Martin.P it's surprizing that you have a 20% speed boost whereas other crunchers doesn't; is it same size of WU for each ?
I really think there is significant speedup on G4 while almost none on G5, probably due to the second FPU the G5 has.
Martin.P, you're running a G4, right?
BM
Bernd,
I run a G5 Dual 2.7 GHz and a G5 Dual 2.5 GHz. Both are single-core. I migrated the 2.5 only yesterday while the 2.7 is running the new version for 9 days already. Unfortunately the validated WUs disappear very quickly so I cannot directly compare the WUs from the 2.7. However, I did notice that the difference between the 2.5 and 2.7 is linear with the frequency, and I did see a significant speed-up when I ran the first WUs with 4.29 (at that time there were a few older WUs in my stats-page).
Check out the 2 machines: PowerMac G5 Dual 2.7 GHz and PowerMac G5 Dual 2.5 GHz. In all BOINC projects the 2.7 is appr. 8% faster than the 2.5. However, in E@H running the different clients it is more than 20% faster. I expect the first results with 4.29 on the 2.5 machine sometime this night, so let's wait and see.
O.K., here is the first result with 4.29 om my G5 Dual 2.5 GHz:
92473053 37133435 19 Feb 2008 0:46:40 UTC 8 Mar 2008 0:46:40 UTC In Progress Unknown New --- --- ---
92446592 36834594 18 Feb 2008 9:42:10 UTC 20 Feb 2008 8:01:19 UTC Over Success Done 45,985.37 237.40 237.40
92407266 37104288 18 Feb 2008 9:44:17 UTC 7 Mar 2008 9:44:17 UTC In Progress Unknown New --- --- ---
92219782 37027279 11 Feb 2008 19:21:33 UTC 18 Feb 2008 11:14:16 UTC Over Success Done 55,919.25 236.71 236.71
92176083 37015893 6 Feb 2008 6:04:48 UTC 16 Feb 2008 0:07:49 UTC Over Success Done 56,724.16 236.71 236.71
As you can see it takes 45,985 seconds versus 55,919 and 56,724 with 4.03. Quite a gain.
Bernd; it's surprizing that a vectorized app has no gain while the S5R2 vectorized app was 60% faster.
Yes, it is. I could understand that on a G5 which has two FPUs the overhead of the vector code makes up for the 4x speedup in that (small but important) part of the code, but on G4 I still have no clue.
Note that the checkpoints of 4.03 and 4.29 are incompatible. Therefore the app_info.xml is created so that tasks that were assigned to 4.03 are finished with the actual 4.03 App, not the 4.29 running under the 4.03 label. But you did take this into account, didn't you?
This is odd. Are there still some extremely short WUs out there?
Because my G4 just finished a WU, I think it's the 4th with the 4.29 beta in about 200 000 sec = 30 000 sec less then the previous fastest WU ever.
It's the first 800 after a couple of 700 ones so this may explain the jump. Will monitor the results over the next few weeks, maybe there is a increase of speed just in a certain frequency range? My iBook which still is in the 700 range has made no speed increase yet.
After changing the system preference processor performance setting from automatic to highest my G5 has significantly improved. It appears to have cut nearly in half the run time. The benchmark also noticeably improved.
With the setting on automatic I had 2 tasks with 4.29
h1_0797.50_S5R2__184_S5R3a finishied in 119086
h1_0887.60_S5R3__475_S5R3b finished in 139635
Then I changed from automatic to highest setting (still with 4.29)
h1_0887.60_S5R3__452_S5R3b finished in 65655.8
I will see if this drop in times is repeated on the next task or just a "short" WU fluke in a day or two.
Any other G5 imac users who have noticed this big of a difference between automatic and highest settings on the processor performance? With the highest setting my G5 times are at least around Martin's and [AF...]'s. I just looked at my G5 results from seti@home and the speed up is apparent there as well (so perhaps the "short" WU theory does not apply).
This is odd. Are there still some extremely short WUs out there?
Because my G4 just finished a WU, I think it's the 4th with the 4.29 beta in about 200 000 sec = 30 000 sec less then the previous fastest WU ever.
It's the first 800 after a couple of 700 ones so this may explain the jump. Will monitor the results over the next few weeks, maybe there is a increase of speed just in a certain frequency range? My iBook which still is in the 700 range has made no speed increase yet.
The runtime variation is pretty much within the normal fluctuation of runtime. You can check this by using the "ReadyReckoner" Web Application that Mike, Gary, archae86 and others have put together, see this thread.
Ca 200k seconds should be the minimum runtime expected for your host.
After changing the system preference processor performance setting from automatic to highest my G5 has significantly improved. It appears to have cut nearly in half the run time. The benchmark also noticeably improved.
With the setting on automatic I had 2 tasks with 4.29
h1_0797.50_S5R2__184_S5R3a finishied in 119086
h1_0887.60_S5R3__475_S5R3b finished in 139635
Then I changed from automatic to highest setting (still with 4.29)
h1_0887.60_S5R3__452_S5R3b finished in 65655.8
I will see if this drop in times is repeated on the next task or just a "short" WU fluke in a day or two.
Any other G5 imac users who have noticed this big of a difference between automatic and highest settings on the processor performance? With the highest setting my G5 times are at least around Martin's and [AF...]'s. I just looked at my G5 results from seti@home and the speed up is apparent there as well (so perhaps the "short" WU theory does not apply).
Peanut,
the "Maximum" settings provides appr. 30% more speed to my G5s over the "Automatic" setting. You can also check it out with Hardware Monitor (CPU heat and power usage). This has always been like that.
RE: RE: RE: Martin.P
)
O.K., here is the first result with 4.29 om my G5 Dual 2.5 GHz:
92473053 37133435 19 Feb 2008 0:46:40 UTC 8 Mar 2008 0:46:40 UTC In Progress Unknown New --- --- ---
92446592 36834594 18 Feb 2008 9:42:10 UTC 20 Feb 2008 8:01:19 UTC Over Success Done 45,985.37 237.40 237.40
92407266 37104288 18 Feb 2008 9:44:17 UTC 7 Mar 2008 9:44:17 UTC In Progress Unknown New --- --- ---
92219782 37027279 11 Feb 2008 19:21:33 UTC 18 Feb 2008 11:14:16 UTC Over Success Done 55,919.25 236.71 236.71
92176083 37015893 6 Feb 2008 6:04:48 UTC 16 Feb 2008 0:07:49 UTC Over Success Done 56,724.16 236.71 236.71
As you can see it takes 45,985 seconds versus 55,919 and 56,724 with 4.03. Quite a gain.
Bernd; it's surprizing that a
)
Bernd; it's surprizing that a vectorized app has no gain while the S5R2 vectorized app was 60% faster.
is it the good app in download ?
Some problems with autovectorization ?
RE: Bernd; it's surprizing
)
Yes, it is. I could understand that on a G5 which has two FPUs the overhead of the vector code makes up for the 4x speedup in that (small but important) part of the code, but on G4 I still have no clue.
Note that the checkpoints of 4.03 and 4.29 are incompatible. Therefore the app_info.xml is created so that tasks that were assigned to 4.03 are finished with the actual 4.03 App, not the 4.29 running under the 4.03 label. But you did take this into account, didn't you?
BM
BM
Yes the WU under 4.03 crashed
)
Yes the WU under 4.03 crashed when i put the 4.29 in place
:)
But it's curious no gain while all other CPU have gain with SSE code;
This is odd. Are there still
)
This is odd. Are there still some extremely short WUs out there?
Because my G4 just finished a WU, I think it's the 4th with the 4.29 beta in about 200 000 sec = 30 000 sec less then the previous fastest WU ever.
It's the first 800 after a couple of 700 ones so this may explain the jump. Will monitor the results over the next few weeks, maybe there is a increase of speed just in a certain frequency range? My iBook which still is in the 700 range has made no speed increase yet.
After changing the system
)
After changing the system preference processor performance setting from automatic to highest my G5 has significantly improved. It appears to have cut nearly in half the run time. The benchmark also noticeably improved.
With the setting on automatic I had 2 tasks with 4.29
h1_0797.50_S5R2__184_S5R3a finishied in 119086
h1_0887.60_S5R3__475_S5R3b finished in 139635
Then I changed from automatic to highest setting (still with 4.29)
h1_0887.60_S5R3__452_S5R3b finished in 65655.8
I will see if this drop in times is repeated on the next task or just a "short" WU fluke in a day or two.
Any other G5 imac users who have noticed this big of a difference between automatic and highest settings on the processor performance? With the highest setting my G5 times are at least around Martin's and [AF...]'s. I just looked at my G5 results from seti@home and the speed up is apparent there as well (so perhaps the "short" WU theory does not apply).
Compared to "automatic
)
Compared to "automatic performance" setting the "maximum performance" setting provided about 3% to 7% gain (most of the time)
i don't test this for einstein (i'm always in "maximum perdormance")
RE: This is odd. Are there
)
The runtime variation is pretty much within the normal fluctuation of runtime. You can check this by using the "ReadyReckoner" Web Application that Mike, Gary, archae86 and others have put together, see this thread.
Ca 200k seconds should be the minimum runtime expected for your host.
CU
Bikeman
RE: After changing the
)
Peanut,
the "Maximum" settings provides appr. 30% more speed to my G5s over the "Automatic" setting. You can also check it out with Hardware Monitor (CPU heat and power usage). This has always been like that.
I will definitely leave my G5
)
I will definitely leave my G5 on the max setting from now on. I didn't think it made that much of a difference. I guess I was wrong about that.