Hi there,

I was wondering if there was any detailed info on what type of hardware would be supported for GPU based pyro acceleration?

Thanks,

Andrew

You need a GPU capable of supporting OpenCL. On the Nvidia side, that would be a Quadro x000 series (non-FX quadro; 600,2000,4000,5000,6000) or a GEForce 400 or 500 series. For AMD, you need a FirePro card (not FireGL) or a Radeon 5000 or 6000 series GPU. Intel's graphics are currently not able to support OpenCL, but they do have a CPU-oriented CL library which may work with Houdini in the future.

hi!!!

I have a Quadro 5000, but i think will be not enought for a sim, because of the memory that card have 3.5 gb. I'm wrong?

Is the mobile version of amd radeon supported? Like the AMD Radeon HD 6750M from the new MPB?

Thanks

alejandro
I have a Quadro 5000, but i think will be not enought for a sim, because of the memory that card have 3.5 gb. I'm wrong?

It has 2.5GB of VRAM, which is good enough for some fairly large sims (320x320x320 is the highest I've seen). The exact limit may depend on how many screens your card is driving, their resolution, and other apps using video memory.

adrianlazar
Is the mobile version of amd radeon supported? Like the AMD Radeon HD 6750M from the new MPB?

Mobile variants of the desktop GPUs I noted above will run OpenCL. Their main downfall is a shortage of video memory which will limit the size of the sim you can run. A GPU with 512MB of VRAM is generally only capable of running a 128x128x128 sim. The lower clocks and fewer shaders will also decrease performance quite a bit.

thanks for the valuable info twod, much appreciated

Thank you!!!

Yes, Thank you!

hi there ! first post on the forum !

eager to learn ..

* * *

first question here :

can anybody tell already if the bullet engine in 12 will already run / be boosted on GPU ?

another question that I have in my mind is what happens when You use 2 or more tesla cards. The size of the sim that a tesla card can handle depends on the memory of it, but do the memories of 2 cards add or not. My guess is not, as as soon as the card has to exchange data it slows down. but would be interesting to hear more about it from the sidefx people

Martin

What will Gtx 570 be capable of? it only has 1.28 GB vram.

The 570 probably max out with volume sims around 250^3. However, I have seen GEForce 570's loaded with 2.5GB of VRAM in the wild.

how about gtx 590?
it has dual gpu and memory of 3 gb (1.5 gb each) will it use all 3 gb or it will only use 1.5 gb? and will it use both gpu?

best regards
Patar

I would be very interested to know the answer to this question as well.

cheers

-G

patar
how about gtx 590?
it has dual gpu and memory of 3 gb (1.5 gb each) will it use all 3 gb or it will only use 1.5 gb? and will it use both gpu?

best regards
Patar

SLI or crossfire cards are treated as separate OpenCL devices, so each device will only see 1.5GB. The GTX590 is really just two GPUs on one card with a SLI bridge built-in. I don't think GPU acceleration for H12 will use multiple OpenCL devices, but this could change by release.

so its better to have 580 with 3gb than 590 with 3gb ?(since its single gpu)…

best regards
Patar

yes! 580 with 3gb is better…next step is Quadro with 6 gb.

Question to Houdini developers.
Can used Network of GPUs in Houdini 12???

Hi folks, this is my first post here…gonna be on H12 once out . In order to plan ahead, my GPU-related question is whether it will be possible to specify which GPU ( or simpler: card present in the system ) will be used for simulation. I understood that there´s only one that can be used…but is this necessarily going to be the graphics adapter, or can i let´s say have an FX5800 for display and a Tesla 2070 for sim? That´d be great to know, thanks a lot in advance!

Nvidia's newer drivers automatically route compute tasks to the Tesla card, and graphics tasks to the Quadro, if you have two in your system. This doesn't work with GEForce cards, however; just Tesla+Quadro. It also won't work with QuadroFX cards - just the newer Quadro 600, 2000, 4000, 5000 and 6000.

You can manually set the GPU with HOUDINI_OCL_DEVICENUMBER as well, which is what you'd need to do for your QuadroFX/Tesla combination.

sweet, exactly what i wanted to know. cheers!