Hydra: the end of SLI/Crossfire?

01Boxer · Post by **01Boxer** » 23 Sep 2009, 15:37

http://www.pcper.com/article.php?aid=785

I'm still skeptical but the motherboards are supposed to be out in 30 days. We'll see then.

DaFox · Post by **DaFox** » 23 Sep 2009, 19:12

Yeah this interested me as well, it sounds alot like going back to the roots of SLI with 3dfx etc.

IIRC SLI just worked back then, and you got a 95% increase in FPS, none of this profile bullshit or anything, and games ranging from a 40% increase to 90% increase because of it.

Post by **Paradigm Shifter** » 23 Sep 2009, 21:10

Yeah this interested me as well, it sounds alot like going back to the roots of SLI with 3dfx etc.

IIRC SLI just worked back then, and you got a 95% increase in FPS, none of this profile bullshit or anything, and games ranging from a negative % increase to 90% increase because of it.

Fixed.

Quite a few games show negative scaling when profiles aren't accurate.

...

And yeah, I loved that about 3dfx SLI, too - nearly perfect scaling going from one card to two. Pity 3dfx went belly up, really. :( Their FSAA routines were ahead of the game, too - ATi has just gone back to using the same methods as 3dfx used to, as it's better quality... but with a hell of a performance drop if the card doesn't have spare shaders to burn.

whismerhill · Post by **whismerhill** » 23 Sep 2009, 23:36

[quote]Yeah this interested me as well, it sounds alot like going back to the roots of SLI with 3dfx etc.

IIRC SLI just worked back then, and you got a 95% increase in FPS, none of this profile bullshit or anything, and games ranging from a negative % increase to 90% increase because of it.

Fixed.

Quite a few games show negative scaling when profiles aren't accurate.

...

And yeah, I loved that about 3dfx SLI, too - nearly perfect scaling going from one card to two. Pity 3dfx went belly up, really. :( Their FSAA routines were ahead of the game, too - ATi has just gone back to using the same methods as 3dfx used to, as it's better quality... but with a hell of a performance drop if the card doesn't have spare shaders to burn.
still at first I was impressed but later in several games the difference in the fact that each line was calculated by a different GPU was pretty obvious with an easy to see kind of blur caused by lines not being synchronous !!

Frag Maniac · Post by **Frag Maniac** » 23 Sep 2009, 23:52

This looks to be both promising and cause for skepticism. While I do acknowledge that much of what they're saying about SLI or Crossfire methods being inefficient is true, especially since AF is the most common method and some have speculated it's latency may in fact be the cause of microstutter in dual GPU cards, I don't know if I like that another (3rd party mind you) chip/driver/engine has to be used to do it. Isn't this going to rile the GPU manufacturers, esp Nvidia, whom have always been hard to work with concerning SLI? What's to keep them from borking their GPU drivers so Hydra tech doesn't work reliably with them, esp if multi-vendor GPU setups are possible?

I also don't get why this sentence...

"Lucid is offering up scaling between GPUs of any KIND within a brand (only ATI with ATI, NVIDIA with NVIDIA)"

...is followed by this...

"This is the most important system running both an AMD Radeon HD 4890 and GeForce GTX 260 and scaling!"

...does not one contradict the other?

(EDITED)
Never mind, found a better article that explains it.

http://techreport.com/discussions.x/17636

"Initially, Lucid's driver will only support multiple GPUs from a single vendor. However, the firm plans to introduce multi-vendor capability in a driver update early in the first quarter of 2010. (A beta version of that driver will roll out before then.) At that point, folks running Windows 7 will be able to pair up Radeons and GeForces to their heart's content."

Mesh · Post by **Mesh** » 24 Sep 2009, 07:12

And shortly after release these chips will join aegia physx in obscur designs that never match the hardware potential due to software pitfalls.

Post by **Paradigm Shifter** » 24 Sep 2009, 11:23

Then nVidia will buy them. ;)

Mesh · Post by **Mesh** » 24 Sep 2009, 19:44

And looked how that turned out, instead of making a standard both sides can use, they tried to use it as yet another proprietary standard that's only good for tech. demos and utterly useless in actual games.

Post by **Downtown1** » 25 Sep 2009, 03:13

Interesting, guess I'm waiting a bit longer before upgrading my new PC to take advantage of EyeFinity.

I am wondering if it'll actually be better than either CF or SLI in the super high (e.g. 3x1680x1050 and higher) scenarios. Otherwise there's not much point in buying that is there? ATI/Nvidia cards typically have performance equivalents, and ATI already lets you mix and match cards for CF (not sure about Nvidia).

Frag Maniac · Post by **Frag Maniac** » 25 Sep 2009, 04:47

On the one hand Lucid makes good points in that SLI can be a problem in some games, with FPS plummeting in spots even on GTX295, not to mention some games just not working at all in SLI. On the other, it will no doubt be very difficult to keep the drivers functioning well with a myriad of ATI and Nvidia GPUs.

Personally I think dynamic multi GPU scaling has more potential than the existing fixed profile scaling of AF. The load balancing and efficiency would no doubt be better. It's just a shame that Nvidia, ATI or Intel haven't implemented it in their MB chipsets so a 3rd party chip/engine/driver isn't needed to run it.

I think it's safe to say though that if a big name MB manufacturer like MSI is willing to incorporate it into their flagship gaming MB, it must have been tested fairly thoroughly for game compatibility and multi GPU rendering effectiveness.

Through two articles I've read on this and Lucid's demo video, I don't recall once them saying games have to be written for Hydra like SLI and Crossfire do. Instead they refer to Hydra as "intercepting" the data sent to the DirectX API and then assigning tasks to the GPUs.

If the games don't have to be written for it, that eliminates a huge stumbling block that is no doubt the primary cause of some games not handling SLI properly. All Hydra does is act as an intermediary between DX and the GPU drivers. It has nothing to do with the actual GPU, game, or OS.

Conversely, Aegia's PhysX that one likened this to is redundant by comparison because what the CPU did before to handle physics, DX11 will on the GPU. This tech goes a direction no one else has before and I think will be a boon to simple upgrades for many, which could make it a big seller out of the gate.

Post by **Paradigm Shifter** » 25 Sep 2009, 11:04

Let's hope this 'interception' by the Hydra chip doesn't impact display lag or response. Wouldn't be much use for gaming if it introduced more than a couple of ms lag... maximum.

Frag Maniac · Post by **Frag Maniac** » 25 Sep 2009, 16:07

Let's hope this 'interception' by the Hydra chip doesn't impact display lag or response. Wouldn't be much use for gaming if it introduced more than a couple of ms lag... maximum.

Keep in mind though it doesn't have to intercept, analyze and assign on top of SLI/Crossfire, it's doing it in place of it. I've seen some numbers on AF latency in SLI and it isn't as fast as you might think.

whismerhill · Post by **whismerhill** » 26 Sep 2009, 21:48

[quote]Let's hope this 'interception' by the Hydra chip doesn't impact display lag or response. Wouldn't be much use for gaming if it introduced more than a couple of ms lag... maximum.

Keep in mind though it doesn't have to intercept, analyze and assign on top of SLI/Crossfire, it's doing it in place of it. I've seen some numbers on AF latency in SLI and it isn't as fast as you might think.
I agree on this one
and I'll add that if that solution is mostly hardware based (that is with a majority of the workload on hardware)
in this case it could :
-free more some processor usage in multiGPU scenario
-get less latency because of it

however that's possibilities, it all comes down to how robust the solution is ...

Frag Maniac · Post by **Frag Maniac** » 27 Sep 2009, 00:32

I think the main thing is keeping up on GPU driver compatibility. One may think that means only ATI and Nvidia drivers when released, but with ATI coming out with drivers on a monthly basis and Nvidia slinging out all these beta drivers, it's a lot to keep up on.

CPU load wise, DirectX 11 in itself will take quite a load off the CPU because with it multi threading, physics, etc on the GPU will be possible. So the timing couldn't be better for this kind of product because all things considered, CPUs will get a bit of a rest and more will be done on the GPU itself. In fact I'm thinking DX11 may require less of a need to OC CPUs even with multi GPU setups.

01Boxer · Post by **01Boxer** » 01 Oct 2009, 05:48

http://www.legitreviews.com/article/1093/1/

Post by **Paradigm Shifter** » 01 Oct 2009, 11:23

I suspect that nVidia will update their drivers soon which breaks something for this. I can't see them being happy at vendor-agnostic multicard working. Of course, I can't see AMD being particularly chuffed either, but they've not got the record of driver disabling that nVidia currently sports.

Frag Maniac · Post by **Frag Maniac** » 01 Oct 2009, 22:38

Someone on one forum said ATI helped Lucid develop this tech, but the above article through it's Q&A with Nvidia seems to indicate it was in fact Intel, which makes more sense. The last question posed to Nvidia was even more curious, asking if they might be interested in buying out Lucid like they did Aegia. One has to wonder at this point if it would be a better business venture than the Aegia acquisition was. Of course if they did they'd no doubt halt the possibility of cross vendor GPU mixing, even though the vast majority of users stick within one GPU camp.

The Q&A with Lucid was a bit troubling. A speculative price of $1.50 per Pci-Ex lane for Hydra equipped MBs does not sound good at all. Assuming the MSI Big Bang is designed for use of at least 32 simultaneous Pci-Ex lanes like the X58s are, that means an MSRP of at least $480. If that's the case, it will be very hard for Lucid to get other MB manufacturers to jump on the Hydra bandwagon, and that could very well mean a startup company that has trouble kicking over the engine like Aegia experienced.

However, it would appear from the games demoed (Project Origin and Bioshock), which have been out for some time now, the Hydra tech may not have to have games written for it like SLI and Crossfire. This in itself would mean more consistency out of Hydra vs the competition I would think. I also in fairness to Nvidia find what they're saying about sticking with same GPU model matches in SLI, makes sense since AF is still the primary means of multi GPU scaling.

ATI allowing non matching GPU models may be why Crossfire doesn't scale quite as well as SLI for those whom use non matched GPUs. In AF scaling each GPU is handling the same content onscreen and just alternating frames back and forth. To allow non matching GPU models in such a scaling profile would mean to some degree being limited by the slower card.

01Boxer · Post by **01Boxer** » 12 Nov 2009, 03:40

The latest round of previews seem to indicate that the technology seems to be working. The downside is that dual GPUs on a single card like the GTX295 are detected as a single GPU and only half of the card's power is used. I find the delay to Q1 2010 annoying because i was planning to build my new computer soon.

whismerhill · Post by **whismerhill** » 12 Nov 2009, 07:57

it would be interesting to know if this would work along with EyeFinity

01Boxer · Post by **01Boxer** » 07 Jan 2010, 19:25

A review is out:
http://www.guru3d.com/article/msi-big-bang-fuzion-lucid-hydra-review-test/

389$ is a little too rich for my blood.

Widescreen Gaming Forum