Multiple different GPU support

Hi,

I have 3070 RTX and 4090 RTX, is it possible for Topaz Video AI to use both to render one movie ?

(without any manual work like dividing file to two pieces and running two instances of Topaz)

You have 2 GPUs completly different, in the same PC case??? :roll_eyes:

1 Like

@r4089187 you could start two intsances of the application by holding down the shift key when double clicking onto the application to open it if you are on windows. then process one part of the video with one card and the other part of the video in the second application with the other card. when finished just merge both parts together with a tool of your choice.

1 Like

@JakSpoon Yes, for computing.
@Imo Did you even read what I wrote ?

On my computer, (Windows 11), there’s an option for ā€œAutoā€ and also ā€œAll GPUsā€ under Preferences… Processing… AI Processor.

Maybe monitor GPU activity while trying both options? Good luck

Sorry to break it to you, but nowadays having more than one GPU in the same PC is only counterproductive (at most, you can actually ā€œuseā€ the integrated graphics in the CPU + the discrete GPU such as a 4090)… but having even two identical GPUs connected in SLI/Crossfire, does not produce a doubling of performance, on the contrary… sometimes the performance is even worse!

In cryptocurrency mining this does not apply, good CPU got enough PCI-E lanes to work efficiently with even 8 GPU’s at a time, the problem is if certain software does take the advantage of multiple GPU’s and SLI/Crossfire is not needed for this.

2 Likes

So if there is an option ā€œAll GPU’sā€ it should take advantage of all GPU’s… will take a look at this.

basically the question you are asking if Topaz is coded to utilize multiple GPU.
I personally don’t know.

It is. If you have two matching GPUs and select ā€œAll GPUsā€ in TVAI, you’ll get a performance improvement somewhere around 30-35%. But if your GPUs don’t match, the slower one will bottleneck the faster one and performance will be degraded.

I use two GPUs in my system. TVAI is set to use the faster one for its AI processing. The slower one does everything else. This prevents TVAI from slowing down other apps and the OS in general.

Very interesting

But if your GPUs don’t match, the slower one will bottleneck the faster one and performance will be degraded.

You mean slower than running 1 GPU that’s faster or that the faster card will not be fully utilized ?

When I tried it here, the two GPUs together produced a frame rate slightly faster than the slower GPU alone and way, way slower than the faster GPU alone.

I’d be very interested to see what happens if someone runs TVAI on an 8 GPU crypto mining rig.

Would running two instances really be better than running both GPUs together in one instance? Also, does the ā€˜All GPUs’ option ignore integrated cards? How would you prevent TVAI from including an integrated card - would you have to disable the card?

Yes, you’d need to disable any GPU you wanted to exclude. I posted a request for better GPU selection a little while ago, but I guess there aren’t many people interested in using multiple GPUs with TVAI.

If TVAI is fully able to use all GPUs, it really comes down to having a CPU that can keep up with more GPUs.
If you did have a rig with 8 GPUs that were as good as let’s say a 3080, you would need the new threadripper with 64 cores to be able to get the most out of them.

1 Like

Yes, it would definitely be a monster. Open frame with GPU slots and multiple power supplies. Would probably need its own circuit breaker.

Mining rigs don’t seem to need a lot of CPU, though. I’ve seen some running Celeron processors. I suspect that as long as the CPU is adequate for Windows and TVAI and the AI processing is going to the GPUs, that would probably be enough.

TVAI is not the same as other AI processing.
TVAI require a powerful CPU, otherwise it will bottleneck your GPU.

Considering all the people grumbling because they were happily running TVAI 3.x but discovered they can’t run 4.x because their 10 year old CPUs don’t have AVX2, swapping in a new motherboard and CPU that could handle TVAI would probably pale compared to the cost of the GPUs and house wiring.

Apart from any performance issues or gains, I just found out that running an RTX2060 and an RTX2060 Super under VEAI 4.04 with the All GPU option causes the GPU temperature to drop from 83 degrees Celsius (1 GPU only) down to 65 degrees Celsius on both GPUs … according to GPU-Z. I think that’s cool, sort of. VEAI IRIS Previews and Exports seem to run faster but that could ā€œwishful thinkingā€. (I didn’t plan on running 2 GPUs in the same PC. I bought the second GPU for my other PC only to find out that the Motherboard doesn’t have enough clearance for a 2-slot width RTX GPU).

If you set it to Auto, does it use 1 GPU or both of them?