Updated results after upgrading my CPU and RAM. I didn’t build this pc for Topaz, but just thought it would be interesting to see how the results here were affected by the CPU.
The screenshot below compares each benchmark side by side after the CPU and RAM upgrade. These are on slightly different version of Topaz. I assume that doesn’t really have much effect on how the models perform.
As for Starlight Mini, processing a 2 second, 29.97fps 1920x1080 video took me 2m20s at 1x, and 9m30s at 2x with these latest specs. I saw virtually no difference when setting the max VRAM to 100% or 90% (and I confirmed in MSI Afternburner that Topaz kept to the 90% limit in 7.0.1).
Before this upgrade the 2x run with Starlight Mini took 11m30s so a slight speed improvement there too. Though I suspect that is entirely due to the RAM and not the CPU as I was previously maxing out my 32GB of RAM. With 96GB of RAM the peak RAM usage I saw was 63GB so plenty of headroom now.
yes, it’s faster settings for the RAM provided by the RAM manufacturer. It should almost always be used since TVAI performance quite benefits from fast RAM, especially when any image upscaling is involved.
Btw, it’s called Expo for AMD systems and XMP for Intel systems. In both cases, the settings are applied in the BIOS. Sometimes there are multiple profiles. And in case of high performance RAM modules, it can be a good idea to check the cooling for the RAM since those profiles often use higher voltages (=more heat).
With the newest Epyc CPUs (AMD Epyc 9115, 16C/32T, 2.60-4.10GHz max. DDR5-6000 (PC5-48000) you can reach 576GB/s, a Ryzen 9950X has 89,6 GB/s with 5600 RAM.
Here’s some fresh tests with 7.0.1. I also benchmarked Starlight mini by generating jpg images insead of a video. With this method i got reliable start and stop times. I included the test values from @PC100 in this little table:
It looks like you have 64gb of RAM. I know it’s easier/cheaper to get fast (6000 cl30) RAM with tighter XMP/EXPO sub-timings with lower-capacity kits. I over-spent on my AMD sim rig and got 64gb. (By the time I actually need 64gb for a sim, there will probably be better-binned RAM available.) I’m running 6000 cl30 but the subtimings are looser than the 32gb kits. (The kit also runs 1.4v too)
Are you already running 6000 cl30 RAM?
Also, Nvidia 5000 series seems to have a LOT of headroom for VRAM overclocking, that could yield a good performance boost.
Temperatures could also be looked at, but the benchmark run doesn’t heat up hardware that bad. My room temperature was ~72f when I ran the benchmark.
Edit: Looking at your benchmark results closer, it looks like you’re getting the extra performance of the 5080 over my 4080S in the benchmarks that lean on the GPU most. (1X, Gia, etc) But are slightly slower in CPU-haeavy benchmarks like the 4x stuff.
The gap is kind of small in both scenarios either way.
I was wondering how an aging platform based on the nearly 12-year-old Xeon E5-1680 v2 (overclocked to 4.4GHz) non-AVX2 processor would perform compared to new hardware with the same GPU (RTX 4070 Ti). I found a ForSerious result with the latest AMD Ryzen 9 9900X processor.
Wow, I even managed to win in Nyx 1X test by +6.97%
Conclusion: Topaz, don’t drop support for processors without AVX2 and follow Adobe’s path!
The folks at Adobe (dirty scoundrels) recently blocked Photoshop 2025 from running on processors without AVX2 - but in a few minutes, I created a patch (also for Lightroom Classic 2025) and everything works perfectly!
Nice!
Nyx 1X is weird though. I know for a fact that when processing real FHD videos, it gets like 9.7 fps. With TVAI 7 the benchmark has started reporting lower numbers, but the real speed has stayed the same.
So the current benchmark test does not test Starlight mini….
Which is the model I will use 90% of my rendering.
But am sure you folks have tested pretty extensively and at least can draw some conclusions.
Given the latest version of Starlight mini which is in its infancy and probably not well optimized.….
It “appears” that Starlight mini is not affected greatly by powerful CPU…. Does your testing coincide with this?
Is Stalight mini affected greatly by amount of RAM? 32 GB vs 64 GB….DDR 5 vs DDR 4?
I understand completely that of course there will be improvements when using better hardware…..more expensive and latest is simply better.
But the degree of “better” I have seen from others jumping from lower end to the latest high end…..on Starlight mini….seems minuscule and not worth a $5000 investment. A 5090 alone is a $3000 chunk of change. All that for a half a frame improvement?
I would rather build a semi low cost rig and let it be my stand alone render rig going for 2-3 days or however long….fire and forget….and drive on with life.
video with a resolution of 1280 x 720 was processed on my 3060 Ti at a speed of 0.4 FPS. Where I live, one 5090 costs as much as 14 3060ti. 14x0.4=5.6fps The conclusion is this - you need to assemble a video farm.
