Message boards :
Number crunching :
The Attack of the Killer 58.7s
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
zombie67 [MM] Send message Joined: 22 Apr 04 Posts: 758 Credit: 27,771,894 RAC: 0 |
just checked, chipset is a 5000X, configured with 4x1 GB FB-DIMMs (Memory set at 5/5/5/15/20 @ 667MHz) and 2 5150 Xeons. If there are more than 4 DIMM slots, I think it matters which slots are populated to get the full bandwidth. Dublin, California Team: SETI.USA |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
...(and I'll have to pester a Dell rep). In order to get the interlaced dual-channel memory working right, the right sockets must be used, and of course, that depends on the manufacturer of the motherboard. My Supermicro X7DAE needs to have sockets 1,3,5 and 7 populated in order to get the interlaced dual-channel (any other way, other than all 8 sockets and it will revert back to regular dual-channel or even straight access). If I can get my copy of SiSoft working again, I'll see what my board reports. As for the total bandwidth efficiency, that's simply calculated by seeing how close the RAM ratio is running to the FSB. If you notice, even dual-channel DDR400 (which can technically fill the 8.4GB/s bandwidth of P4s and PDs) is even labeled as only having an efficiency of 55 to 65%, because it's running at 200MHz double pumped while the FSB of most P4s and PDs is 200MHz quad pumped. That's one of the problem with synthetic benchmarks such as SiSoft Sandra. |
Sisyfos Send message Joined: 22 Jul 00 Posts: 7 Credit: 2,796,632 RAC: 0 |
I was fiddling with a large cache (10 days), just at the time these units were send out. I think I got 200 of them on my C2D :-( Anyways, just for kicks I tried Chicken's 2.0 Generic SSE2... My C2D crunch times went from ~5400s to ~4200s http://setiathome.berkeley.edu/result.php?resultid=434671531 http://setiathome.berkeley.edu/result.php?resultid=434827238 I have a lot of 58.7s to cover yet, so I don't know how other types of WU would fare. Current setup E6600@3456(9x384), DDR2@960 5-5-5-15. |
Pooh Bear 27 Send message Joined: 14 Jul 03 Posts: 3224 Credit: 4,603,826 RAC: 0 |
All I say is for those aborting them, someone has to crunch them, so why not take the bad with the good? Aborting just slows down the science. Just an observation. It probably will not make a difference, but someone had to say it. I'm in it for the science, and will not abort the units. So, I get less credit per hour on a few units. It's how it goes. My movie https://vimeo.com/manage/videos/502242 |
KWSN - Chicken of Angnor Send message Joined: 9 Jul 99 Posts: 1199 Credit: 6,615,780 RAC: 0 |
...(and I'll have to pester a Dell rep). Good point! CPU-Z tells me my modules are in slots 1-4, which seems stupid (unless they do some funky numbering), may be holding performance back quite a bit. Would be nice to get a basically free performance upgrade, anyway. Regards, Simon. Donate to SETI@Home via PayPal! Optimized SETI@Home apps + Information |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
...(and I'll have to pester a Dell rep). My Dell (workstation motherboard) is also a 5000X, with the same memory timings, but so far only 2 x 1GB DIMMS in slots 1 and 2 (as supplied by Dell). I've just downloaded and run SiSoft Sandra Lite 11.17b: it reports Int 3362, Float 3373, 39% efficiency :-( - I hope the DIMMs I've ordered aren't stuck in the Christmas mail for too much longer. One advantage of Dell systems is that you can use the system tag (serial number) to get quite detailled technical information which is supposedly right for that exact machine. For mine, they're saying I should install DIMMs in pairs, starting with the lowest numbered slots - so (1&2), (3&4), (5&6), (7&8) in that order. But if your Dell rep says anything different, please let us know. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
All I say is for those aborting them, someone has to crunch them, so why not take the bad with the good? Aborting just slows down the science. I fully agree, and will do likewise. FWIW, I haven't seen any sign of any units being aborted. I think what happens is that on some machines (like msattler's), they take much longer than expected, which bumps up the RDCF. So that machine's cache registers as being much fuller than it really is. Which inhibits work fetch for a while. So nothing gets reported. So everyone else's pending credit goes up, and RAC goes down. Does that make sense to anyone? |
KWSN - Chicken of Angnor Send message Joined: 9 Jul 99 Posts: 1199 Credit: 6,615,780 RAC: 0 |
Enabling Quad-Channel operation on the Dell may also be a BIOS option - I'll have to check it out in more detail when I have physical access to that box again. Regards, Simon. Donate to SETI@Home via PayPal! Optimized SETI@Home apps + Information |
Sisyfos Send message Joined: 22 Jul 00 Posts: 7 Credit: 2,796,632 RAC: 0 |
I was fiddling with a large cache (10 days), just at the time these units were send out. Hmmm...I thought the Core 2 people would be excited about this, but I just noticed that I forgot to mention, that the ~20% decrease in 58.7 credit WUs crunch time was going from Chicken SSSE3 1.41 to Chicken SSE2 2.0 Generic |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Hmmm...I thought the Core 2 people would be excited about this, but I just noticed that I forgot to mention, that the ~20% decrease in 58.7 credit WUs crunch time was going from Chicken SSSE3 1.41 to Chicken SSE2 2.0 Generic Well, there's a challenge for Simon and the coop [I like that formulation - pun on chicken, or you could read it as co-op(erative), which is a good description] Could you/they write a little app which would examine the headers of downloaded WUs, and dynamically choose which optimised app to crunch them with???!!! That should keep you out of mischief for an hour or two, LOL |
Sisyfos Send message Joined: 22 Jul 00 Posts: 7 Credit: 2,796,632 RAC: 0 |
Could you/they write a little app which would examine the headers of downloaded WUs, and dynamically choose which optimised app to crunch them with???!!! That should keep you out of mischief for an hour or two, LOL Not a bad idea I have started 2 58.7s with SSE3 Intel 2.0, just to see if there's more to be gained. I certainly have enough of those pesky WUs to try all the different apps for effectiveness (is that a word?). |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
(is that a word?). Yes, the word "that" is a word. <giggle> <chuckle><going back into my corner now> |
ML1 Send message Joined: 25 Nov 01 Posts: 20334 Credit: 7,508,002 RAC: 20 |
(is that a word?). Are you going over there for a byte or a nibble also? Happy crunchin', ;-) Martin ( ps: Sorry for inflicting multiple groans :-/ ) See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Hmmm...I thought the Core 2 people would be excited about this, but I just noticed that I forgot to mention, that the ~20% decrease in 58.7 credit WUs crunch time was going from Chicken SSSE3 1.41 to Chicken SSE2 2.0 Generic I have suggested this at least a couple of times in Simon's forums. Add both types of code to the app, and branch to the one that would be most effective for the WU to be processed. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Clyde C. Phillips, III Send message Joined: 2 Aug 00 Posts: 1851 Credit: 5,955,047 RAC: 0 |
I think he's talking about lots of high-noise WUs that are taking 58.7 seconds. MSattler, I took a look at one of your results and it looks like you may not be using the latest version of Simon's cruncher. I saw "V1.41" which may be the previous edition. I used that version, too, and those 58-credit units crunched slowly. If you update your cruncher you might (I repeat MIGHT) get slightly better results. In my case the later version does better on the most plentiful units but does worse on others. It handles some VLARS very well but does more poorly on the 0.7-to-1.12 and the ones above three degrees. Overall my PD950s do slightly better (maybe just a couple percent) with Version 2.0. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
I was fiddling with a large cache (10 days), just at the time these units were send out. Thanx for the idea!!! I am in the same boat as you are. I had a RAM crash on my rig a while back, errored out a bunch of WUs, and downloaded a bunch of these nastys which are now in my cache waiting to be processeed like little land mines. I'm gonna switch to the SSE2 app for a while, see if it helps, and maybe stick with it until I get the little buggers cleared out. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
I think he's talking about lots of high-noise WUs that are taking 58.7 seconds. I just switched to the SSE2 app for a while (see other posts in this thread), but Simon has always maintained that the 1.41 was the best to use only on core 2 rigs. It obviously has one Achilles' heel. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Sisyfos Send message Joined: 22 Jul 00 Posts: 7 Credit: 2,796,632 RAC: 0 |
...Anyways, just for kicks I tried Chicken's 2.0 Generic SSE2... You're welcome :-) These just in... SSE3 Intel 2.0 App http://setiathome.berkeley.edu/result.php?resultid=434827413 4221s http://setiathome.berkeley.edu/result.php?resultid=434827417 4179s Not much of a difference from SSE2 Generic 2.0, but still a 20% improvement from SSSE3 1.41. Off to try SSE2 Intel 2.0... |
Sisyfos Send message Joined: 22 Jul 00 Posts: 7 Credit: 2,796,632 RAC: 0 |
SSE2 Intel 2.0 App http://setiathome.berkeley.edu/result.php?resultid=434827370 4130s http://setiathome.berkeley.edu/result.php?resultid=434827429 4188s Now trying SSE2 Intel-PM 2.0 App |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Well, I think it's more than just the lack of interleaved dual-channel memory, as even AnandTech was having low throughput with the FB-DIMMs, and the problem could be attributed to many things, including (but not limited to): available BIOS settings (differs between manufacturers), immature MCH (this is Intel's first foray into FB-DIMM whereas they've had some time to work with DDR2), immature chipset drivers, etc. A colleague of mine that was reading my post pointed out a different perspective to me, one that I thought I'd share. He pointed out to me that even if it's a BIOS setting/limitation or an immature MCH/driver, it still doesn't change the fact that the Core 2 Duos/Quads are performing better with their DDR2 RAM as of right now, which would technically make the Core 2 Duos/Quads a better choice cruncher over the equivalently clocked Xeons as of right now, and until Intel can work on whatever is preventing them from using the FB-DIMMs to their fullest potential, it's going to hold the performance back. I still hold to my theory that something is preventing the Xeons from performing at their fullest potential, but I must concede to the point that as it stands, it appears the Core 2 chips crunch better overall. Also, as a follow up, I can't seem to get SiSoft to run on my new system. I have a legal full version of Sandra 2005 SR3 that I can install, but as soon as I launch the app, it simply exits without an error message. I'll have to see if I can upgrade to the latest version (I'll have to check with my accountant [girlfriend] on that matter) and see if that fixes the issue. I'm thinking that Sandra doesn't like 32bit Windows running with 4GB RAM or something. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.