View Single Post
Old 10th June 2017
  #22
Lives for gear
 

Quote:
Originally Posted by mattiasnyc View Post
Yes, he did. And I said "if" for a reason. Someone else linked to another person doing tests saying that increasing memory speed makes a big difference. So they can't both be right.
If you're talking about the previous linked post by J. Roseberry, then at the top he just assumes, it will be faster due to CCXs communication.. there are no comparable DAWBench results say at 2133 vs 2933, because he switched to 64bit DAWBench from his previous test at lower RAM clock.

So far, the most comprehensive memory tweak review for Ryzen 7 is there
AMD Ryzen 7 Memory And Tweaking Analysis review - An introduction at a double data-rate
Where you can see results for multiple games and applications. Results vary quite a bit depending on workload.. If our considered DAW load is closer to say Cinebench in MT mode, which is already very parallelized for start, then it's easily bring no improvement.

One remark, that their top clocked 3200 memory sticks has also lower CL (14 vs 16). Running such memories at this high clock is generally know to be very problematic with regards to use with all 4 RAM banks populated according to multiple other report also by my friends.. so it's questionable, if it will be even possible to run for example 4x 8GB setup like that.

Quote:
Intel not benefiting from faster RAM is completely irrelevant if the issue is the cross-ccx data sharing. Again, if.....
I'm aware of that and mentioned that..
Frankly, I'm glad, so far it looks like, it's the same with AMD for DAW workloads, because majority of people buy or consider Ryzen setup either for power/price ratio (or sentiment ).. and such fast RAM sticks, just adds to total price of their builds possible and increase possible compatibility problems at fully populated setups. So IF someone finally proves, it's irrelevant for this massively threaded workflow.. it's good news IMO.

Michal