Hi everyone!

While I've been a folder for quite some time, I was recommended by a friend to give Rosetta a shot. As such (it being the only BOINC project I've run), it's been fairly easy to get going. I did have a question or two though as Rosetta obviously operates a bit differently than the flags in F@H.

1. In the SMP core version of F@H, I can simply set the flag and F@H will auto-crunch one project with as many threads and cores as you have available. I see in the task manager for BOINC that it's downloaded and working on 16 projects at once (which is right as I have 8 physical / 16 HT cores). What I'm wondering though is that much like how I can put -SMP 15 in the F@H config to allow the GPU to work on one project as well, is there a noticable benefit to have Rosetta "tone down" on how many cores or threads it runs on? Should I just leave all 16 cores cranking along? I noticed after extensive F@H testing that the calculations were working a bit faster if I kept one thread free for the GPU work. If I maxed the CPU, it'd actually have the GPU slow down the CPU and both sets of projects would take longer. Is Rosetta like this or different?

Is there an "ideal" way to run Rosetta? I still want to use the GPU for folding though if possible.

2. Any particular settings that'd help Rosetta crunch faster? I'm allowing it plenty of disk space and enough RAM for all the projects it's downloaded. Is simply OC'ing enough or would tighter RAM timings help (currently at auto which I think is something like 11-11-11-36), though the RAM is fine at 8-8-8-24 or 6-8-6-24. Are there any other optimizations I can do or just let it work?

Thanks for the help.

Hi everyone!

While I've been a folder for quite some time, I was recommended by a friend to give Rosetta a shot. As such (it being the only BOINC project I've run), it's been fairly easy to get going. I did have a question or two though as Rosetta obviously operates a bit differently than the flags in F@H.

1. In the SMP core version of F@H, I can simply set the flag and F@H will auto-crunch one project with as many threads and cores as you have available. I see in the task manager for BOINC that it's downloaded and working on 16 projects at once (which is right as I have 8 physical / 16 HT cores). What I'm wondering though is that much like how I can put -SMP 15 in the F@H config to allow the GPU to work on one project as well, is there a noticable benefit to have Rosetta "tone down" on how many cores or threads it runs on? Should I just leave all 16 cores cranking along? I noticed after extensive F@H testing that the calculations were working a bit faster if I kept one thread free for the GPU work. If I maxed the CPU, it'd actually have the GPU slow down the CPU and both sets of projects would take longer. Is Rosetta like this or different?

Is there an "ideal" way to run Rosetta? I still want to use the GPU for folding though if possible.

2. Any particular settings that'd help Rosetta crunch faster? I'm allowing it plenty of disk space and enough RAM for all the projects it's downloaded. Is simply OC'ing enough or would tighter RAM timings help (currently at auto which I think is something like 11-11-11-36), though the RAM is fine at 8-8-8-24 or 6-8-6-24. Are there any other optimizations I can do or just let it work?

Thanks for the help.

There are soooo many tweaks and settings to play with that the only real answer is to try them and see what works best on your pc. Yes if you leave one cpu open for the gpu to use, the gpu will run better but the trade-off is that you reduce your cpu crunching. The key is to find the sweet spot for both, I have personally done it both ways and found that if the gpu uses a low amount of cpu while it is crunching, then I get more credits when I crunch with the cpu too. Of course each project uses a different amount of the cpu to feed the gpu, so you will have to see what it is with your projects. Collatz is only using 0.01% of the cpu, while PrimeGrid can use up to 0.08% or more and DNETC is using 0.05% of the cpu on most of my gpu's. I do have a gpu in a single core machine and right now the cpu is not being used for crunching, however since the gpu is only using 0.08% of the cpu that might change in the next few days.

Biggest impact in Rosetta performance (after CPU performance) is L2 cache size. Not RAM speed.

AFAIK, you can crunch with a CPU core AND let that same core manage a GPU at the same time (as long as the GPU managing has a higher priority)... at least for Collatz Conjecture (GPU-enabled project). When it's using 100% a GPU, it only uses 1% of CPU. So I crunch with both cores while I crunch with the GPU (Same with my laptop ;) ). Now, of course, you can limit how many cores BOINC will use to crunch. Just go preferences and set the number of desired cores to crunch at any given time. In your case, put 15, if you want to leave a single virtual thread free of any duty.

As far as performance goes for Rosetta: Just have enough RAM (at LEAST 512MB per core), and a strong CPU. Rosetta is all about CPU brute-force.

Welcome to Rosetta!.

Edit: It seems BOINC asks for a percentage, rather than number when it asks for "how many cores to use"... Try to crunch with 100% of cores crunching and see how it goes... It doesn't give me a problem when crunching Collatz.

---

Right if you run R@h on all cores, the GPU may be impacted because GPU processing DOES require a small amount of CPU. If an idle CPU is waiting around, the GPU can count on use of it instantly, if not, it waits in line in priority order (R@h runs at low priority). However, running R@h on 15 cores rather then 16 is reducing your productivity on the R@h side. So, as long as you have enough memory to keep them all happy, your overall greatest throughput should be running all 16 cores and GPU just sneaks in and briefly interrupts when it needs to (same as R@h is interrupted when I open a new page on my browser here and it is crunching in the background).

---

Rosetta Moderator: Mod.Sense