Trying to find 2.8 benchmarks for rx 5700 xt, Sapphire / PowerColor (Red Devil) also Gigabyte ?
Brand does not matter for performance (they will all be in 1-2% variation). If you want to see how loud and how hot they are just check gaming tests.
Thanks but specifically looking for the difference in brand.
There is no difference they are all the same. Buy the cheapest coolest and quietest.
Fine I will then. :-p
If you have no other specific reason to buy amd than I would suggest a 2060S or 2070S. CUDA is more stable widely supported (not only blender but for other computation loads) and most importantly Nvidia drivers are a lot more reliable. Nvidia has implemented RTX cores to cycles renderer via OptiX api which makes them almost 50% faster in average. (still not 100% supported but they are working on it)
If you really want AMD go for it but these things needs to be told.
I’ve looked into it and apparently RTX 2060 Super is only just comparable to GTX 1080 but slower ram and the 1080 is less performance RX 5700 XT. Price the 5700 comes out on top but there is another problem with Blender itself at the moment, 2.81 AMD seems to render slower than 2.8.
Though I do like AMD I have to agree that Ton and Blender team are not putting as much effort into supporting OpenCL computing (along with AMD) as they are with RTX/Nvidia.
Though all I have is AMD systems and GPU’s, recently 2.80 to 2.81 there was nearly 50% rendering reduction. after a ticket was raised, they did “fix” most of it so now rendeirng is only 20% slower then it was…
Though AMD is sponsoring Developers, it seems the primary concern will be Vulcan migration (impacting Eevee only) though I do have hope that they will also address the above and even more so drive more performance out of OpenCL…
Honeslty I’m very tempted to use AMD’s RenderPro engine to replace Cycles, as it also supports Hybrid rendering (mixing Vulcan for general rendering and then Ray tracing as a final pass) giving best of Cycles and Eevee in one go. Issue is that not all shaders are direclty transferable.
So though I do enjoy AMD cards, its hard to recommend right now due to limited Blender intrest in openCl
yes, 2.81 is slower, as per my above post, there is a bug that they haven’t fully fixed (they mentioned its something to do with the new Noise shader… but not much beyond that. more details in the below thread
Just tested official release 2.81a (RX 5700 XT) and the render was even faster than 2.8, will test again but I know there was a mention of 2.81a having performance improvements so looks like they may have fixed it.
What scene did you use to test the performance jump?
I want to duplicate the result with my setup and compare if I see improvement or drop.
A personal project, would u like the file ?
can you test the BMW or Classroom scene? that’s available on Blender site.
If indeed you ahve improved render speeds, then this points to a Driver issue where RX 5700 has speedup where as Vega and Polaris cards have reductions…
If you can’t test the BMW or Classroom scene, then yes i’d like to test the same project/file you are usign to confirm
Strangely got a different result today, yesterday’s test of the BMW scene in 2.8 was 1:10 but today’s results …
BMW 2.8 = 1:27 / 2.81a = 1:26
Classroom 2.8 = 2:56 / 2.81a = 2:51
My scene 2.8 = 5:25 / 2.81a = 5:19
With the Classroom I set to GPU & samples 256, I let the CPU load run first and cancelled then run again to get just the GPU result.
Hmm… that is getting weird…
at least your 2.8 and 2.81a have same results which is far better then most of us non RX 5700 users…
What OS are you using and what drivers?
So did a BMW test
2.79 - 1m24s
2.80 - 1m24s
2.81a - 1m43s
2.82 - Failed to render - todays build seems to be broken… “Split kernel error: failed to load kernel_path_init”… will try tomorrow and see iff its’ fixed.
Still the fact that you are getting identical times between 2.80 and 2.81 (well actually faster) is really good to see… but wonder why most AMD users dont
Windows 10 and recent AMD 2020 software & drivers. What about you?
Same… Windows 10 and recent adrenalin drivers from december…
hmm this is now really really strange… wonder if AMD made changes in OpenCL part of the driver that affects Polaris and Vega negatively where as it has no impact on Navi…
Now I want to get a Navi card and test it on my system hehe…
Still it’s good to see that at least some AMD users like yourself have no issues
What card are you using ? U have quicker results than mine.
I have Vega 64, water cooled, and clock are slightly higher.
stil as you stated earlier BMW scene in 2.8 was 1m10s… and for soem reason 17 seconds were lost.
Do note that I also test few times. First render always has that “kernel compile” that adds additional seconds. as second and further re-renders dont’, I exclude the first render.
So if you take the above into consideration your times are most likely faster then mine when directly comparing.