View Single Post
Old 10th June 2017
Lives for gear

Originally Posted by mattiasnyc View Post
I don't have a Ryzen chip. If I had one I would have run the benchmark suite and shared the results.

It's interesting though because if the lower latency performance on Ryzen is not as good as people want because of the cross-ccx communication then increasing the memory speed would increase performance, since the memory speed and infinity fabric speed over which the ccx' communicate are tied together. Higher memory speed = lower latency when the ccx complexes share data.
Hmm.. so in this case, I probably wouldn't talk about increased memory clock like important attribute for DAW performance, until someone really proves that.

I directly asked Pete about that, because he did Ryzen tried it with different clock speeds and Dawbench.
And yes I agree with you, Ryzen's architecture with separate 4 core CCXs brings some challenges (as tradeoff for cheaper price and mfg. yield rate with smaller die) with regards to scheduling and sharing of cache data via Infinity Fabric, which is clocked together with memory. So while higher clock might be important for some workload with more intensive data sharing between CCXs, it definitely won't apply to everything. I can imagine, heavily threaded DAW workload with many individual track workers doesn't necessarily benefit so much from smaller latency at infinity fabric.
Which might be bit different than some games or compression algorithms with more intensive cache sharing architecture, where that increased memory clock speed might give you couple of percent.
Although different architecture.. with Intel setups, speed of RAM doesn't really matter and when I tested it with various DAW workloads, difference was at level of statistical error.