![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Tour de prime 2024
- Thread starter Holdolin
- Start date
I yoinked a mega yesterday, but after 50k PPSE's not a one. This little 7950X3D is a nice way to hammer the hell outta PPSE tasks.Holdolin How goes the prime hunting? I'm continuing to have PPSE primes drip in every day or so, but still nothing at all from the GPUs
Ooh. Still haven't found a mega. But did get my first GFN-16 at least. GPUs aren't completely worthless apparently
'grats on the mega!
Nice nice, Gratz on your GFN-16!! I find it quite satisfying to find a prime.Ooh. Still haven't found a mega. But did get my first GFN-16 at least. GPUs aren't completely worthless apparently
'grats on the mega!
Thanks
Unfortunately I appear to have kicked the ethernet cable out of one of my systems before I left town, so it's not going to be reporting anything until Sunday. Hopefully with the custom version of BOINC I built that increases the task limit it has enough work cached to keep it working for most of that, so it'll have a nice stack of WUs to report upon reconnection.
I hate when that happens. Here's to hopin for a nice huge cache to chew onThanks
Unfortunately I appear to have kicked the ethernet cable out of one of my systems before I left town, so it's not going to be reporting anything until Sunday. Hopefully with the custom version of BOINC I built that increases the task limit it has enough work cached to keep it working for most of that, so it'll have a nice stack of WUs to report upon reconnection.
wareyore
HDCOTY 2023
- Joined
- Jan 1, 2014
- Messages
- 1,890
Do tell. What else does this thing do?
Just that. By default BOINC limits you to 1000 WUs in progress. I've noticed with the small tasks, and a large CPU count + a couple of GPUs, it tends to end up in a state where it runs out of work for the GPUs, and won't fetch more because it hits the 1,000 task limit. So I cloned the GitHub repository, upped the limit to either 5k or 8k (don't remember) and then compiled an x86_64 Linux version. I built it when 7.23 was the latest release, so I called mine 7.23.1.
Finally pulled a mega!
Dear Primefinder,
Congratulations! Our records indicate that a computer registered by you has found a unique prime number. This computer is running BOINC, is attached to the PrimeGrid project, and is assigned to the Generalized Fermat Prime Search n=17 Mega (GFN-17-Mega). What makes this prime unique is that it's large enough to enter the Top 5000 List in The Largest Known Primes Database.
Since you have auto-reporting selected, the following prime was submitted on your behalf:
Added 137528 : 207753480^131072+1 (1090198 digits)
This prime was found on this workunit which will automatically show as a prime result after verification by the Largest Known Primes Database.
If you have any questions or concerns, please contact us and we will surely resolve any problems.
Once again, congratulations on your find! Thank you for participating in PrimeGrid.
PrimeGrid staff
Dear Primefinder,
Congratulations! Our records indicate that a computer registered by you has found a unique prime number. This computer is running BOINC, is attached to the PrimeGrid project, and is assigned to the Generalized Fermat Prime Search n=17 Mega (GFN-17-Mega). What makes this prime unique is that it's large enough to enter the Top 5000 List in The Largest Known Primes Database.
Since you have auto-reporting selected, the following prime was submitted on your behalf:
Added 137528 : 207753480^131072+1 (1090198 digits)
This prime was found on this workunit which will automatically show as a prime result after verification by the Largest Known Primes Database.
If you have any questions or concerns, please contact us and we will surely resolve any problems.
Once again, congratulations on your find! Thank you for participating in PrimeGrid.
PrimeGrid staff
Awesome sauce, Gratz!Finally pulled a mega!
Dear Primefinder,
Congratulations! Our records indicate that a computer registered by you has found a unique prime number. This computer is running BOINC, is attached to the PrimeGrid project, and is assigned to the Generalized Fermat Prime Search n=17 Mega (GFN-17-Mega). What makes this prime unique is that it's large enough to enter the Top 5000 List in The Largest Known Primes Database.
Since you have auto-reporting selected, the following prime was submitted on your behalf:
Added 137528 : 207753480^131072+1 (1090198 digits)
This prime was found on this workunit which will automatically show as a prime result after verification by the Largest Known Primes Database.
If you have any questions or concerns, please contact us and we will surely resolve any problems.
Once again, congratulations on your find! Thank you for participating in PrimeGrid.
PrimeGrid staff
wareyore
HDCOTY 2023
- Joined
- Jan 1, 2014
- Messages
- 1,890
I was thinking along the same lines. Got a handful of the PPSE and 16's early on. Shut all that down last week and let a couple gpu's chew on the 17's. Waiting for the 17 to be proven then may switch to the 18's until the 29th. Thinking I may go back to PPSE and the 16's on the Mountain stage. Just for the "why not" of it.
It's no mega but at least I finally found a prime.
Dear Primefinder,
Congratulations! Our records indicate that a computer registered by you has found a unique prime number. This computer is running BOINC, is attached to the PrimeGrid project, and is assigned to the Proth Prime Search (LLR2). What makes this prime unique is that it's large enough to enter the Top 5000 List in The Largest Known Primes Database.
Since you have auto-reporting selected, the following prime was submitted on your behalf:
Added 137533 : 7599*2^1906514+1 (573922 digits)
Found by my lowly Ryzen 2600x.
Dear Primefinder,
Congratulations! Our records indicate that a computer registered by you has found a unique prime number. This computer is running BOINC, is attached to the PrimeGrid project, and is assigned to the Proth Prime Search (LLR2). What makes this prime unique is that it's large enough to enter the Top 5000 List in The Largest Known Primes Database.
Since you have auto-reporting selected, the following prime was submitted on your behalf:
Added 137533 : 7599*2^1906514+1 (573922 digits)
Found by my lowly Ryzen 2600x.
Still a prime. Great find!!
Congrats -- got yourself an entry in the top 5k!
wareyore
HDCOTY 2023
- Joined
- Jan 1, 2014
- Messages
- 1,890
Just under a week left. There is still time to jump in a score a couple badges for this year.
Those crazy Italians are still grinding hard.
Those crazy Italians are still grinding hard.
wareyore
HDCOTY 2023
- Joined
- Jan 1, 2014
- Messages
- 1,890
Still a few days left in the month. Also, the Mountain Stage starts on the 29th zero dark hundred UTC.
I would love to get one during the mountain stage so that I can grab that badge. And on average, I'd get it...got more primes than days so far. But highly inconsistent. Plenty of days with 0, and then occasionally I'll hit 3 or 4 in one day. I might take some of my slower systems and suspend their network access tonight to buffer up a couple days of tasks on them to submit all in one go on Wednesday night. The faster ones chew through work so fast there's no benefit from buffering on them. At least not without increasing the WU cache to 40k or something and incurring Michael's wrath
I just released about 18k WUs I had bunkered across a few of my systems. Not that many all things considered, but if it gets me something in the final stage tomorrow, I'll take it. Wish I had a version of BOINC that would let me store 20k, and I could have done a lot better. But oh well.
wareyore
HDCOTY 2023
- Joined
- Jan 1, 2014
- Messages
- 1,890
The extra day this year made no real difference, but, will mess up Y/Y comparisons for the next three. Scoring linked below. gg's.
What's next?
What's next?
How many megas did you end up pulling off of all of those 4090s in the end?
5, and 2 little ones. I'm happy with that. Well, not really but it is what is is
Are/were you running PPSE or similar on the CPUs, or are you all-in on the GPUs? I pulled mostly PPSE primes, with only a few (3?) GFN-17s and a couple GFN-16s. Did a couple of the bigger GFNs too on the 4090 just for the heck of it, but nothing from there unsurprisingly.5, and 2 little ones. I'm happy with that. Well, not really but it is what is is
I was mostly doing the GFNs on the GPUs, but I tossed some PPSE runs in when I wasn't using my dev box to actually work on. Amazing how many PPSE tasks a modern CPU can chew through.
Yup! With SMT off, I'm seeing the 7950x chew through them in under 8 minutes each. Good stuff. 3x as fast as the Broadwell E5 Xeons I've got.
Got the same with my 7950x3d. 16 concurrent tasks getting done in under 8 minutes each. That's a lotta work getting done. Also tested it on my 7960x Threadripper and got the same results. This got me to thinking: gee, this is only a 24 core cpu. What about the 64 core which then got me thinking about something Michael said: thanks to all the sieve we have 10 years of PPS/E work to do. Hmm, wonder how fast I could drain the.....cache you sick-minded people.
It's making me think that it might be time to send some of the older Xeon equipment off to the great computer god in the sky. The 7950x does 50% more work than one of the dual-socket Xeon systems at approximately half of the total power draw. That really adds up over time.
Aye. It adds up quickly. I hate having to 'retire' hardware, but the build cost is a one-time thing. The electric bill is an on going every month thing. I realized how much my old Xeons are holding me back when I compared my Einstein crunchers. One is a Xeon 1697A v4 with dual Radeon Pro VIIs and one is a Threadripper 7960x with a single Pro VII (my 4th card made it's own journey to that great silicon wafer in the sky) and while my single GPU box isn't as fast as the dual card, it's much more than 50% as fast. It's not the GPUs themselves, but when the project hands off to the CPU to work at 50% and completion of the task. I've timed the Threadripper and it takes 80 seconds each time. The Xeon takes about 3 minutes. I've never actually timed it, but watching it work for an hour or so it's a good ballpark. I only actually counted on the TR system cause I wanted to see just how much faster a cutting edge CPU would be.
Have you experimented with running multiple tasks on the GPUs at once? I've done that for years going back to WCG, and again with the Sieve tasks and the smaller GFN tasks here. Particularly the tiny _c tasks run so fast that with just one at a time, the GPU is mostly spending its time idle while the BOINC schedule and CPU get caught up. So I was running anywhere from 2 to 4 of the GFN-16/GFN-17 tasks at once. Still, the larger tasks seem to do a better job on the more powerful cards.
Ya, for PG tasks it makes little difference. If i run 2 concurrent it takes twice as long. On Einstein, I found the sweet spot to be 4 concurrent tasks on the current cards. I have an Instinct Mi100 due in tomorrow I'll be experimenting on.