I think part of this is simply driver optimization. Unfortunately, at the moment the 5700 series seem to be in a weird spot regarding compute. Blender works well, but most other things appear to be broken, there’s some word going around that this may have been intentional to prevent miners from eating up the entire stock on launch. Even on Linux, AMD have avoided mentioning anything about support for the 5700 series in their ROCm compute platform. Additionally, their architecture documentation implies that there are still optimizations and extensions for new features in Navi that haven’t been released yet.
how did you get these resutls, as in how does one find other comparisions, want to see Vega 56/64 but no clue how to pull that.
I can’t find a way to pull other results either, at best there is https://download.blender.org/institute/benchmark/latest_snapshot.html for a manual comparison. These ones were just my own from running Blender Benchmark on my machine (same driver), the utility gives you the option to upload your scores at the end.
I guess time to run the Blender Benchmark on my machine for comparison.
Side question. Are you experiencing any “Tile performance issues” for my Vega’s on Windows 10, 128x128 tiles work great, but 32x32 is nearly twice as bad. Let alone that rendering on 2 GPU’s is only loading the GPU’s at 60-70% per GPU.
I haven’t had either issue you’re mentioning, both GPUs get loaded to 100%, taking ~44s to render bmw27 at 256x256 tile size, and ~55s at 32x32 tile size (in 2.8). Really great value tbh, considering that those numbers are apparently Titan V territory, which is a much more expensive card.
Dam… I’m happy you are getting the proper scaling. I’m still investigating on my side as my 2 Vega 56 dont’ cooperate and low tiles are so slow that rendering with CPU mixed in is a performance degression
Mixing in the CPU slows mine down significantly too (GPUs finish up all their work fast, so the render ends up just sitting around waiting for the CPU to finish its tile), I just leave it at the two GPUs.
Intresting that this CPU+GPU performance regression is happening on AMD setups. yesterday I raised a bug report for my GPU not running at 100%, as I did everything.
But I’ll continue to investigate the CPU+GPU, as in the past, I had a noticable/measurable rendering reduction when mixing my TR + Vega card at 16x16 and 32x32
But happy that at least you are getting 100% from each GPU. I love Vega’s for rendering. at peak they dont’ even consume 150W (vega 56s)
