With new Intel/AMD CPU’s being released and hopefully the 5090 card released early next year, I’m looking to build a new PC - or upgrade what I have now.
For context, I have absolutely tons of SD (PAL) footage that I upscale using x2 Upscale with the Rhea setting that gives me good results… but even with my current i9-13900KS / 4090, an hours worth of video is taking just under an hour (so almost real time).
Do AMD threadripper CPU’s have any benefit over intel in Topaz Video AI or are they just a waste of cores that don’t get used? I notice my GPU is almost maxed out when Topaz is running but the CPU only uses about 50%.
I have x2 machines both with 4090 cards. While the 5090 performance specs are unknown still, would putting both 4090 cards into one machine be beneficial than upgrading both machines to the 5090 or do you get diminishing returns when using more than one card? IE: Does using x2 cards double the performance or not?
Both machines have 64GB ram, would upgrading to 128GB have any benefit?
The impact of video resolution and model selection on load and generation rate with a constant CPU of 12700K, dual channel DDR4 3200 memory channel, and 4090 graphics card. Using WPA for analysis with GPUView assistance
1920 * 1080 P60 FPS video, Artemis model. It can be seen that the CPU usage is stable and low, with a maximum wait time of 0.9 seconds and intermittent load on the graphics card. Each bend before the curve is waiting for memory to load data, which is severely limited.Here is an example of test data.
The advantage of using thread disruptors, such as the 7995WX, is that it has high-frequency DDR5 eight channel memory, which provides sufficient memory bandwidth and single core multi-core peak performance, thus eliminating the memory bottleneck of multi card machines.However, I have only tested it on the Xeon Platinum 8468 machine for now, and it has indeed improved the speed to a certain extent
I actually have more data analysis on single card or multi card workloads under different working conditions. After the CPU reaches a certain performance level, it is basically due to memory bottlenecks that the running speed cannot be further improved. At this time, servers with multi-channel CPUs, even those that are two generations behind, threadripper, and the Xeon W series, can effectively eliminate bottlenecks with a large number of 6-12 memory channels.
Despite the other reply, I’d say go for a 5090 or 4090 build. They are significantly faster than the 4080.
Also don’t skimp on CPU/Memory speed. Both must be very fast, or else you’ll bottleneck your GPU during upscales like in my 10900k + 4090 w/ DDR4 memory rig.
