PrimeGrid Challenges 2021

[Endgame124]

You guys keep posting interesting contest threads and I’m going to have to get up at 3am again to get another 5950 - only 13 more days till I can buy another from microcenter.

 

[StefanR5R]

Since @Ken g6' stats posting,

• @Icecold is still 17th, but I am no longer 21st,
• 7AA7 lost its nice prime-numbered rank of 7, and occupies the boringly even-numbered rank 6,
• we made 270 kPPH on average during the last 12 hours, while Ultimate Chaos made 210 kPPH. Extrapolated to the remaining 37 hours from when I looked this up: 25.1 M for us, 25.6 M for them.

 

[StefanR5R]

I added a 22-core Xeon now. (It began like this: "No keyboard detected! Press F1 to Run SETUP".) It may make 4.8 kPPH perhaps, that's 0.17 M in the remaining 36 hours.

Edit, also added three 4-core computers for maybe 0.14 M combined for the remaining hours.

 

[TennesseeTony]

Well, since it is going to be such a close finish, I have added the proverbial kitchen sink.

 

[Skivelitis2]

Run 'em down boys....

 

[Endgame124]

So it looks like a 5900 might be better than a 5950 for primegrid just from skimming this thread. If I wanted to play around with random projects, better to get another 5950 or a 5900?

 

[Icecold]

I have every FMA3 capable CPU I own running on it at this point(as far as I know). Does anybody know if the app is optimized at all for AVX? (not avx2). I'm pretty sure that was a thing back in the day but I don't see mention of it on their website. I'm just wondering if booting up an old Sandy Bridge i5 is worth it.

Edit- this was a silly question from me, please ignore it. Of course it's worth it, even if the PPD is not great. An i5 3470s(Ivy Bridge, I had it mixed up for a second, but still no FMA3) and an i3-4030u (has FMA3 but low cache/clock speed/core count) have been added to the cause.

So it looks like a 5900 might be better than a 5950 for primegrid just from skimming this thread. If I wanted to play around with random projects, better to get another 5950 or a 5900?

If price isn't a big factor I'd go with another 5950x You can always run less cores/threads on the 5950x, but you can't make more cores/threads that don't exist on a 5900x. If price is a big factor, I'd consider a 3900x. I see them going used on this forum and others occasionally for around $350 and they're still a very capable processor.

 

[StefanR5R]

So it looks like a 5900 might be better than a 5950 for primegrid just from skimming this thread.

No, 5950X is definitely better than 5900X. They have the same (default) PPT, same amount of cache, and same amount of RAM bandwidth. Yet the 5950X has got more execution units. Therefore, the 5950X wins. (In absolute performance, and in performance/Watt at default PPT. But perhaps not in performance/investment, depending on how it is accounted.)

 

[Skivelitis2]

I have every FMA3 capable CPU I own running on it at this point(as far as I know). Does anybody know if the app is optimized at all for AVX? (not avx2). I'm pretty sure that was a thing back in the day but I don't see mention of it on their website. I'm just wondering if booting up an old Sandy Bridge i5 is worth it.

Edit- this was a silly question from me, please ignore it. Of course it's worth it, even if the PPD is not great. An i5 3470s(Ivy Bridge, I had it mixed up for a second, but still no FMA3) and an i3-4030u (has FMA3 but low cache/clock speed/core count) have been added to the cause.


If price isn't a big factor I'd go with another 5950x You can always run less cores/threads on the 5950x, but you can't make more cores/threads that don't exist on a 5900x. If price is a big factor, I'd consider a 3900x. I see them going used on this forum and others occasionally for around $350 and they're still a very capable processor.

I'm seeing just under 2hrs per task on my i5-3470 running all four cores per task. Only 6mb L3 cache.

 

[Icecold]

I'm seeing just under 2hrs per task on my i5-3470 running all four cores per task. Only 6mb L3 cache.

That seems really good(and better than I would have expected), hopefully mine is similar. I have it running all four cores as well. I'm seeing about an hour and 35 minutes per task on an FMA3 capable i5-4570 also running all 4 cores, so maybe it does use AVX, and then the jump from AVX to AVX2/FMA3 is less substantial than the initial performance jump to AVX. I'm probably not going to setup any of the old pre Sandy Bridge hardware I have here to test that, though.

 

[Icecold]

2 Hours, 5 minutes on the first task completed by my i5-3470s. No complaints here on that task time and nice to see an Ivy Bridge i5 still doing fairly well on a project.. I would have started it up sooner had I thought of it(especially knowing it performs pretty decently). It's running nice and cool too with the 120MM CPU cooler I have on it(a generic-ish ripoff of a Hyper 212, don't remember what brand)

 

[Skivelitis2]

I also have an i5- 4570 running and our run times are almost identical (as they should be I suppose). All of you guys have some incredible hardware...must be nice. My latest processor is an i7-3770 and best GPU a GTX 1650 super! (although my i3-8100 is the fastest on a per core basis I think)
Edit: I'm sure your numbers regarding the 3470 are more accurate. I only looked the estimated time for a cached work unit and those estimates are likely skewed by a few double check tasks I may have had.

 

[Ken g6]

All of you guys have some incredible hardware...must be nice.

Not all of us. I have an i7-6700 I bought used from the For Sale/For Trade forum here, a Xeon I bought from @TennesseeTony, an i3 in a tiny HTPC, and a Sandy Bridge laptop I just added to this Challenge. The rest of my hardware I'm renting from Digital Ocean. (They gave me some free credits I'm burning through.)

 

[VirtualLarry]

I didn't realize that the race was so close. I've thrown in my third 6C/12T Ryzen, it's only a 1600 "AE" version, though, and a couple of PrimeGrid races ago, it was showing errors in CPU WUs, so that's why I hesitated to throw it into the race. Let me know if I should disable it. (I did disable the micro-OPs cache or whatever the "fix" was, but that didn't seem to completely fix the issue. Maybe the problem is RAM-related?)

 

[Skivelitis2]

I need to look into these various cloud computing services eventually to see what kind of free deals I can get. Thought of doing it for this years Tour de Primes but doubt I'll get around to it by then and I'd rather use them for a TeAm event anyway.. Next PG CPU challenge series event will do nicely. Strike that...Pentathlon! Gives me time...

 

[Icecold]

Edit: I'm sure your numbers regarding the 3470 are more accurate. I only looked the estimated time for a cached work unit and those estimates are likely skewed by a few double check tasks I may have had.

Mine is the S model with the slightly lower TDP, so it may also not be using turbo boost as much as your's. It seems to be running at 3.2GHz pretty much at all times currently.

 

[Skivelitis2]

I have an i7-2400S and never knew what that meant and was too lazy to look it up. Thanks for the info.

 

[VirtualLarry]

I have an i7-2400S

That would indeed be a rare beast.

(Hint: The 2400 was an i5.)

 

[Icecold]

It's been mentioned before, but there definitely is a lot of cloud on these challenges. I checked the person right ahead of me, since it's super close, and their computers weren't hidden(I was curious what hardware I was up against) - http://www.primegrid.com/hosts_user.php?userid=1209161

It doesn't bother me, and I could rent some cloud as well if I wanted to so it's not like it's an unfair advantage, but I am curious where the rankings would line up across the board with no cloud involved.

I guess I don't know 100% that those are cloud instances, but the Xeon golds with 3 cores per(plus the other machines listed) definitely make me think that.

 

[Skivelitis2]

That would indeed be a rare beast.

(Hint: The 2400 was an i5.)

Oooops yet another typo from sir two-fingers. Should have said 17-2600S. Nice catch!

 

[lane42]

I guess I don't know 100% that those are cloud instances,

Same with this Guy, same Team
Computers belonging to Sean (primegrid.com)

 

[Skivelitis2]

Just saw a post at PrimeGrid that work units with 512k FFT size have started to appear. Not sure how this affects cache size etc. No time to investigate, got an emergency service call to deal with.

 

[StefanR5R]

Just saw a post at PrimeGrid that work units with 512k FFT size have started to appear.

Now that you say so, I 'grep'ed through stderr.txt of the currently running work and saw a few of those.
grep 'FFT length' /var/lib/boinc/slots/*/*

As long as we are dealing with a mix of 480K and 512K workunits, the settings which were optimized for 480K-only should still be good. When 512K workunits become the majority, such settings may get a little tight if set at 4 MB cache per task. But it will be a gradual thing at worst.

My own production is mostly on Xeons with large unified cache, task count configured such that there are 5 MB cache per task, and on Zen2 Epycs which I configured for merely 1 task per CCX (not the bandwidth optimum for the tested 480K-only workload, but very close; it's their efficiency optimum). So in short, there is no impact for me from these larger WUs.

 

[biodoc]

I switched my 3700X's from 8 tasks, one thread each to 4 tasks, 4 threads each.

 

[lane42]

to 4 tasks, 4 threads each.

Same here. Doing about 2 hours, give or take a couple min.
on a 3800x on windows.