Home
Friday, August 18, 2017
11:03:43 AM
Users online: 0   You are here >> Home > Graphic Hardware

Forums | Graphic Hardware Forums search
Forum FAQ
   
  1 | 2 Next Page 
mixing gpus now possible?
layzi 
23/8/08 10:28:06 PM
Learner
well after doing some tech reading as i normally do i stumbled across this fascinating new chip which supposedly enables you to mix different brands of gpus. so what do you think?

http://www.lucidlogix.com/technology/technologies.html

-----
future gaming=realism=physics processing=nvidia.

Jeruselem 
23/8/08 10:42:07 PM
Guru

Interesting! :)
Hope it works.

-----
PC 1: XP Home SP2, Opty 165@1.8Ghz, GEIL 1GB PC3200, 320GB SATA Cuda ES,XFX 9600GSO 580/700x2/1450, Seasonic S12+ 550W
PC 2: XP Home SP3, XP 3000+@2.24 Ghz, 1GB PC3200, 80GB IDE,ASUS nVidia 6800 512MB, Antec PlanetWatts 380W

layzi 
23/8/08 10:52:41 PM
Learner
hmm sorry i may have jumped the gun abit, i realised that this chip only enables motherboards to use different brands of gpus rather than scaling different brands of gpus together. =/

http://www.pcper.com/article.php?aid=607


Edited by layzi: 23/8/2008 11:28:45 PM

-----
future gaming=realism=physics processing=nvidia.

superfireydave 
23/8/08 11:35:52 PM
Titan

To me, what it looks like is this:

You have 2 GPU's hooked up (vendor and model is irrelevant) and the Hydra card.

You run a 3D game in fullscreen mode on the one monitor.

The command to draw the screen is sent to the hydra card which then allocates tasks and then delivers seperate tasks to each GPU.

From what I understand, each card will render a seperate part of the scene completely, therefore textures do not need to be shared, and both cards are working on the one frame at a time.

Once finished, they send it back to the Hydra, and then the Hydra sends it back to the GPU which is hooked up to your screen to output.

So from what I understand it's fucking brilliant. I'm really interested to see how it turns out.

-----
Mreow?
http://www.atomicmpc.com.au/forums.asp?s=2&c=9&t=17306

TheFrunj 
23/8/08 11:37:55 PM
God

That's essentially it dave.

Anandtech have a good summary/explanation of it here:
http://www.anandtech.com/video/showdoc.aspx?i=3385

Very excited on my part too, will hopefully see a great increase in every game from an upgrade.

-----
Atomic – Tech Writer

Quote by Shikimaru
TheFrunj doesn't guess, when he forms words with his lips, the world will Warp and change to do his bidding.

SceptreCore 
24/8/08 12:11:51 AM
Guru

Awesome.... eliminating the limitations of the SLI or CrossFire chipsets, and letting a smarter, more powerful third party card do the work (I wonder if even this thing is overclockable). However this thing needs to be between the two cards yes... and it also needs a PCI slot, unfortunately I only have the PCIe 1x between mine.

-----
Greatest Sayings Of Our Time:
Quote by sora3
...What drivers are you using? If you say the drivers on your CD, I will phone you and tell you that I will bone your f*cking dog...



m0zes 
24/8/08 8:18:46 AM
Champion
I've been doing some reading about this over the last few weeks and whilst it does sound good, i haven't seen any particularly useful information out yet, for instance:
-No discussion on how it will handle architectural/software differences in cards, for example if i've got an nvidia based based card and software that i'm running is coded specifically to take advantage of a particular way my card can render shadows and the other card in the array doesn't support this what happens?
-How about vista? No use having an array of nvidia and ati cards as you can only use a single driver.
-How about dx revision, keep your old card but what happens if your combining different dx revisions?

So is this going to end up being useful, i think so, but at this point in time i see it being more so for getting the most out of identical cards rather than the ability to mix and match different cards. But word has it the company producing this is heavily funded by intel, so you've also got to wonder whether this is more about larrabee than anything. For instance if the way larrabee is designed it can be programmed to do or imitate any card then combining an ati + intel, nvidia + intel solution would be incredibly easy. It becomes a means for intel to make gains in the market regardless of whether they have the best product available.

cheers


Edited by m0zes: 24/8/2008 8:28:20 AM

-----

superfireydave 
24/8/08 11:05:30 AM
Titan

"-No discussion on how it will handle architectural/software differences in cards, for example if i've got an nvidia based based card and software that i'm running is coded specifically to take advantage of a particular way my card can render shadows and the other card in the array doesn't support this what happens?"
Nothing bad, because the tasks are seperated.
I imagine that you may have shadows that look different if there were a major difference in drivers, but from what I understand there's no mixing of tasks. One card gets assigned tasks, and the other gets assigned completely seperate tasks so it's irrelevant if both cards have the same specs.

You're right about Vista tho, however with DX, cards are backwards compatible. I assume the Hydra is smart enough to know which DX version it has, and to set tasks based on that. It might mean that you need two DX10 GPU's to do DX10 however, I'm not sure if you could render part of the scene in DX10 and the other part in DX8 for example, I'm guessing it just renders with the highest version of DX available by both cards.

-----
Mreow?
http://www.atomicmpc.com.au/forums.asp?s=2&c=9&t=17306

xguntherc 
24/8/08 1:35:43 PM
Overlord

how could you use ATI driver and Nvidia at the same time.. uh trouble.

it's not for mixing GPU mfg's.

-----
eVGA 750i FTW- -Q6700 GO @3.4Ghz w/Tuniq- -eVGA GTX 260- -2x2GB G-Skill- -Seagate Cuda 500GB 32mb 7200.(11)+ 500GB Ext- -ASUS 20x ODD- -Samsung 226BW & 2253BW Dual 22's- -Bose Companion 3-

http://tinyurl.com/3DMark06


superfireydave 
24/8/08 3:16:05 PM
Titan

Quote by xguntherc
how could you use ATI driver and Nvidia at the same time.. uh trouble.

Fairly easily. Vista doesn't allow it though, most likely to avoid potential driver issues.

It's certainly possible, and works fine.

it's not for mixing GPU mfg's.
What the hell is an mfg?
If you mean vendors (e.g. ATI, nVidia etc), then yes, it's possible and that's one of the aims of this technology.

Of course, you'd know that if you read the thread, or the articles -_-

-----
Mreow?
http://www.atomicmpc.com.au/forums.asp?s=2&c=9&t=17306

battlefield_gir 
24/8/08 3:48:36 PM
Titan

should get a job there :P

Kfar Netter, Israel.

-----
Minister for Education, Innovation, Science & Research.


R.I.P Josh Woods

Goth: Am I the only one who somehow swam through the wall of the vagina and got lost/stuck inside




"There is no probl

Sparky 
24/8/08 8:42:31 PM
Guru

Actually I don't belive the specific intent was to mix cards, just that the reports read slighly wrong. I think what they were trying to say was that it will work with any card make but the cards in an array (graphics array - sound oh so cool) need to be similar.

But I could be wrong.

Also larrabee is an x86 processor so there is no reason as to why you would not just use smp for multi gpu larrabee rendering. (nvidia looks like it needs to consume via to stay competitive in this area)

-----
The Olympic spirit http://xkcd.com/284/

superfireydave 
24/8/08 9:13:09 PM
Titan

Quote by Sparky
Actually I don't belive the specific intent was to mix cards, just that the reports read slighly wrong. I think what they were trying to say was that it will work with any card make but the cards in an array (graphics array - sound oh so cool) need to be similar.


Perhaps.

I read it as the Hydra allocating tasks over the PCI bus to the cards and them sending it back, eliminating the direct cable link between cards for SLi and Crossfire, and enabling you to use any cards together for distributed processing.

-----
Mreow?
http://www.atomicmpc.com.au/forums.asp?s=2&c=9&t=17306

layzi 
24/8/08 10:17:19 PM
Learner
i think its main advantage is how it scales multiple gpus (load balancing) compared to traditional sli/cfx by allocating different tasks (on the fly)across each gpu as to split frame rendering/alternate frame rendering, thus resulting in near perfect scaling (yet to be proved) regardless of how many gpus. so basically you can hand pick any graphics card within a vendor e.g hd4870x2 + 2900XT + 4850 hell make it another 4870x2 providing the mobo supports 4 pci-e x16 slots and you will have a graphics processing monster that will last for years. if this concept works well then sli/cfx days maybe numbered.


Edited by layzi: 24/8/2008 10:20:05 PM

-----
future gaming=realism=physics processing=nvidia.

osama_bin_athlon 
27/8/08 1:10:57 PM
Hero
Immortal


the Inquirer says:
http://www.theinquirer.net/gb/inquirer/news/2008/08/20/lucid-makes-multi-gpus-easy

-----
....it's called a 'snuke'.

Vanne 
27/8/08 6:28:27 PM
Champion

mixing gpus is possible already, the DFI ultra D boards had this tech way back already. Both drivers for Nvidia and Ati can be intslalled.

(that said, i never did it, nor do i know anyone who did)

-----
DFi Lanparty Ultra D-Hardmod to SLi
AMD X2 4400+ Holy Toledo Batman
TeamXtreem Cronus 3-3-2-8 DDR500
2xAsus 7900GTX (GA SLi bridge)
Enermax Liberty 620W SLi
2xApple 20 Inch LCD's
Ageia PhysX

Gubble 
27/8/08 8:38:54 PM
Apprentice

From my understanding the technology can't mix cards that require different drivers (Nvidia with an ATI card)

http://techreport.com/articles.x/15367
Quote for the lazy.
"...One limitation is that the GPUs involved must all use the same graphics driver, so mixing a GeForce with a Radeon won't work."

However it could take 2 cards that use the same driver and deliver near perfect scaling
eg. A 3850 with a 4870. The technology is supposed to use algorithms to balance how powerful each card is and then send a balanced workload to each one.

So lets pretend the 3850 was half the speed of the 4870. Hydra would send approx 33% of the workload to the 3850 and roughly 66% to the 4870.

http://www.pcper.com/article.php?aid=607
Has some pics showing what gpu1 and gpu2 are rendering seperatly, and some info on it


Edited by Gubble: 27/8/2008 8:41:04 PM


Edited by Gubble: 27/8/2008 8:43:13 PM

-----
I am what I am and I do what I can

Vanne 
27/8/08 10:16:34 PM
Champion

http://www.viperlair.com/reviews/cpu_mobo/other/s939/dfiud/

Quote "Expanding on the Dual Graphics I installed a 2nd video card which was a little cumbersome, mainly because my main graphics card is a HIS X850XT which encompasses two slots. The 2nd video, or DXG slot, is only 2 slots away making the fit snug. To challenge myself and the DFI solution, I installed an NVIDIA based ASUS N5900 Extreme to pair to my Radeon based X850XT. Surprisingly the physical installation was the hardest piece of it. The X850XT's fan was rubbing on the N5900 and causing it to halt, which, I am sure, would cause an overheat condition in time. I placed a thick elastic between the cards ensuring not to get it close to rub the fan or anything important. Another note about the DXG solution, with the cards so close together, only a card designed such as the HIS solution (kind of a catch 22 isn't it :P) could work properly as airflow would be drastically reduced between the cards, giving your high end card (in slot 1) the least amount of airflow. The HIS solution brings air in from the rear back of the card and flows it across the HS out the back of the case, almost as if designed to be in a tight space.



The software and drivers for both NVIDIA and ATI ran fine (albeit not overclocked) in this system. The solution appears to work seamlessly, nicely dispersing the load between the cards (manual selection of course) and allowing up to 4 monitors, unfortunately I only have 2." unquote

so thats nothing new, even drivers play nice.

-----
DFi Lanparty Ultra D-Hardmod to SLi
AMD X2 4400+ Holy Toledo Batman
TeamXtreem Cronus 3-3-2-8 DDR500
2xAsus 7900GTX (GA SLi bridge)
Enermax Liberty 620W SLi
2xApple 20 Inch LCD's
Ageia PhysX

layzi 
28/8/08 9:09:41 AM
Apprentice
Quote by Vanne
http://www.viperlair.com/reviews/cpu_mobo/other/s939/dfiud/

Quote "Expanding on the Dual Graphics I installed a 2nd video card which was a little cumbersome, mainly because my main graphics card is a HIS X850XT which encompasses two slots. The 2nd video, or DXG slot, is only 2 slots away making the fit snug. To challenge myself and the DFI solution, I installed an NVIDIA based ASUS N5900 Extreme to pair to my Radeon based X850XT. Surprisingly the physical installation was the hardest piece of it. The X850XT's fan was rubbing on the N5900 and causing it to halt, which, I am sure, would cause an overheat condition in time. I placed a thick elastic between the cards ensuring not to get it close to rub the fan or anything important. Another note about the DXG solution, with the cards so close together, only a card designed such as the HIS solution (kind of a catch 22 isn't it :P) could work properly as airflow would be drastically reduced between the cards, giving your high end card (in slot 1) the least amount of airflow. The HIS solution brings air in from the rear back of the card and flows it across the HS out the back of the case, almost as if designed to be in a tight space.



The software and drivers for both NVIDIA and ATI ran fine (albeit not overclocked) in this system. The solution appears to work seamlessly, nicely dispersing the load between the cards (manual selection of course) and allowing up to 4 monitors, unfortunately I only have 2." unquote

so thats nothing new, even drivers play nice.




theres something really wrong about this picture, unless someone can clarify on how dispersing the load between the cards is possible without some kind of specialised driver or chip.

-----
FTW!
http://overclockingpin.com/final%20gtx280/790isli-triple-tek9-40.jpg

mark84 
28/8/08 10:02:38 AM
Hero
Guru


Quote by layzi
theres something really wrong about this picture, unless someone can clarify on how dispersing the load between the cards is possible without some kind of specialised driver or chip.



The Anandtech article goes into a bit of depth on it. Basically the Hydra gets feed back from each gpu when it finishes rendering. Then it checks to see how long it took the GPU to do it. If one takes longer than the other it'll balance the next task so the slower one gets less to do.

This happens for each frame being rendered too, so if a gpu is particularly better at one type of scene compared to another the Hydra will balance the load in that GPUs favour on the fly.

Really is an innovative way of going about multi GPU tasks.



I think having this chip start off on a 4870X2 type card in place of the PLX chip would be a good proving ground for this tech. Of course getting it on the mobo is the better goal.

Only down side that I can see with it, if it can do everything it says it can, is that it's only PCIE 1.0 compatible. I'm sure they'll get a 2.0 model in their next revision if this takes off though.


Also what happens when DX11 comes out? The chip interprets DirectX calls on the fly, but what happens when a new version comes out? Would you like flash the chip to update it or is that controlled through the Hydra drivers or are you stuck with what ever version the chip has on it?

-----
QX9650 | HD4870 | 2x WD3000GLFS RAID0 | Swiftech Storm | 2407WFP |

AtomicMPC Firefox extension
http://tinyurl.com/2qabr7

The Truth About Graphics Power Requirements V2
http://tinyurl.com/cj3pw

  1 | 2  | Next Page 
Forums | Graphic Hardware