PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : 321 Prime Search : 321 Sieve is being SUSPENDED (Message 144496)
Posted 10 days ago by Profile Rafael
So what the next default subproject is planned?

PPS-SV
2) Message boards : Number crunching : WW testing (Message 144441)
Posted 11 days ago by Profile Rafael
Tested the GPU app on a Gtx 950 with 456.55. Numbers match, but there's a bit of screen lag; a B option similar to GFNSV would be appreciated if possible.

1 - 67029FBFBC1AEFB0
2 - 2F32CAD23BA5B1CA
3 - B8B040FA3813FEB2

I've also ran it on an Intel HD630 (8400 iGPU) with 26.20.100.7642 for the sake of validation. And I'm happy to report that it's working as intended.

1 - 67029FBFBC1AEFB0
2 - 2F32CAD23BA5B1CA with 5257
3 - 2276306935816523 is a Wieferich special instance (-1 -3 p)
B8B040FA3813FEB2
3) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 144410)
Posted 11 days ago by Profile Rafael
Thank you very much for taking the time to reply guys. I do appreciate it.

Raf -
I haven't had that much experience driving Prime 95. If I've done it right, a quick test on my 3950 came up with this -

Timings for 400K FFT length (16 cores, 1 worker): 0.42 ms. Throughput: 2367.28 iter/sec.
Timings for 400K FFT length (16 cores, 4 workers): 0.39, 0.38, 0.38, 0.38 ms. Throughput: 10525.06 iter/sec.
Timings for 400K FFT length (16 cores, 16 workers): 1.39, 1.40, 1.38, 1.41, 1.41, 1.41, 1.40, 1.39, 1.42, 1.41, 1.45, 1.44, 1.40, 1.39, 1.37, 1.44 ms. Throughput: 11374.65 iter/sec.


So 1 task for each of the 16 cores is slightly better than 4*4.... is that correct?

Yes, that's what it says. However, I suggest you try benching 16 cores with 8 workers, aka you'd be running 2 cores per task; it might give out higher performance. Or it might not. Who knows.
4) Message boards : Fermat Divisor Search : Fermat Divisor Search (Message 144405)
Posted 11 days ago by Profile Rafael
Can't prime95's benchmark be used to determine how many threads to run?

Yes it can.
So why spend time hypothesizing about FFT size and number of cores instead of just running the benchmark? :)

Because it's good to understand why things work and how they are supposed to behave. For instance, we praise CPUs with FMA3 support because it greatly boosts performance, yet some older AMD CPUs have it and still perform the same whether you use it or not - why is that? Same with AVX512, some xeons do it well, whereas others don't. You could just run the benchmark, true; but some of us like to tune our machines to the limit and/or build machines tailored for Primegrid, and trying to understand the combination of hardware and software is very helpful, and at the very least fun.
5) Message boards : Number crunching : Sub-Project Credit (Message 144392)
Posted 11 days ago by Profile Rafael
Hello PrimeGrid!
Since 321 SV is shutting down, I have been binging it pretty hard for the last 3 weeks.
During that, I noticed that it gave me a lot of credit/task, like 600, and that it ran fast on CPU as it
was designed for it.

Are there any almost-as-efficient sub-projects out there?
Thanks in advance!

Actually, 321 SV isn't particularly great performing. For something like your i7 4790, which has support for FMA3, the LLR projects can make full use of that instruction set to greatly boost performance when compared to older-than-haswell machines.

As to which one, there's a plethora of options. You can hunt for badges like most of us do, or chose a specific one that you find nice and deserving of your computation power; it's really up to you.
6) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 144373)
Posted 12 days ago by Profile Rafael
So after reading all these posts, I'm still a little confused about how many tasks to run.

Stream said "First of all, the multi-threading thing must be used anyway. It will improve overall efficiency of your PC because internal CPU cache will be utilized better. You will finish more tasks per day with multi-threading." but then others have said it may be possible to run say 1 task/core with a 3950X (I guess that depends on the size of the FFT?)

Right now, I have no idea for DIV how many tasks I should run simultaneously on my -

3950X
1950X
3930K
4930K
(and I'm not worried about my 4c processors, I'll just run 1 task across 4 cores)

I've currently chosen 4*4 on the 3950x & 1950X, 2*3 on the 6 core processors. Would that be optimal?


There is a simple way to find out: test it yourself! It's really not hard to just run a couple benchmarks for your systems to figure out what to do.

For instance, here's my Ryzen 3600:

Timings for 400K all-complex FFT length (6 cores, 1 worker): 0.34 ms. Throughput: 2945.72 iter/sec. Timings for 400K all-complex FFT length (6 cores, 2 workers): 0.47, 0.46 ms. Throughput: 4299.22 iter/sec. Timings for 400K all-complex FFT length (6 cores, 3 workers): 1.28, 0.68, 0.47 ms. Throughput: 4396.30 iter/sec. Timings for 400K all-complex FFT length (6 cores, 6 workers): 1.29, 1.28, 1.27, 1.28, 1.26, 1.26 ms. Throughput: 4713.76 iter/sec. Timings for 400K all-complex FFT length (6 cores hyperthreaded, 1 worker): 0.41 ms. Throughput: 2436.01 iter/sec. Timings for 400K all-complex FFT length (6 cores hyperthreaded, 2 workers): 0.44, 0.44 ms. Throughput: 4527.77 iter/sec. Timings for 400K all-complex FFT length (6 cores hyperthreaded, 3 workers): 1.25, 0.62, 0.44 ms. Throughput: 4697.96 iter/sec. Timings for 400K all-complex FFT length (6 cores hyperthreaded, 6 workers): 1.35, 1.22, 1.21, 1.24, 1.24, 1.20 ms. Throughput: 4825.86 iter/sec.

As you can see, apparently it's better to run 6 -t2 tasks, it's marginally faster than going with 6 -t1; with that said, given this is my daily driver, I'm willing to take the penalty in the name of leaving a couple resources free for my other stuff.

And next is my i5 8400, which has no HT, but much smaller cache: looks like running 3 -t2 is the best bet:
Timings for 480K all-complex FFT length (6 cores, 1 worker): 0.80 ms. Throughput: 1255.14 iter/sec. Timings for 480K all-complex FFT length (6 cores, 2 workers): 0.83, 0.72 ms. Throughput: 2584.94 iter/sec. Timings for 480K all-complex FFT length (6 cores, 3 workers): 1.12, 1.04, 1.03 ms. Throughput: 2821.25 iter/sec. Timings for 480K all-complex FFT length (6 cores, 6 workers): 3.27, 2.66, 2.59, 2.79, 2.67, 2.59 ms. Throughput: 2186.32 iter/sec.
7) Message boards : Fermat Divisor Search : Sweep stake % complete by end of challenge (Message 144300)
Posted 13 days ago by Profile Rafael
52%, 1 prime
8) Message boards : Number crunching : Badges III (Message 143754)
Posted 32 days ago by Profile Rafael


According to my calculations, I should be able to get to Jade within 2weeks. Let's see if turns out fine.
9) Message boards : Number crunching : How good is i7-10750H? (Message 142713)
Posted 65 days ago by Profile Rafael
How good is Intel's i7-10750H? This processor type doesn't show on the "Fastest CPUs" page. I need a new laptop for my studies and the web stores offer this processor above all else.

It's a laptop, which is another way of saying that cooling will be the problem, not the CPU itself. In theory, it's a low clock, 6c FMA3 capable chip, so it'll be decent at any of the CPU projects, cooling notwithstanding.
10) Message boards : Problems and Help : Task has been pending Validation for multiple weeks (Message 142507)
Posted 75 days ago by Profile Rafael
As said, sometimes you get "unlucky" and it takes you a while for the second person to return a valid result. Don't feel bad, though, you'll eventually get it, even if you've been waiting since January like this person:

https://www.primegrid.com/workunit.php?wuid=638404410


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.89, 2.96, 3.17
Generated 31 Oct 2020 | 12:40:34 UTC