Hello
My first time picking up a few dual processors for pennies - two boards and 4 processors for less than a Ryzen 3 for a total of 48 threads, and probably a huge jump in my electric bill. Still, I'd rather they get some use, if only for a few cold months of the year.
A few questions:
I'll be running linux - Ubuntu 1804.
I have 32 GB of DDR3 ram for each. Will this be enough or will I find bottlenecks in either the ram or processors themselves with Rosetta taking all cores/threads?
I'm assuming target runtime just means using a little more or less bandwidth and has nothing to do with daily credit/points, or is there optimal targets that give the best results?
How does Boinc handle 2 processors? Can I run, for instance, CPDN at 50% (16 concurrent on the 2670) is boinc smart enough to run it on the true cores rather than hyperthreading with Rosetta taking up the rest?
I also have the chance to buy dual Opteron 6128 but that just seems like a very underclocked Phenom II, even with 16 threads, with much higher power consumption to boot.
I know these are years old at this point, but 48 threads is 48 threads, and I'm excited to throw these together.
It's no 3rd generation Ryzen 7 or Ryzen 9, but it will be mine, at a fraction of the up front cost at least.
Sorry for the questions. I'm just not sure if I'm getting in over my head.

Hello,
Hope I undestood the question, but 32GB of Ram should be enough for 32 rosetta instances. They never eat over 1GB/instance, but sometimes it can be close. Most projects needs a lot less ram than rosetta.
For runtime I think the 8 hour default is ok for most cases. If you run only rosetta 24/7 and never got calculation errors, then you can increase runtime.
For threads it depends more of operating system, cpu and bios etc than boinc, how the cpu is used. For 50% CPU settings Boinc calls for 16 hardware threads(on 32thread system) and has no control over what threads it gets.
But it most likely runs on "real cores" in that case.

Hello,
Hope I undestood the question, but 32GB of Ram should be enough for 32 rosetta instances. They never eat over 1GB/instance, but sometimes it can be close. Most projects needs a lot less ram than rosetta.
For runtime I think the 8 hour default is ok for most cases. If you run only rosetta 24/7 and never got calculation errors, then you can increase runtime.
For threads it depends more of operating system, cpu and bios etc than boinc, how the cpu is used. For 50% CPU settings Boinc calls for 16 hardware threads(on 32thread system) and has no control over what threads it gets.
But it most likely runs on "real cores" in that case.

Hi,
Yes, you understood correctly.
I guess the ultimate question is will I get more credits running 3 8 hour tasks a day on each core or run 1 24 hour task on each core?
Or will it be much of a difference?

I general, the length of time a WU runs is proportional to the amount of credit granted. The exception would be when a specific model wonders around and does not zero in on a good solution, and the coding to detect such cases does not notice it is not progressing to a better solution. In those cases, the "watchdog" kicks in if a WU goes more than 4 hours past the runtime preference, and ends the task. Some people shorten their preferred runtime to reduce the odds that more than 4 hours of work are lost, and to speed the return of their completed results back to the project.

Personally, I prefer the smaller number of WUs and lower bandwidth requirements of the 24 hours runtimes.

---

Rosetta Moderator: Mod.Sense

I general, the length of time a WU runs is proportional to the amount of credit granted. The exception would be when a specific model wonders around and does not zero in on a good solution, and the coding to detect such cases does not notice it is not progressing to a better solution. In those cases, the "watchdog" kicks in if a WU goes more than 4 hours past the runtime preference, and ends the task. Some people shorten their preferred runtime to reduce the odds that more than 4 hours of work are lost, and to speed the return of their completed results back to the project.

Personally, I prefer the smaller number of WUs and lower bandwidth requirements of the 24 hours runtimes.

Does this happen quite often or is it rare?
I'll likely just experiment to see what works best I suppose.
In other news, it may be 2 2670's instead - so 64 threads total.

And as it turns out, I apparently have a rare processor on Rosetta anyway. Just got 2 Xeon 2670v2's (I thought they were original 2670s)... so these are Ivy Bridge with 10 cores and 20 threads each. 2 motherboards with 4 cpus will make for fun processing here.
I can't find these listed on the list of processors here, just the original Sandy Bridge 2670, which, at least in theory, crunches some very nice results, even compared to my Ryzen 1800x.
I was hearing things about l3 cache, though - should I limit the number of workunits based on that, or is that a contributing factor these days?

Not certain if you mean level 2 cache, or Intel i3 processors. But basically, on-board memory cache will improve performance of memory intensive applications, and Rosetta is a memory intensive application. Since the cache is generally shared across multiple cores, it can get saturated if all of the cores are demanding high use of it. So, I believe the answer to your question is yes, it is possible that you get better throughput by running a few less concurrent R@h work units. You might achieve this any number of ways. One would be to simply add a BOINC project that has a lower memory requirement to your mix of work, and adjust the resource shares to control the typical mix of work you will be running.

---

Rosetta Moderator: Mod.Sense