Dual mismatched GPUs getting better performance

This is an early post on this topic, as I literally just put this to the test and was pleasantly surprised with the result. VERY surprised, but still in a good way.

First, I have been told running dual GPUs is only beneficial if they are a matched set, like dual rtx 3080s, or dual rtx 3090s. I still haven't gotten around to testing the 3060ti + 3090ti setup...BUT...I realized the i9-13900k has an IGP unit, and it is surprisingly powerful. Not 3090ti powerful by any means...but still worthy of note.

So, I decided to just test it. I opened and ran a fairly fusion heavy 4k project with the 3090ti only. No proxies, no render cache...nothing. Just pure GPU grunt to crank out the frames. I kept the task manager open on the side to see if the CPU was getting overly involved, just to make sure the notable differences were going to be attributable to the GPU...and it was.

Next, I closed DR, switched the monitor to the IGP, and restarted DR. I could see the extra activity kicking over on the IGP, while the 3090 was sitting idle. I had the DR preferences set to automatic. So, it was allowing DR to display via the IGP (no other option), but could still opt to utilize the 3090. I set the playhead, and hit play. The playback performance was DRAMATICALLY better. Sections that previously were dropping entire sections of 30 frames were now playing back. Yes, they were still dropping frames, but the playback was stuttering bad...not slideshow bad.

Not sure what to make of this just yet...and like I noted, I still have to do the 3060ti + 3090ti test. Maybe it could be worthwhile to do a triple mismatched GPU test??? Given what I have been told on the mismatched GPUs not playing well together, and what I just witnessed for myself...I have a theory. The reason for the reported subpar performance of dual mismatched RTX GPUs may be due to a limitation of the Nvidia driver when handling two mismatched RTXs. This cannot be the case for the IGP and the 3090, since they each have their own drivers. Maybe DR is able to leverage both at maximum capacity due to the discrete drivers handling their own GPU interconnect. All I know is, DR was demonstrating notably superior playback with the mismatched IGP + 3090ti.

Alrighty...I have an update on this. I finally got a chance to do a playback test for straight 309ti vs IGP as the main display with the 3090ti as the backup. It's mixed results.

When I used the 3090ti exclusively, the fps reported in the color page jumped erratically from 0 up to 15 fps, spending most of the time bouncing between 3 & 4 fps. When I had the IGP handling the display, the fps was a reasonably stable 4-5 fps. So, technically speaking, the 3090ti alone was hitting higher fps numbers...but not consistently. The playback looked more stable with the IGP handling the main display.

As it is, I just don't know what to make of it. Clearly, the avg fps was marginally higher with the IGP handling the main display. I guess I will need to do a render test to see if there is a notable improvement with render speeds...although, I am doubting there will be any notable change...but there is some uncertainty. The i9-13900k has the AV1 encoder, and I do intend to start exporting to AV1 instead of h.265. That's going to be entirely on the IGP. So...maybe there will be some beneficial collusion between the IGP and the 3090ti during the render.

I did 4 different render tests. The clip was 1:16:15 @ 30fps 4kUHD. It had some fusion in it. Most of the video footage was 5k, with a few clips being 6k. There were places with several layers of footage stacked with transforms, etc applied. Each render was to quicktime h265.
The setups were:
1) IGP driving main display + 3090ti, w/encoder set to NVIDIA
2) IGP driving main display + 3090ti, w/encoder set to Intel Quick Sync
3) 3090ti driving main display + IGP, w/encoder set to NVIDIA
4) 3090ti driving main display + IGP, w/encoder set to Intel Quick Sync

The results--
1) 1:21
2) 1:15
3) 1:14
4) 1:11

Clearly, DR is faster with the 3090ti as the main display driver, but not by as much as I would have expected. And, the intel quick sync is a definite speed booster. I will be changing my default to use the IQS from now on. That said, I am still a bit mystified why the playback is smoother with the IGP driving the display.

Anyway, make of it what you will.