Rally Nov. 21st-23rd

[mdettweiler]

Originally Posted by em99010pepe

Only at 2.96 GHz, can't go much higher. You can ask Flatlander about his quad timings.

Hmm, I see. Now if only I could overclock my machine (that is, my dualcore)...alas, I have an Intel-branded motherboard, and Intel motherboards are notorious for not being overclockable. Which begs the question: are you by chance familiar with any software-based overclocking programs for Linux? I found a number of such programs for Windows with Google, but for Linux everywhere I looked just said that there isn't much of any software overclock programs for Linux, and to overclock from the BIOS instead--but that's no help for me since my motherboard doesn't have any overclocking settings!

Max

[gd_barnes]

Originally Posted by mdettweiler

Hey, I just thought of an idea about G4000: since right now we're kind of in the dark as to how much load it can safely handle, why don't we run a pre-rally load test on it? Maybe we could all try to put as many cores as we can on G4000 for, say, 12 hours, and at least get a minimum figure as to how much we can put on it before it breaks.

For those who are interested, here's the specs for the server G4000's on:
-AMD Phenom 9750 quad-core CPU
-2 GB RAM
-running Ubuntu 7.04 Desktop 32-bit (yeah, it's old, that's a long story. We'll upgrade it some time or other, especially now that Ubuntu stopped supporting 7.04. )
-lighttpd web server for the status page stuff

The LLRnet server has a jobMaxTime of 5 days, and a prunePeriod of 15 minutes.

Max

Well, I never! See if I let you put servers on MY machines any more! If you don't like 32-bit 7.04, just move the server over to one of my Intel's running 64-bit 8.04. lol

Seriously, shall I attempt to upgrade the AMD's also; at least the one running the server? Obviously we'll need to coordinate that with any stress test on the server.


Gary

[gd_barnes]

Originally Posted by mdettweiler

Okay, I've marked you down in the table; IB400 should be good.

Gary, we've got 61 cores on IB400 at this point. At which point do you think we should start telling users to begin putting their cores on G4000 or C443 instead? We've been able to support upwards of 100 cores on IB5000, so presumably IB400 can handle a similar load; however, we probably want to make sure that, especially during a rally, we stay somewhat under that limit, especially considering as how we have plenty of servers anyway.

We had ~120 cores on port 5000 for the final ~2 days of the 3rd drive with no problems. Port 400 should be able to handle the same.

David, is there any reason to think that port 400 should not be able to handle the same load as port 5000?

I would say this: If ONLY our main (mostly) full-time searchers here run the rally, port 400 will work. (I'm including Buster in that.) We could even throw Bruce in the mix and we'd still be < 100 cores on that port since Carlos will likely run his port 443. (I think Bruce has ~20-28 cores.) BUT...if Lennart or Beyond run the rally, we should probably set up port 5000 just for them...or Lennart can run his own server like he's done in the past.

In other words, if a heavy hitter comes, let's definitely set up another server. If not, ports 400 and 443 should be sufficient; at least at this point anyway.


Gary

[Flatlander]

Originally Posted by em99010pepe

Only at 2.96 GHz, can't go much higher. You can ask Flatlander about his quad timings.

Edit: 321-325 secs between all cores with llr 3.7.1c.

I'm getting 335 secs at 3.30Ghz on LLRNet
(I'm hoping your maths is out and you mean 6mins 21-25secs?)
Or maybe it's because the client is using the old LLR?

I'll try to bring six cores to the rally. The C2Quad at 3.3GHz and a C2Duo at 2.4GHz.

[gd_barnes]

Originally Posted by Flatlander

I'm getting 335 secs at 3.30Ghz on LLRNet
(I'm hoping your maths is out and you mean 6mins 21-25secs?)
Or maybe it's because the client is using the old LLR?

I'll try to bring six cores to the rally. The C2Quad at 3.3GHz and a C2Duo at 2.4GHz.

I think Carlos caches his LLRnet pairs, runs thems manually, and then processes them back to the server. Even over-clocked, it's hard to imagine 321 secs. at 2.96 Ghz. for n=~540K-550K on LLRnet.

Not over-clocked on my 2.4 Ghz Intels, I'm getting 420 secs. on LLRnet at n=~545K.


Gary

[IronBits]

All my llrnet servers and ports run off the same quad core Q6600 with CentOS 64bit and 12gb of ram and twin 300gb velociraptor hdds. :)
I have very high bandwidth and a smoothwall router to negotiate, all on Gbit nics.
Bring it on

[mdettweiler]

Originally Posted by gd_barnes

Well, I never! See if I let you put servers on MY machines any more! If you don't like 32-bit 7.04, just move the server over to one of my Intel's running 64-bit 8.04. lol

Seriously, shall I attempt to upgrade the AMD's also; at least the one running the server? Obviously we'll need to coordinate that with any stress test on the server.


Gary

LOL--nah, actually it's no big deal. I didn't mean to imply that 32-bit 7.04 was inadequate; I was only saying that since I was sure that, otherwise, somebody was going to post something like "what the heck? you're still using 7.04? uh, is that a typo?"

As for upgrading them: Go ahead and upgrade the AMD's when it's convenient, all except crunchford, which we may want to upgrade a little differently so that the server stuff doesn't get messed up. I'm thinking we may want to just do a "soft" upgrade to 8.04 32-bit on that one, instead of a clean install of 64-bit 8.04, so we don't have to mess with setting up the server stuff all over again. Mind you, I'm not saying that we *couldn't* do it if we had to, just that it would probably be easier if we didn't, and besides, 9 machines should be plenty in case you ever need to do some massive 64-bit sieving work.

[em99010pepe]

Originally Posted by Flatlander

I'm getting 335 secs at 3.30Ghz on LLRNet
(I'm hoping your maths is out and you mean 6mins 21-25secs?)
Or maybe it's because the client is using the old LLR?

I'll try to bring six cores to the rally. The C2Quad at 3.3GHz and a C2Duo at 2.4GHz.

llrnet is 10 % slower than llr3.7.1, if you used the manual client you would get 301.5 sec per candidate . The math is correct.

683*2^543866-1 is not prime. LLR Res64: 9765CFAB4E4DBD34 Time : 322.017 sec.
755*2^543866-1 is not prime. LLR Res64: D2DE375020EA2E61 Time : 322.196 sec.
785*2^543866-1 is not prime. LLR Res64: CBDFCBFBF4E36076 Time : 321.877 sec.
843*2^543866-1 is not prime. LLR Res64: EFDD6FD18B542F65 Time : 322.509 sec.
935*2^543908-1 is not prime. LLR Res64: 6936EC34653F2D0A Time : 324.475 sec.
439*2^543909-1 is not prime. LLR Res64: 466899EA25FC9E78 Time : 324.035 sec.
465*2^543909-1 is not prime. LLR Res64: AD22B217A3BED7F7 Time : 324.833 sec.

Originally Posted by gd_barnes

I think Carlos caches his LLRnet pairs, runs thems manually, and then processes them back to the server.

Correct.

[em99010pepe]

Just saw this on XS forum, an Intel Core I7 965 overclocked (air cooling, 4 cores, 8 threads) to 4.4 GHz is 3 times faster than my Q6600 on wprime benchmark!!!!
I have to get one of those.

Carlos

[mdettweiler]

Originally Posted by em99010pepe

Just saw this on XS forum, an Intel Core I7 965 overclocked (air cooling, 4 cores, 8 threads) to 4.4 GHz is 3 times faster than my Q6600 on wprime benchmark!!!!
I have to get one of those.

Carlos

HOLY COW! 4.4Ghz? Rarely do I ever hear of a CPU being overclocked *that* far.

One thing, though: what the heck is wprime? Do you mean Prime95?

[mdettweiler]

Originally Posted by mdettweiler

One thing, though: what the heck is wprime? Do you mean Prime95?

Never mind, I found the answer on Google.