Is GPU Stacking The Answer To All Your Performance Concerns?
Kfar Netter (Israel) - Sure, you may have heard about SLI and Crossfire and you may run up to four graphics chips in your PC. But what if you could add many more GPUs to your system, what performance in GPU and GPGPU applications would your PC be capable of? It seems that we are about to find out.
There is a big debate how parallel computing may evolve over the next few years and unlock what appears to be an opportunity of dramatic computing horsepower. On the one side, there is Intel, which believes that the future are many-core x86 cGPU devices, scheduled to first debut with the Larrabee accelerator in the second half of 2009. On the other side, there is Nvidia which bets that GPUs could take over much more functionality in future PCs and accelerate applications through general purpose GPU computing. AMD is somewhere in the middle, as it can go in both directions.
We have little doubt at this point that GPU-based stream computing will have its place in future PCs. One more indication that we may not yet fully understand what GPUs may be capable of is a new technology called Hydra. Created by LucidLogix, Hydra delivers a hardware-software interface that can run a number of graphics cards - not just two or four - within one system. The benefit, of course may not be increased graphics performance, as most current games have trouble to scale even beyond two graphics chips. The benefit could be to scale the GPGPU performance deep into supercomputer territory. If you have the necessary power supply, the 5 or 10 TFlops could be within reach.
According to LucidLogix, Hydra combines a system-on-chip with proprietary software technologies that load-balance graphics processing tasks. The company promises "near-linear to above-linear performance with two, three or more graphics cards".
"We’re currently focusing our Hydra engine on improving everyday PC performance, but our vision is to bring more scalability, flexibility and power to all visual processing markets including graphics professionals, power gamers, design/engineering professionals and others," said Offir Remez, LucidLogix co-founder and president. "
LucidLogix said it has begun working with "OEMs, ODMs and PC technology providers" to create platform solutions that are expected to come to market in the first half of 2009. The company said that the Hydra engine does not require a modification to existing graphics software applications and supports common graphics standards such as DirectX.
What makes the technology especially interesting is the fact that it apparently works with any GPU. Compared to Nvidia’s SLI and ATI’s Crossfire, both of which use AFR (Alternate Frame Rendering) and SFR (Split Frame Rendering) techniques, Lucid claims that it uses multiple graphical techniques that are used during the process of each frame in order to avoid performance latencies. Hydra is also promised to reduce the graphical processing resources required from the main processor using internal memory and a RISC processor.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Despite GPU scaling issues in current games, LucidLogix claims that it has been able to speed up games such as Crysis, Unreal and Bioshock by 85-95%.
No further technical details were provided.
Wolfgang Gruener is an experienced professional in digital strategy and content, specializing in web strategy, content architecture, user experience, and applying AI in content operations within the insurtech industry. His previous roles include Director, Digital Strategy and Content Experience at American Eagle, Managing Editor at TG Daily, and contributing to publications like Tom's Guide and Tom's Hardware.
-
bloodymaze Oh my gee -- 85% - 95% increase in performance? I love it. Crysis = PWNED! lol -- GOOD NEWS FOR THE CONSUMER I LOVE IT!Reply -
seatrotter Let's just hope that standards come up soon to allow just one interface to handle any kind of GPGPU hardware. This'll pave a way for developers to start accommodating such features (since they wouldn't be concerned about vendor-specific interface). Also, it will encourage and help accelerate development of GPGPU hardwares.Reply -
homerdog ^I don't get what you're trying to say, but for what it's worth it was Intel's Israeli design team that built the Core architecture.Reply -
Mr_Man I wonder what this will mean for manufacturers making high end and low end cards. If you could have 5 8800GT's scaling linearly, that would beat even 2 GTX 280's, and cost less, too. Of course, you'd have to have a large case and 5 PCI-Express ports.Reply -
homerdog Reply
What's more interesting is their claim of application-independent scaling. Two 8800GTs would be nearly unstoppable (even in Crysis) if they scaled linearly across the board.9083218 said:I wonder what this will mean for manufacturers making high end and low end cards. If you could have 5 8800GT's scaling linearly, that would beat even 2 GTX 280's, and cost less, too. Of course, you'd have to have a large case and 5 PCI-Express ports. -
Luscious It's been a long time since I've heard the word, but if RISC hardware is the secret to multi-GPU performance then the ODM's will be looking at radical board designs. Makes perfect sense though when you think about it - when scaling is done in hardware, it would allow performance level increases across the spectrum.Reply
I imagine 4 4870x2 cards humming away with 8 GPU's and 8GB of ddr5 memory would only be a start of things to come if this technology becomes mainstream.
Now, if we could use that power for something more useful than Crysis... that would be the real computing achievement.