Lucid Hydra Benchmarks Finally!

[gigahertz20]

http://www.pcper.com/article.php?aid=815&type=expert&pid=1
http://hothardware.com/Articles/Lucid-Hydra-200-MultiGPU-Performance-Revealed/?page=1


Articles were just released today, the performance was what alot of us predicted. It works, but it's no where near close to the linear performance scaling that was mentioned in alot of articles on the net. Performance will improve even more over time, as Lucid continues to work on their drivers, so it'll be interesting to see how far they can push the performance scaling. It might give SLI and Crossfire a run for their money eventually.

 

[MarcVenice]

Thank you for this

Looks like it works, although not as good as people were hoping it to be. Nonetheless, if you had a GTS 250, and buy a GTX 285, the GTS 250 can actually be quite a bit of help. If that remains true across multiple generations, like you had a 4870 and it helps a fair bit if paired with a 5870, then it's quite nice imo. Instead of selling the 4870 at an absolute loss, you can actually usefully use it, and it'll heat your room too

 

[Soccerman06]

Obviously you don't remember the performance sli/xfire provided when they were released. They were lucky to get this kind of performance boost. When drivers mature we will see the 90% performance (hopefully).

 

[Quiksilver]

As long as it is able to scale better than Crossfire or SLI, I think Lucid will still have a winner, even if it isn't true 100% scaling.

 

[JSt0rm]

As long as it is able to scale better than Crossfire or SLI, I think Lucid will still have a winner, even if it isn't true 100% scaling.

I agree 100% is a little crazy for scaling, but these results are respectable for sure.

 

[thilanliyan]

Even less than 90% scaling would be alright with me...it means you don't always have to get 2 identical cards and can do it over several generations. For example I could use the 8800GT I have sitting in a drawer doing nothing with my 4870 and see performance gains.

As some others have pointed out, I'm curious to see how the IQ settings work...does it default to whichever card is primary?

 

[OVerLoRDI]

That test rig is completely ridiculous, but it makes me happy as a hardware nerd. Will be great to see the final product, and I would also like to see it on an AMD platform, I have some mix and matched cards I'd like to see paired up.

 

[RussianSensation]

Add Techreport to the mix:
http://www.techreport.com/articles.x/17934/3

In every single benchmark tested GTX 260 + 4890 was faster than dual GTX 260s. More shocking was Call of Juarez and Stormrise performances where GTX 260 + 4890 netted almost another 100% performance increase over GTX 260.

Bang for the buck GTX 260 + 4770 cleaned up the house

wow!

 

[veri745]

I had trouble believing the "linear scaling" rumblings when I originally heard them, and it looks like I was right, but I'm still extremely impressed.

 

[Shmee]

not bad. Any news of an x58 big bang? or other x58 mobo's with the lucid hydra chip? Seems kinda silly to have all this potential of graphics power and not have gulftown potential

 

[nitromullet]

This is definitely impressive. The thing with mixed vendor cards is just baffling IMO. I'd really love to see that in action.

 

[shangshang]

This is pretty impressive.

But the real question is will NV or AMD lock this out?

I wonder if an NV card is used in Hydra, can it still be used for Physx?

But i'm not sold on Hydra yet though. I'll wait to hear the complaints first from early adopters before I'll even consider Hydra

 

[dflynchimp]

This is pretty impressive.

But the real question is will NV or AMD lock this out?

I wonder if an NV card is used in Hydra, can it still be used for Physx?

But i'm not sold on Hydra yet though. I'll wait to hear the complaints first from early adopters before I'll even consider Hydra

Yeah that's what I'm wondering too. It seems entirely possible that if Nvidia was able to vendor lock ATI cards out of the PhysX equation detecting the motherboard chipset then choosing to support or lock it out is a distinct possibility

 

[Hyperlite]

wow. very, very promising. I can see, in the future, being able to basically dynamically assign either card for GPGPU, or straight GPU, or physx, etc....very cool.

 

[Cookie Monster]

If this technology matures abit it looks like it will attract quite alot of customers.

What i really like is how you can drive optional displays with your second, third (etc) card. For people who use multi-monitors, this is a good thing since it would not cripple the primary card's power saving features.

So basically, do you need to install two drivers?

And, wonder how the overall IQ will be like. Like if you use different AA between the cards (say 16xCSAA on the nVIDIA card and 8xAA on the ATi card) , how would this work?

 

[EarthwormJim]

If this technology matures abit it looks like it will attract quite alot of customers.

What i really like is how you can drive optional displays with your second, third (etc) card. For people who use multi-monitors, this is a good thing since it would not cripple the primary card's power saving features.

So basically, do you need to install two drivers?

And, wonder how the overall IQ will be like. Like if you use different AA between the cards (say 16xCSAA on the nVIDIA card and 8xAA on the ATi card) , how would this work?

Wondering about that too, since it seems to split up parts of the scene, maybe the parts on the Nvidia card would do 16xCSAA and the parts on the ATi card would do 8xAA.

 

[n7]

Yay, finally real numbers.

 

[Drakcol]

Indeed and Lucid Hydra actually looks quite appealing esp since I have an old 9800gtx+ lying around after I upgraded to an ATI 5850.

 

[jvroig]

What's really interesting is how they managed to get the scaling that they did. Before the benchmarks were published, a lot of people kept moaning about how Hydra is impossible because "AMD/ATi's own engineers couldn't even scale crossfire as ambitiously, nor could nVidia's own engineers scale SLI as ambitiously, how the hell can Lucid engineers achieve such scaling on vid cards, especially for a mix of ATi and nVidia". So really, as a tech guy, what's interesting for me is how they were able to do it. I'm sure they can't spill the beans since it's a trade secret, but I hope we do get atleast a small clue.

Lucid gets a big bump in the cred department for this in my book

 

[firewolfsm]

The best scaling obviously comes from mixing vendors, it seems that if either card is better at a particular game it takes over and is supplemented by the other so that the performance more than doubles that of the weaker card. It's like covering the weaknesses of each architecture.

 

[TurtleCrusher]

I would love to mix and match my 4830, 3850 and my 8800GTS 320mb.

 

[jvroig]

Right. Making two of cards of the same vendor scale isn't impressive; it's the ability to scale while using a mix of nVidia and ATi cards. I wonder how complicated the Hydra driver is - could it just be a library of API calls for both nV and ATi? Is that even possible?

 

[thilanliyan]

This is pretty impressive.

But the real question is will NV or AMD lock this out?

I doubt ATI will lock it out. There are no XFire license fees like there are SLI fees so they don't really have a whole lot to lose. Unless people start to think that the IDEAL configuration is a mixed setup...then they lose potential sales.

 

[jvroig]

Unless people start to think that the IDEAL configuration is a mixed setup...then they lose potential sales.

Not as much as some people might imagine. In the high-end, very few sales are made compared to low- and mid-range offerings. Selling one card as opposed to "what-could-have-been-two" isn't so bad, especially since people would factor out getting two AMD cards in CF (let's say 5870) versus an AMD card (5870) and an NV card (GTX 285 or the upcoming Fermi) and the $40-70 price premium for the hydra chip.

Significant lost sales for AMD would probably matter if Hydra "steals" from their mainstream, low to mid-range cards. And that's just not gonna happen - or at least, doesn't seem too likely, unless the price premium drops and MSI (or whoever) makes a board marketed at using two midrange cards that scale better than CF/SLI and thus offers high-end performance for cheap. Like I said, not looking very likely right now, so AMD probably only has half an eye looking at Lucid Hydra right now.

 

[BFG10K]

Look at the bottom three graphs here:

http://techreport.com/articles.x/17934/3

Here a multi-GPU 4980 is slower than a single 4890.

Lucid implied this shouldn’t happen because application specific optimizations in the drivers are not required. Clearly that isn’t the case.