A Quick Look At The Blender 2.82 Performance On Intel + AMD CPUs

[phoronix]

Phoronix: A Quick Look At The Blender 2.82 Performance On Intel + AMD CPUs

With Blender 2.82 having released on Friday, this weekend we've begun our benchmarking of this new Blender release as the leading open-source 3D modeling solution currently available. Here are some preliminary v2.81 vs. v2.82 figures on different higher-end Intel and AMD processors...

http://www.phoronix.com/scan.php?page=news_item&px=Blender-2.82-Quick-Benchmarks

[bridgman]

Interesting... I hadn't thought about Blender scaling well all the way to 2 socket server systems.

[ms178]

Any idea why Barbershop regressed significantly on 2P configurations?

[numacross]

Interesting... I hadn't thought about Blender scaling well all the way to 2 socket server systems.

Why wouldn't it? Tile-based rendering is usually embarrassingly parallel

[bridgman]

Why wouldn't it? Tile-based rendering is usually embarrassingly parallel

I guess I didn't know enough about Blender internals.

[oiaohm]

I guess I didn't know enough about Blender internals.

https://www.flamenco.io/

Blender is one the programs more cpu cores/threads the better. Dual core 64 core EPIC cpu( 128 core, 256 threads) is still technically on the small side for what it tested with blender tile rendering. Remember blender tested well past 4096 core solutions using render farm software to cluster computers into a single solution. Of course CPU NUMA to NUMA latency is way better to the network lag the tile rendering has been tested with.

If it possible to make quad cpu socketed motherboards with the EPIC cpus blender render farm might eat this up. 4x64 core EPIC 256cores and 512 threads. Even if making this did block out the possibility of using multi gpus.

Some configurations of scenes in blender are not GPU friendly thrashing between GPU and CPU to compute stuff can end up slower than staying fully on the CPU cores.

As you can see in Blender Institute render farm software it made for mixed configuration as in full compute nodes and compute nodes with GPU acceleration with the objective to farm the workload out as required.

With the EPIC single and dual cpu systems neither look like a balls to the wall we are maxing out CPU compute solution. There may be a untapped market. Of course I might be wrong as well. Even so it could be a fun item to leak of different bench marking sites to causing media attention.