RTX Realtime Multi-GPU performance quiestions

I did a little testing in Create version 2022.2.1

My local PC has 4K display and single A6000 GPU ( not the latest released but the 2021 ver ).
In Create I have 1280x720 resolution and in this first test I keep “fill viewport” ON.

As you see, on my local single A6000 I get 30-34 frm/sec.

But with DUAL 2xA6000 on Infinite-Compute I get only 25-27 frm/sec.
On an Infinite-Compute cloud Virtual Machine with a single A6000 I get roughly the same results (this is not recorded on the above video).

Then second test, resolution 1280x720 but this time “fill viewport” OFF:

First my local single A6000 I get 23 frm/sec.
On the same cloud machine upgraded to triple A6000 I get only 21 frm/sec.

Note: I don’t have the ability to install dual or triple A6000s locally at this time

I feel like:
A.) A single A6000 (or even the soon to be out 2x speed cards announced at GTC) will not not be fast enough for some of my large projects… so dual, triple and even more if performance scales reasonably would be very useful now and in the future.

B.) As you can see above… I’m experiencing “reverse scaling” three GPUs are slower than one GPU.

I know that in rendering in Path-Traced mode I get 4.44 more power with 3xA6000 at Infinite-Compute and that is awesome!
I would love to benefit from that hardware also in RTX Real-Time mode :)

Yours,
Pekka

Hi Pekka,

There is a possibility that with such a low viewport resolution and perhaps some CPU-side limitations you will run into a perf degradation when enabling MGPU.

Can you try to bump up the viewport resolution? Also can you enable the profiler window with F8 and mark down the CPU and GPU times, it would be interesting to see the bottleneck.

Matt

1 Like