What would be interesting here is to see what happens with DLSS at Balanced and Performance modes, specifically for the DLSSQ-Transformers base line (58 FPS one).
I wonder how much FPS is gained by running "Balanced" preset VS the artifacts/quality degradation. Maybe the latter is near indistinguishable, while you gain 10-15 more FPS, which would make Framegen much better.
It would especially be interesting to see for card like 5080/5070Ti, because that might be all that is needed to make Framegen usable. As you write, it needs some basic decent FPS to begin with to work decently well.
So, I've done some additional testing. I still need to update page six or whatever of the review, but here's some additional data for you. This is still using Full RT (RT-Overdrive), but now I've got DLAA results tossed into the picture. What's key here is that DLAA at 4K gets just over 30 FPS average, with latency of 65 ms. So, first the numbers:
Code:
CP77 FullRT DLAA 4K AVG: 30.99 99pMIN: 23.6 Latency: 65.4 CPUCLK: 4809.3 GPUCLK: 2584.7 GPUTemp: 77.0 GPUPWR: 588.8 GPU%: 97.5
CP77 FullRT DLSSQ 4K AVG: 56.97 99pMIN: 38.0 Latency: 41.6 CPUCLK: 4525.7 GPUCLK: 2633.2 GPUTemp: 74.2 GPUPWR: 551.7 GPU%: 94.7
CP77 FullRT DLSSB 4K AVG: 67.34 99pMIN: 46.5 Latency: 38.2 CPUCLK: 4533.9 GPUCLK: 2656.2 GPUTemp: 73.0 GPUPWR: 523.9 GPU%: 92.4
CP77 FullRT DLSSP 4K AVG: 79.38 99pMIN: 47.9 Latency: 35.4 CPUCLK: 4530.6 GPUCLK: 2685.4 GPUTemp: 71.6 GPUPWR: 486.9 GPU%: 87.3
CP77 FullRT DLSSUP 4K AVG: 97.54 99pMIN: 55.9 Latency: 31.4 CPUCLK: 4522.1 GPUCLK: 2758.1 GPUTemp: 64.3 GPUPWR: 371.6 GPU%: 71.7
CP77 FullRT DLAA MFG2X 4K AVG: 57.15 99pMIN: 43.9 Latency: 77.4 CPUCLK: 4833.6 GPUCLK: 2598.5 GPUTemp: 75.6 GPUPWR: 593.1 GPU%: 96.7
CP77 FullRT DLAA MFG3X 4K AVG: 84.48 99pMIN: 53.2 Latency: 82.6 CPUCLK: 4865.0 GPUCLK: 2598.2 GPUTemp: 75.9 GPUPWR: 589.1 GPU%: 97.2
CP77 FullRT DLAA MFG4X 4K AVG: 110.30 99pMIN: 66.5 Latency: 86.3 CPUCLK: 4845.8 GPUCLK: 2596.5 GPUTemp: 76.0 GPUPWR: 583.3 GPU%: 96.7
Now, if you're only looking at performance, note how DLSS Quality upscaling is basically the same FPS (or frames to monitor) as DLAA + Framegen. Both give around an 83% boost to "performance" — in quotes because framegen isn't really performance as such. And look at the input latency. With DLSS Quality, latency dropped from 65 ms down to 42 ms. The game feels very playable at this point. With framegen, latency increases from 65 ms to 77 ms. It's not terrible, but it's also not better.
DLSS Balanced gives a modest boost to 67 FPS, and latency drops to 38 ms. It feels about the same, honestly, as DLSS Quality. Visually, the new transformers model also looks very good. You can absolutely play this way and I suspect most people wouldn't be able to guess whether they were running native or upscaled based on the visuals. (Performance is a dead giveaway, though.)
DLSS Performance mode upscaling gives 79 FPS with 35 ms latency, and DLAA + MFG3X gives 84 FPS with 83 ms latency. Those framerates are close enough to
look equally smooth, but the feel of the game and the responsiveness absolutely favors DLSSP.
Finally, DLSS Ultra Performance 9X upscaling gives 98 FPS and 31 ms latency. DLAA + MFG4X gives 110 FPS and 86 ms latency. Visually, you would think Ultra Performance would look pretty awful, but it's not really that bad. Again, most might not even notice. If you're versed in graphics and upscaling and know what to look for, you can tell there's upscaling enabled, but I'd almost venture to say that DLSS Transformers in Ultra Performance mode at 4K probably looks pretty close to FSR3 Quality mode! And the feel is way better than MFG4X.
I think the DLAA numbers with MFG give a good idea of what we can expect from MFG and DLSS on something like a 5070. If your base FPS (before framegen) is only 30 or so, MFG does look better — whether at 2X, 3X, or 4X — and I'd even say it tends to feel more playable as well. But that's a bit nebulous. As someone else pointed out, our eyes and brains are complex, so even though the input latency is worse, the visual smoothness counteracts that to some degree.
I ran around in Cyberpunk for 10 minutes or so with DLAA + MFG4X, as that's basically the "worst-case" option for the 5090 right now. It was absolutely playable. A bit sluggish feeling, yes, but for me I'd say not as bad as playing with a controller on a console (shots fired!) And because our brains are complex, you do get used to the slightly higher input latency after a minute or so.
The net results of MFG are interesting as well. So we have DLAA with 31 FPS, sampling input every 32 ms. With one frame of latency, that works out the 64 ms and basically matches up perfectly with the benchmarks.
Turn on MFG2X and we get 57 FPS with input sampling every 35 ms (ie, running at 28.5 ms). Now there's approximately four frames of latency (relative to the MFG rate), because the GPU renders frames 1 and 3 and generates frame 2, and frame 1 gets sent to the monitor probably around the time that the GPU is rendering frame 4. That seems to be how the math works out. Four frames of latency at 57 FPS would be 70 ms, and the measured value is 77 ms, so that's relatively close — it's 4.5 frames of latency.
MFG3X bumps the framerate to 84 FPS, but input sampling is only happening every third frame. So input sampling at 28.16 FPS and happening every 35.5 ms. And I think now we end up with about six frames of latency relative to the 84 FPS. So 84.48 FPS would be 11.8 ms and times six gives 71 ms... which means it's actually seven frames of latency at 82.6 ms.
MFG4X gives 110 FPS, input sampling at one fourth of that or 27.6 FPS. Which means it now happens every 36.3 ms. Really not much worse than the MFG3X result. Except because of the higher generated framerate, we're now getting about 9.5 frames of latency (86.3 ms) relative to the generated 110.3 FPS result.
What does that mean? If you want a really low latency with MFG4X, you're probably going to want a generated framerate well above 240 FPS. And a 240Hz or even 360Hz / 480Hz monitor. I think any latency result below about 40 is going to be plenty smooth for most gamers. And if you're getting 300 FPS generated rate, that's a 75 FPS base rate before MFG4X.
But, as I've been saying since framegen came out, it's totally not right to pretend that generated framerates are the same as rendered framerates. They can look similar, but the feel can be very different, and the gap just increase. If you have MFG4X running at 110 FPS, as shown here, it looks way smoother than a game running at 30 FPS, but it feels just as laggy as a game running at something closer to 25 FPS.
Facebook
Twitter
Instagram