AMD Raedon Pro duo

[Vividus]

Hi everyone,

Does anyone have the card below?
There is not much info online and I am wondering if it makes sense to have this as a main rendering card or it sounds too good to be true

https://www.newegg.com/Product/Produ...82E16814105070
https://www.amd.com/en/products/prof...ro-duo-polaris

Thanks!

[richard_rosenman]

VRay RT is optimized for Nvidia cards using CUDA. With AMD cards, I believe you would only be able to use OpenCL for rendering, which I believe is not as efficient as CUDA.

For that price, why wouldn't you consider an Nvidia card? Of course, I could be wrong and I am only stating what I have read thus far.

-Rich

[Vividus]

Hi Richard,

Thanks for your answer. The reason why I am interested on that card is because it is two cards in one so i would end up having 32 gb ram and I do not have that posibility with any envidia card at that price range as far as I know

Regarding cuda and opencl, interesting that you mention that, I believe it could be true, but I cannot find any doc where it says that. Do you mind pointing me to where it says that?. I remember reading or listening to an interview with vlado saying that vray works with both the same way but I could be mistaken but I also cannot find any article where it goes deeper into the topic

Thanks!

[super gnu]

just to note, for RT gpu you would only have 16 GB to use with that card. gpu renderers cannot share memory between seperate gpus, as the data transfer between the cards (even if on one card) is way too slow. that is, unless you use an nvidia card with nvlink. unfortunately those are quite outrageously expensive. also, as stated above, the focus of vray gpu development is on cuda devices now.


having said that, ive not noticed this card before.. two 16gb gpus for that price is quite impressive. id be interested to see benchmarks.

https://www.chaosgroup.com/blog/blag...ased-amd-cards

that should give an idea to work with.

[savage309]

Check www.benchmark.chaosgroup.com
Also, keep in mind that V-Ray GPU CUDA has more features than V-Ray GPU OpenCL.

Best,
Blago.

[super gnu]

Check www.benchmark.chaosgroup.com
Also, keep in mind that V-Ray GPU CUDA has more features than V-Ray GPU OpenCL.

Best,
Blago.

ive looked on the benchmark page.. dont see any search results for radeon pro, or radeon duo, or duo, or polaris, and a lot of the radeon results seem to be geforce results.. the page could do with some more precise filters!

[Vividus]

super gnu, thanks for the info, It seems that article is from 2016 and they talk about a different card and i believe they were talking about vray rt 3.6 not vray next. It seems that the pro duo is the only one that has the sharing memory feature as it is 2 cards in one if you take a look at the specs it says it is 32gb, 16gb per card

savage309, I had the same problem, I cannot find anything that is not nvidia related. Do you know where we could look at the features that are not supported by opencl comparing to cuda in vray next?

Here are the few videos and explanations i could find online:

https://www.youtube.com/watch?v=FVFz0c8yAcE
https://www.youtube.com/watch?v=To_N2xT6Xa8&t=605s
https://www.youtube.com/watch?v=A4dT2dxarY0

[super gnu]

just because it says "32 GB total" does not mean it can be shared for gpu rendering. AFAIK AMD does not have an equivalent of NVLINK, and even if they did, A) they would shout about it and B) i doubt highly it would be present in a 900$ card. This is a simple dual GPU board, of which there have been *many* previously, and they always list the total gpu ram as it sounds more impressive.


Nvidia saves NVLINK for their eye-wateringly expensive datacentre oriented cards. even the $3k TitanV has it disabled.

The link i posted is the closest info i could find to this card, as it uses practically the same gpu chip (same family, Polaris, so performance is likely to be in the ballpark of 2x that result, since this card is 2x Polaris GPU's. )


in itself this would be quite a good result, putting it in the ballpark of a 1080TI running Cuda, despite using the slower opencl. a 1080 TI equivalent with 16 gb of usable ram for $900 is not too shabby.. power consumption might be a bit high though.

as you mention, that result was for 3.6 though, and i know the gpu engine has been rewritten for NEXT.. so im not sure how the delta between opencl and cuda has changed... i fear Cuda is just even faster now!

[Vividus]

Yeah you could be right as It is hard to find info about and it looks too good to be true waiting will be a wise decision then.