AMD and Nvidia graphic cards at same time

[drewthuss]

is it possible to run (in my case) a gtx 970 and a RX 460 at the same time. Basically utilizing CUDA and OpenCL at the same time. I figured the gtx would handle displays and the RX would just handle OpenCL for FCPX. Is this idea feasible if so any obstacles i would run into?

 

[Revod]

That question could be reply by @Gigamaxx

 

[rwillett]

Don't do this. Bad idea. FCPX gets terribly confused.

You can't specify which cards FCPX uses and you could end up in a right mess.

Rob

 

[Gigamaxx]

I haven't run FCPX so I can't attest to that but it does work for a non igpu solution for RX cards in X99 boards. If you have them try them.

It works in Luxmark (open CL) benchmarks, and Uningine (open GL) benchmarks without any major hiccups. Nvidia 8400gs and RX 460 and RX 480 were tested by me others have tested various cards with the same success. The RX needs to go in the number 2 slot or you will suffer black screen issue.

Running the 970 with Nvidia drivers may present issues but a 960 was just tested and it ran with an RX card.

Disclaimers: Most CPUs only have 16 pcie lanes so when you use slot one alone it's gets 16 lanes (max performance). When you use slot 1 and slot 2 slot 1 gets 12 lanes, and shares 4 with slot 2. So performance for both cards gets reduced slightly.

X99 systems (and or Xeon chipsets) may have more pcie lanes supported so it shouldn't effect performance as much if at all.

If it works, let us know how well, I'll bet you could get some good benchmarks with that combo.

 

[rwillett]

I have done this with FCPX, though not with these cards and not recently.

FCPX doesn't like multiple types of GPU. This could be nVidia/AMD, AMD/Intel, Intel/Nvidia or if you are feeling really difficult nVidia/Intel/AMD.

FCPX doesn't know which to use and could end up trying to do all the hardwork on the slowest/cheapest/nastiest card in your mix.

You cannot tell FCPX to just use the AMD cards (or even the latest nVidia cards). It tries to work it out for itself and its not too bright

You might well get excellent benchmarks but a dog-slow FCPX system

Rob

 

[Gigamaxx]

I have done this with FCPX, though not with these cards and not recently.

FCPX doesn't like multiple types of GPU. This could be nVidia/AMD, AMD/Intel, Intel/Nvidia or if you are feeling really difficult nVidia/Intel/AMD.

FCPX doesn't know which to use and could end up trying to do all the hardwork on the slowest/cheapest/nastiest card in your mix.

You cannot tell FCPX to just use the AMD cards (or even the latest nVidia cards). It tries to work it out for itself and its not too bright

You might well get excellent benchmarks but a dog-slow FCPX system

Rob.

.

In the Radeon Sierra testers thread these issues have come up with Intel and the RX 480, FCPx the latest version was dropping out probably due to this and by playing with clover settings and removing certain sound fixes the user was able to get it working. By simply reverting back to the last version/prior version FCPX they had no problems but they wanted to work in the latest fcpx and with some minor tweaks they were able to run it without issues. Injecting Intel was one of the main issues and in the clover ACPi folder removing HD4600 aml. So by removing these fixes and creating a native approach to Intel it worked out.

By hiding the HD 4600 and still getting metal 30 bit color for the RX card it sees only the RX card, when using inject Intel it was dog slow. Using fake ID 0x04128086, and ig platform 0d220003 allows support without injecting Intel.

For Skylake models use fake ID 19120003, and ig platform ID 0a206080 and you will get SKL CRB HD graphics without injecting Intel. By injecting Intel you get HD 530. These two setups allow your RX to function at peak OS Sierra levels and not trigger an issue in most apps due to conflicting graphics injections/labeling.

 

[rwillett]

Thats good to know, but your average user is going to barf his/her coffee over their machine if you ask them to do that.

I have no doubt some people can do this but most people just want FCPX to work without having to inject anything.

Rob

 

[Gigamaxx]

Thats good to know, but your average user is going to barf his/her coffee over their machine if you ask them to do that.

I have no doubt some people can do this but most people just want FCPX to work without having to inject anything.

Rob

Actually, you're NOT injecting and getting native metal support.

 

[rwillett]

You are correct, I misread your post.

However most people will still struggle with doing all this to get AMD and nVidia working together to get this all working.

 

[Gigamaxx]

You are correct, I misread your post.

However most people will still struggle with doing all this to get AMD and nVidia working together to get this all working.

If you want easy you buy a real Mac. This is hackintosh, we're rebels.