my AMD 6000+ machine with Kubuntu 7.04 Linux has taken a significant performance hit with the switch to the 'R3 run.
So Linux R3 is bad also on AMD. I had a theory that constant memory access was holding back performance on the new app but then it should have affected AMDs less than Intels. OTOH, could still be the case, AMDs smaller and less efficient caches could be adding to the already frequent memory access.
For R2 4.37 i saw a significant reduction in crunching time when going from memory settings 5-5-5-15 to 4-4-4-12, the app is accessing RAM almost constantly. If the difference was just a few hundred seconds i wouldnt have said anything, but look at these R2 results done with different memory settings: http://einsteinathome.org/host/1016454/tasks
So R3s problem in Linux could be no cache hits and constant accessing ram? Same as in R2, just worse.
Yes, this is wahat I'm thinking as well. The code generated by the Microsoft Compiler is arranged a bit cleverer to avoide stalls from having to wait until data is read from RAM in case of a cache miss.
I've noticed on a Q6600 WinXP host that running 3 4.07 S5R3 and 1 4.40 S5R2 I see much more variation in power draw than I did running 4 4.40 S5R2 last week.
The measurement is by a power meter with a once per second update rate and a resolution of 1 watt. Formerly, most sequential seconds would have the same value, and until a SETI task swapped in, would lie in a very narrow band of a couple of watts near 250 (total box power). SETI was both higher and more variable than Einstein.
Now I see changes in most sequential seconds of two or more watts, with a range from 250 to 262 seen in a an observation period of about a minute.
So something has changed quite a bit. One possibility is that the memory access activity has become more clumped on a time scale of a second, in addition to the previous poster's suggestion that overall memory activity may have increased.
As I said, it would also explain why my system seems to do so badly for a Windows box, so it seems very plausible to me. As soon as my laptop gets S5R3 I'll be able to get some more data on this (the laptop has DDR-2 memory so it should be affected less badly when I'm running Windows).
Annika,
DDR1 vs DDR2 on Athlon64 shouldnt make much difference i think, DDR1 got lower latencies and the higher frequencies of DDR2 doesnt do much when moving small amounts of data as is the case with E@H.
The link in my previous post is dead after merging hosts into this one, basically it went like this (R2):
th3, thanks for pointing that out. Don't think it makes a difference? Well, OTOH I'm glad that using DDR-1 doesn't hurt my puter's performance too much (I'd have to get a new mainboard and everything if I wanted to switch so I'm delaying it until I need a serious upgrade anyway). But- why then do I have performance problems like this?
I've noticed on a Q6600 WinXP host that running 3 4.07 S5R3 and 1 4.40 S5R2 I see much more variation in power draw than I did running 4 4.40 S5R2 last week.
So THAT'S why my CPU temperature fluctuates rapidly between 50C and 60C!
Here are some results from my
)
Here are some results from my Pentium 4 with Gentoo Linux. The first two are S2R3, the others are old S2R2.
87138571 34802744 25 Sep 2007 19:40:16 UTC 28 Sep 2007 16:44:19 UTC Over Success Done 141,101.55 218.91 pending
87100911 34789277 23 Sep 2007 11:47:18 UTC 27 Sep 2007 21:13:48 UTC Over Success Done 141,531.52 218.91 218.90
87070399 34760882 19 Sep 2007 23:05:26 UTC 25 Sep 2007 16:52:56 UTC Over Success Done 290,186.35 632.69 632.69
87043636 34772920 21 Sep 2007 10:31:19 UTC 27 Sep 2007 5:44:04 UTC Over Success Done 291,815.01 657.66 pending
87002376 34752315 18 Sep 2007 8:30:24 UTC 23 Sep 2007 11:47:16 UTC Over Success Done 293,408.67 632.33 632.33
RE: my AMD 6000+ machine
)
So Linux R3 is bad also on AMD. I had a theory that constant memory access was holding back performance on the new app but then it should have affected AMDs less than Intels. OTOH, could still be the case, AMDs smaller and less efficient caches could be adding to the already frequent memory access.
For R2 4.37 i saw a significant reduction in crunching time when going from memory settings 5-5-5-15 to 4-4-4-12, the app is accessing RAM almost constantly. If the difference was just a few hundred seconds i wouldnt have said anything, but look at these R2 results done with different memory settings:
http://einsteinathome.org/host/1016454/tasks
So R3s problem in Linux could be no cache hits and constant accessing ram? Same as in R2, just worse.
Team Philippines
That would certainly explain
)
That would certainly explain the reason my box sucks so badly, since it still uses DDR-1-memory.
Hi! Yes, this is wahat I'm
)
Hi!
Yes, this is wahat I'm thinking as well. The code generated by the Microsoft Compiler is arranged a bit cleverer to avoide stalls from having to wait until data is read from RAM in case of a cache miss.
CU
H-BE
I've noticed on a Q6600 WinXP
)
I've noticed on a Q6600 WinXP host that running 3 4.07 S5R3 and 1 4.40 S5R2 I see much more variation in power draw than I did running 4 4.40 S5R2 last week.
The measurement is by a power meter with a once per second update rate and a resolution of 1 watt. Formerly, most sequential seconds would have the same value, and until a SETI task swapped in, would lie in a very narrow band of a couple of watts near 250 (total box power). SETI was both higher and more variable than Einstein.
Now I see changes in most sequential seconds of two or more watts, with a range from 250 to 262 seen in a an observation period of about a minute.
So something has changed quite a bit. One possibility is that the memory access activity has become more clumped on a time scale of a second, in addition to the previous poster's suggestion that overall memory activity may have increased.
As I said, it would also
)
As I said, it would also explain why my system seems to do so badly for a Windows box, so it seems very plausible to me. As soon as my laptop gets S5R3 I'll be able to get some more data on this (the laptop has DDR-2 memory so it should be affected less badly when I'm running Windows).
Annika, DDR1 vs DDR2 on
)
Annika,
DDR1 vs DDR2 on Athlon64 shouldnt make much difference i think, DDR1 got lower latencies and the higher frequencies of DDR2 doesnt do much when moving small amounts of data as is the case with E@H.
The link in my previous post is dead after merging hosts into this one, basically it went like this (R2):
CL5:
73,791.62 - 658.15
73,780.67 - 658.15
74,575.04 - 658.16
74,429.39 - 658.16
CL4:
70,861.47 - 658.18
70,890.76 - 658.15
Team Philippines
Sorry, wrong thread.
)
Sorry, wrong thread.
th3, thanks for pointing that
)
th3, thanks for pointing that out. Don't think it makes a difference? Well, OTOH I'm glad that using DDR-1 doesn't hurt my puter's performance too much (I'd have to get a new mainboard and everything if I wanted to switch so I'm delaying it until I need a serious upgrade anyway). But- why then do I have performance problems like this?
RE: I've noticed on a Q6600
)
So THAT'S why my CPU temperature fluctuates rapidly between 50C and 60C!