always been based off the Geforce line. Is this a turnaround? Are they going to design for the Tesla line and then remove features for the Quadro and Geforce? I hope the pricing of these isn't going to reflect all the capabilities this chip has that will be completely unused by the majority of geforce owners.(other than Folding@home)
wow the money in my pocket is getting really hot
If it does not bottleneck it's self and can manage data flow it could do well. "Just one DVI output - keep in mind this is NOT a gaming card, but the Tesla model for super computing. " I would like to put a card in that takes care of everything else in the background of my computer that bogs the cpu. If it helps with programming languages perform better like C, Java, Python, OpenCL and DirectCompute why not but it have to be more then 10% increase for me to be interested.
I use to see tests with 4 gpus what happen to those days? I still would love to rip threw games and have no game bring me to my knees. I am not sure if motherboards have enough bus to take advantage of this. hmm?
well nice looking gpu but with its delayed entry, focus on computing rather than gaming, and the high possibility that it will be pretty high on the cost scale. i would have to say the odds that they will one up ATI are slim. just due to the fact that their chips are already out, they are priced reasonably and their chips have already shown to haul serious ass in gaming, and with their soon to be full line of dx 11 products covering all price levels of the market i would be surprised if nvidia can pull this one out of their hats.
I think it's safe to assume that if it has the stated capabilities then it really won't have any issues at all playing games, even the next-gen stuff. nVidia would be insane to sell a card in the consumer market without making it a kick-ass gaming beast, or the reviews would tear it apart and within a month all gamers would be buying ATI, and we know they would capitalize on that shift so much that it would force nVidia to go in crisis mode! No need to worry, these card will be premium gaming cards, with the added benefit of an expanded potential. I HOPE!
...will this run crysis?
Ahh... as technology marches on. It is odd that Tom's put their logo on top of not just the chart, it's over the part of the chart showing info on the new GPU!
Woah.... Thats insane. wish I programmed now.
A guite a monster for GPU... Great for folding@home and PhysX I supose.
And don't worry, with that kind of compute power it is ok allso in gaming ;-)
The other thing is completely... is this more sensible than ATI 58XX series for games. With 3.0 billion transistors, this is not going to be a "cheap" alternative to 5870. The Nvidia is going to take the fastest card tittle at any cost. But even GPU genre needs to have it's Ferrari. It may not be sensible to normal use, but it can be fun if you have the money for it.
This seems to be reasonable modular, so there is some hope for new generation Nvidia middle range cards... ATI is doing a great job in that aspect, we need some competition also there. If Nvidia makes (again) facelift for G80 in the low and midlle range, the graphic development can stagnage greatly.
One nice article from Anandtech:
http://www.anandtech.com/video/showdoc.aspx?i=3651
From Anandtech article
So this seems to be scale able, so we really can see competition allso in not so high end segments. But how soon?
Allso interesting is that ATI use smaller chips and use more of them for more power. Nvidia makes one big chip and reduce SM units or something like that to make smaller chip for cheaper prize range.
All in all it's good that Nvidia has a plan and there is going to be some real competition next year!
Maybe cheaper 58xx cards?
Yes that's really nice for scientific purposes and stuff. Now where's your DX11 GPU and new motherboard chipsets?
I want to see if they can come up with something interesting enough to make them my next choice, otherwise I'll definitely be moving over to AMD/ATI.
if NVDA can provide the floating point and processing power required to render in REAL TIME (which is a feat in itself since one realistic image (1 frame, of 60 in 1 second) takes 23 hours on a quad core CPU.) They could take a chunk from intel and amd. I'd be all over support NVDA, but I have been since their first 3D Accelerator (i believ they were diamond at the time, or bought diamond at the time)
if NVDA can provide the floating point and processing power required to render in REAL TIME (which is a feat in itself since one realistic image (1 frame, of 60 in 1 second) takes 23 hours on a quad core CPU.) They could take a chunk from intel and amd. I'd be all over support NVDA, but I have been since their first 3D Accelerator (i believ they were diamond at the time, or bought diamond at the time)
Hell, if this happened nVidia would have a MASSIVE advantage. I'm sure many pro 3D designers will be willing to shell out $5k+ for a card like this considering the time saved in the long run.
Then again, this probably won't happen very soon. Also I'm assuming it will be a true ray traced rendering.
This sounds like the new line of nVidia GPU to compete with the new HD 58xx....
I think this is a logical step by Nvidia to head in the GPGPU direction if it's to survive. Hats off to them if they do release the product.
But rest assure AMD/ATI will come out with a GPU in Q2/Q3 2010 with similar features. You never know, the GT300 might not make out to the market by then :-)
This sounds like the new line of nVidia GPU to compete with the new HD 58xx....
It sounds like a new gpu line that will crush all of their FirePro 3D line
The professional applications and scientific computing market is big, not as big as the video game business, but the current number of professional and scientific applications taking advantage of CUDA after only a couple years is huge, they have an installed base way larger than ATI. If they can keep that base, as they likely will, the entire graphic design, movie production, animation, GIS, oil & gas industries and academia may shift to optimizing their tools for CUDA. I personally think they'd be better of designing for Open CL, so that they can use ATI or nvidia cards for GPU acceleration, it might not be as optimized as if it was specifically tuned for one architecture, but would hedge their bets. Considering we can all by the end of this year have maybe 20 Tflops with 4 x 5870 X2's on our desktops, it's amazing, what would four of these new cards be capable of?