Multi GPU tech Lucid to take on graphics giants

This story is getting damn boring. We have been hearing about this for ages and seeing bits of kit.

When are they going to make something people can actually use I'm getting sick of waiting. the green and red team will probably have something else out by the time Lucid get their thumb out their bumhole and Intel will probably have the late to market Larabee or wanabee.

This is getting dull. No point in saying this stuff is gonna be miles better than current stuff if its gonna be 2 years after the fact.

YAWN BLOODY YAWN!! HURRY THE FECK UP

Inaccurate to make such a claim. They may not have 'frame' lags, but they would have 'object lags'

Also not all forms of multi GPU rendering are frame independant. Other forms of multi GPU also render different parts of the same scene.

SLI's split frame mode, which doesn't suffer from 'frame lag' but 'tearing' can be observered as one gpu may complete rendering it's part of the frame and moves on to the next, while the other GPU lags behind. With vsync disabled, this lag results in a scene 'tear'

Power VR's TILE based mode suffered the same as SLI, except there were multiple 'tears' in the scene if vsynch was disabled.

I've never heard of object based rendering prior to lucid, however I assume there also would be considerable overhead in setting up which textures went to which card. Solutions would be to share ram, duplicate ram. Sharing a pool of ram creates contention as GPU's and the CPU must access the ram through the same controller (the bottle neck) and duplicating the ram (like SLI does) requires allocating ram on local GPU's for objects being rendered on other GPU's (more expensive, but most believe the better of the two evils)
Coordianting which GPU is to render which object is another area of overhead. I suppose that is what that custom controller is for.

At any rate, I'm looking forward to seeing what these guys can accomplish. IMO, faster, better graphics is faster and better :).

The article isn't completely accurate. Nvidia renders in one of two modes not just AFR as the article implies. It also can use SFR as well.

Oh my god, the post above me looks like the author ran his comment through four or five language translators in google and then converted it back to english. I do have to agree with a couple things you said though man, especially this quote...

"Breaking from 3 &1/2 years of ignoring specs has brought out more garbage upon perfectly good O/S, than heretofore ever."

Nice dude!

Is this Bitboys Oy reincarnated?

Firstly didn't understand this bit "According to JPR, Hydra's scaling isn't quite as good as Crossfire's or SLI's for VGAs yet"
Thought the whole point was that it scaled almost 100% and therefore better not worse than nvidia or ati?

Secondly where is the news in this "news item" other than "jpr" thinks its a good idea....we all came to that conclusion ages ago....all we have seen so far is that gpgpu box they released other than that it sadley does seem to be getting old.

"doesn't scale as well as SLI or cfx"
They said last year they can provide near linear scaling??

I bet Intel will get them to launch it alongside Larrabee.

"Inaccurate to make such a claim. They may not have 'frame' lags, but they would have 'object lags'"

Object lag??? Are you joking? Its not exactly hard to insert a sync ...

And Sylvie's claim is wrong too. The reason the driver queues 3 or so frames is to smooth out sudden changes in whats drawn. There is no difficulty in queueing 3 frames of data and sending each one to a different card. In fact if you do this you will return from the rendering of all 3 frames that bit quicker and hence lag would actually be REDUCED.

But what do I know? Have only been professionally writing 3D engines for 10 years or so ....

Of course you can't get linear scaling. The only way I can think of that you could distribute per "object" as they suggest would be to interrupt each "DrawPrimitive" call (In D3D parlance). Splitting each individual call up would be a pretty harsh task (Though not impossible). Dynamically branching shaders will also create havoc here because you can't know how quick a shader will run for a given pixel let alone a given triangle. If one draw call takes 90% of the time then it doesn't matter how many other cards you distribute the other 10% too ...

In the photo shown, that looks like the Lucid chip has it's own "Motherboard" and the video cards are daughtered into it.

That would be a great idea. You could run your main system off a smaller PSU, and generate a lot less heat in it the main case, which would be great for the CPU and RAMs sake.

Also, having the Video Cards in a seperate box, as shown, you could set up better air flow for them, but having intakes (and possibly fans) directly in front of the cards. I want one.

The only question is though, how does it hook into the main box, and does that connection carry enough bandwidth that you are not bottlenecking the PCI-E feed?

It may have been a while since the Lucid HYDRA technology was announced, but it alive and was on display this week in Tokyo at the Industrial Virtual Reality Expo.

SGI demonstrated the ELSA Vridge system running two NVIDIA Quadros and two Quadro FX 5800 and two ELSA Vixel H200.

Read the news release here: http://lucidlogix.com/news/pr_2009_06_23.html

This article is a bit wierd.
I will try explain how the chip works.
The Lucid chip scales using API calls, if a object is called its sent to a card processed etc. Im no expert but it makes sence to me to split up the API calls and distribute workload.
Near 100% linear scaling is what they have claimed (maybe its lower "now" because their not finished yet? or something), API's supported was DX/D3D and OpenGL.
I have no idea if the chip is upgradable to be able to support DX10, DX11 API calls etc, but i do think so.
distributing API calls to different graphic cards basicly means to me send this texture to that card, create these polygons on that card etc.
they send it back and the chip (basicly intercepts?) then sends the stuff for output on the main card (im sure i make no sence but whatever).

There was some kind of load balancing that would ensure the near perfect scaling even if mixing very old cards with new they would work near 100%.
the chip knows what to send where and keep the cards working maximum, if that is automatic by the chip, or software no idea..
What i do know is that they have said you can use a DX8 card (perhaps even older) and benefit in DX9 games as DX9 api calls go to DX9 card while other calls that perhaps the DX8 card can handle will be sent to it.
If its upgradable to future API's or if you have to get a new chip for future API's i dont know but i would gues yes since they said they will focus on gaming API's implying other API's like GPGPU etc can be used in the same way in the future. Vaguely remember something said about it taking time, resources and they need to concentrate one step at the time.
And no wonder intel is an investor? This would be megakill for their Larrabee (maybe CPU scaling/multicore can be done the same way? to hard and to mutch work atm i think but the potential), and RAYTRACING would finaly be a reality. Aswell as other API's that can use the graphic cards.

And what does all this potentially mean? Say you get a DX12 (lol) card, a DX12 game and have a bunch of junk cards using DX10, DX9, DX8 etc, you can use them playing the DX12 game, assuming the API doesnt RADICALLY change.
What does this mean?
Keep your old cards, wait for a motherboard consisting only of lucid chip and rest PCIe slots and use that beast to fill with cards and play your favorite game at ridiculous framerates/resolutions.
A true graphic miracle possibly dooming our race since we will never leave our computers, now bigger then ever possibly several cases filled with obsolete DX9 junk cards creating a insane beast..

I actualy realy believe Lucids Hydra chip is the real deal we have been waiting for in terms of graphics processing. I think everything sounds ok in terms how it works for a novice like me, im sorry i dont remember more.
Did i actualy make sence to anyone? lol

"i bet my overclocked 486 with an MX440 would kick its ass on 3dmark07"

LOL!!!

I'm just imagining the PSU requirements of the future then..

Its bad enough trying to power ONE card let along 3+

Yes... I believe last year we were told "near 100% scaling" in realtime, put ATI & NVidia to shame etc etc etc... now apparently scaling is worse than CrossSLI. I wonder why less scaling than the big boys across more GPU's (which I'm sure they could do if they wanted, at equally diminishing returns) is an advantage?

More vapourware rubbish

I see Intel is backing them, this sounds like the perfect technology for their up and coming Larrabee multi core CPU/GPU hybrid chip with up to 32 cores or more. Only a thought.

What is stupid "green goblin" name? Author are you ATI fanboy? I think Charli was the one here so look like i was wrong and all inquirer writers are just ATI chear fan boys... or you just a goblins lover? :P

I think lucid hydra is bad idia cose it will be create more lags to sinc work of GPU, but maybe it's better then 4-GPU crossfire/SLI lag. I think now more interesting ATI's DX11 chips. Area of showed DX11 chip is about 180 square mm, it's too small for 256bit GDDR5 interface but enough for 192, but these can be 128bit GDDR5 interface and hight clocked 64bit for fast inteconnect of cores on substrate. These shame allow non AFR and none lag rendering with shared pull of ram between two chips and create opportunity cover all markets from low to hight end withs only one cheap chip, what look more than cute now

Lucid tech CEO archer-quinn
http://www.engadget.com/2008/05/05/archer-quinn-documenting-his-free-energy-project-descent-into-m/

google: ATI tile based rendering

When Lucid will come up with real dame demos they'll say "you got me ;)" ATI tile based rendering is pretty close to our performance but again we never said about it. we were saying their AFR is far slow. wink wink!

Lucid's CEO is Moshe Steiner... a joo, one might presume. Archer Quinn? What are you smoking?

...will it play Crysis?

Man, it was a pun on Lucid, cuz Lucid is also making castles in air. Like archer-quinn fooled engadget and was turned out to be fake. I guess Lucid will also turn out to be similar as archer-quinn.

....should be looked at again.

Running PowerVR with half the rendering HP we have available would be pretty incredible.

Could make some very fast low power cards.