Which CPU should upgrade to?

I have an AMD Ryzen 2700 (8 cores, 16 threads,3.2 GHz base, boost Up to 4.1GHz) but I’m not getting as much out of it as I was hoping in Rack so I am thinking of upgrading it.

I’m looking at these but I don’t know which I should go for, there’s quite a price difference but doesn’t seem like a great deal of difference in the numbers. Would I see a noticeable difference with these?

  • £307.99 - The Ryzen 7 3800X has 8 Cores and 16 threads with a 4.5 GHz Boost, 3.9 GHz base clock, 36 MB total cache and 105 Watts TDP.

  • £191.04 - The Ryzen 3600X has 6 cores and 12 threads, with a 4.4 GHz Boost, with 3.8 GHz base clock, 36 MB total cache, and 95 Watts TDP.

  • £224.99 - The Ryzen 5 3600XT has 6 cores and 12 threads, with a 4.5 GHz Boost, with 3.8 GHz base clock, 35 MB total cache and 95 Watts TDP.

Any others I should consider?

There is also this post so I have no idea what it is I should be looking for.

Ryzen 4000 series?

2 Likes

A 3700X or 3800X gives you about a 20% performance upgrade on benchmarks, but I’m not sure how that really translates to VCV. Are you getting audio stuttering on complex patches?

1 Like

Upping thread count gives diminishing returns with VCV, I’d suggest buying a proc with the fastest single thread performance you can afford.

4 Likes

Yeah I am, it gets to the point where even with really high buffer settings I can’t run the patches in real time. Strangely it’s always when Rack gets to about 60% in my task manager.

How many engine threads are you using in Rack? Admittedly getting more that 50% with any app isn’t easy, but I would think if you bumped the thread count you could do better.

Have you turned on the CPU meters to see if any of the modules is the cuprit?

1 Like

I’m using all threads that I can, 8 for my current cpu. I can certainly push things further with that rather than 1 or 2.

I have tried to find particular offending modules but there doesn’t seem to be anything obvious and no really big hogs. I’ll screen shot a patch tomorrow, perhaps I’m missing something.

1 Like

Have you tried using the processor lasso approach described in the other thread, instead of using more threads?

I’m not sure what it means, restricting the threads to 1 or 2? If I change to 1 when I’m hitting the limits then there is a very noticeable difference, set at 8 I can do a lot more.

No, restricting the threads to a specific processor complex. I seem to get better sound on my Ryzen 1700 by using lasso to limit Rack’s CPU affinity to one CCX (cpu complex). (available cores per CCX depends on the processor)

1 Like

So are they always grouped next to each other? I mean, if I choose 4 threads in Rack how do I know that those 4 are in the same CCX?

In Lasso, you set the CPU affinity for Rack. Then Rack will only see as many threads as Lasso lets it see.

So first you research and find out how many threads there are in a CCX for your CPU. AFAIK, the threads are grouped together in a CCX. Then go into Lasso, locate Rack in the list of running programs, and under CPU affinity select that many threads. The CPU affinity shows each available thread. For example, if your CCX has 4 threads, only select the first 4 threads in CPU affinity.

When Rack runs it will only see that number of threads if Lasso is running. In Rack you can then set it to use all the threads it can see; ignore the “most modules” because that isn’t aware you’re managing the CPU with Lasso.

Does all this stuff make a big difference? If so, it seems like rack should use some API like pthread_setaffinity_np to make the engine threads all run in the same CTX, yes?

In my one test case it seems to have made a difference. But I’ll see if it’s a placebo or not…

1 Like