Update IV - Last update - This will be the last time I tinker with it for a while
Conclusion
Increased memory bandwidth beyond 6000mhz for this workload helps even out performance for this demanding chronos/gaia workload but doesn’t improve overall speed by much.
You have to have a special motherboard and get lucky on the RAM vendor / binning and CPU IMC lottery to achieve stability in the high 7x000mhz range or beyond.
Was it worth it now that I already had 6000mhz RAM - not really.
My FOMO is your gain.
ORIGINAL POST
When using gaia + chronos to upscale 1080p 29.97/30fps to 4K60 I am getting between 6.7fps-9.4fps (fluctuates constantly).
my setup:
13900K
DDR-6000 RAM (@6000mhz, not overclocked due to stability issues)
4090 (overlocked / water-cooled)
I’m wondering if exploring faster CPU RAM will help at all given that these are gpu-heavy models.
Every benchmark I’ve seen is for gaming (1-4% improvement). The only relevant ffpmeg-related posts simply mention general memory bandwidth benefits when going from 1080p to 4K encodes.
Would going to 7200/8000 mhz ram (supported by my mobo) have any appreciable effect on these FPS numbers given that most of the work is being done by the gpu? I have a lot of old tapes to go through and if the $270 in RAM can save me some $$$ on electricity I’m all for it.
Or have I maxed out and just need to deal with it?
In theory you are all correct, but any one of you actually put this down to the test? before and after memory upgrade? it might provide mild improvement, which will be diminishing return.
Alas the last time I tried overclocking it, the system became unstable whenever I tried to mix cpu overlocking / xmp II / ram overclocking . It’s stable right now with CPU @ 5.7ghz (all cores) and XMP II timings
Just purchased G.SKILL Trident Z5 RGB Series (Intel XMP 3.0) DDR5 RAM 48GB (2x24GB) 8000MT/s CL40-48-48-128
Give me until maybe the weekend or so for it to arrive, bios updates, stabilize the overclock, and for me to run a battery of tests. I’ll benchmark as follows:
existing ram + 3.5.x benchmark
existing ram + topaz 4.0 benchmark (if it’s out by then)
existing ram + anecdotal test using my videos.
new ram + 3.5.x benchmark
new ram + topaz 4.0 benchmark (if it’s out by then)
Let me give a shot anyway as I won’t plan on overlocking the memory if I don’t need to.
Additionally, both my CPU and GPU have AIOs and my GPU AIO blows directly outside of the case. That, coupled with lots of case fans means that the inside of the case is never above room temperature.
Maybe try to use two processes simultaneously, one works on the first half of the video, the second one on the other half. A first test made a speed improvement of 20 % with the 4090. After processing merge together the parts with no re-encoding via MakeMkvToolnix.
Interesting. My issue is that my workflow is already long [and storage-space-intensive] enough. I’m also saving to network storage and don’t want to worry about HDD-based bottlenecks (like that merge operation you mentioned).
I use chronos/gaia to go to ProRes 422 HQ in bulk and then use ffmpeg to manually set my crf + slow pass value the way I like it. It’s basically set and forget it (~40-50MB/s constant to NAS) until I need to switch from rendering to encoding (~300-600MB/s constant from NAS) once a week.
My GPU also says that it’s already at 99% utilization with the task at hand (if I am understanding this correctly). If it wasn’t I would definitely double-up so that I could make more use of the CPU.
Might be something to look into with the new cmdline interface in the future.
Memory bandwidth is especially important when upscaling images (x2, x4) because it’s performed by the CPU + RAM. And the higher the resolutions the more important.
. It’s stable right now with CPU @ 5.7ghz (all cores) and XMP II timings
