Lucidchip
Yes, there's already early silicon which apparently delivers on Lucid's amazing claims.

What is Hydra? A SLI/CrossFire slayer, that's what!

Bennett Ring
21 August 2008, 7:12 PM


Multi-GPU technology debuts at IDF; could put SLI and CrossFire to shame.


While not all of us were lucky enough to get an invite to this week’s IDF (damn you David and Seamus!), thanks to the wonders of the blogosphere we’re practically there anyway. And we have to say that for all Intel’s talk of Nehalem, the once piece of news that has made us take note is a little piece of hardware called Hydra, designed by a company known as Lucid.

If you can’t join ‘em, beat ‘em, and that is exactly what Lucid has done with Hydra. Let us explain. One of Lucid’s main investors is Intel, the very same company who can’t get NVIDIA to deliver SLI on Intel chipsets.  Hydra is a hardware solution developed by Lucid that enables multi-GPU solutions that are independent of the GPUs in use – and it has nothing to do with SLI or CrossFire. In other words, it will allow you to whack any combination of video cards into your system to deliver multi-GPU performance. It could also deliver multi-NVIDIA GPU performance to Intel systems, without Intel needing to sign an agreement with NVIDIA. But it gets even better.

Lucid claims that Hydra can deliver near linear performance increases in a multi-GPU setup; run four GPUs and you’ll get four times the performance. Contrast this with today’s primitive solutions delivered by ATI and NVIDIA, where two GPUs deliver anywhere between 1.3 and 1.8 times the performance, (and results getting even worse once we hit three GPUs), and now you might understand why we’re just a tad excited about this technology.

Hydra delivers such impressive performance by intercepting DirectX calls before they hit the GPUs, then deciding on where the load should be sent. It’s a mixed software/hardware approach, that is much more refined than today’s alternate/split frame approach. Hydra could decide to send the rendering of one model or effect to just one card, and let the other cards handle the rest of the scene. It’s a very dynamic approach, with Hydra knowing exactly what each GPU on the market is capable of, and maximising the use of each GPU’s strengths.

For this tech to be shown at IDF, Intel must have confidence that it’s not just great on paper. It has the potential to make multi-GPU systems truly worthwhile, rather than the buggy, overpriced and under-delivering issue that they are today. With some form of release in early 2009, Hydra has suddenly bounced to the top of APC’s technologies-to-watch list.


Post your comment



Comments

RSS feed Email alert
21 August 2008, 7:57 PM (4 years ago)report abuse Send to a friend reply

Me In Oz (User):

I have no doubts that Intel has the resources and clout to make these early claims, but the thing that will need more 'fine tuning' will be the "mixed hardware/software approach". Games and apps (if any) will have to be developed for this tech and I will bet that the games will be exclusive to this design.

The thought of using ANY Graphics GPU combination is outstanding though :)

And I totally agree with the comment that SLI and Crossfire is not an economical way to increase gaming performance. Although we love selling the stuff, I discretely shake my head as the customer hands over $1000 for 2 cards to increase his frame rate from 50 fps to 60 fps !

22 August 2008, 8:37 AM (4 years ago)report abuse Send to a friend reply

Your Average Joe (User):

Looks like another nail in the coffin for OpenGL.
DirectX only games seem not far away now !

22 August 2008, 10:30 AM (4 years ago)report abuse Send to a friend reply

agami (New user):

I wonder how this fits in with Intel's Larrabee architecture?

On the surface the Hydra IC seems to be a DX broker chip. This to me seems like a stop-gap solution as most chip makers are looking to further blur the distinction between CPU and GPU. The future for Lucid could be a a version of Hydra that is an Integer/FPU broker. Then it wouldn't matter if it were DX or OpenGL, Multimedia or Scientific applications.

22 August 2008, 12:48 PM (4 years ago)report abuse Send to a friend reply

anonymous user Anonymous user