I've got 4.32 results from two Conroe-class hosts with enough difference in phase to permit a decent estimate of both the peak and the variance parameters, thus allowing an overall performance improvement estimate.
For my Q6600 3.006 GHz Quad, running 794.65 frequency, I see 18934 CPU seconds at sequence 93, and 24201 at sequence 129. This gives an estimate of 24384 peak time and .286 variance, for an overall estimated runtime average of 19939.
On 4.26, my peak and variance estimates were 31,000 and .21. Those correspond to a runtime average of 26856. The 4.32 productivity is thus estimated at 35% higher than on 4.26
For my E6600 3.006 GHz Duo, running 792.15 frequency, I see 17931 CPU seconds at sequence 90, and 22447.73 at sequence 0. This gives an estimate of 22448 peak time and .286 variance, for an over all estimated runtime average of 18923.
On 4.26 for the Duo, my peak and variance estimates were 29453 and 0.200. Those correspond to a runtime average of 25703. The 4.32 productivity is thus estimated at 36% higher than 4.26
Conclusions: For Conroe-class hosts, 4.32 is probably between 30 and 40 % more productive than 4.26, and likely pretty near the middle of that range. As my variance estimate has increased, it appears that there was proportionally more improvement in the code which executes a varying amount between peak and valley than in the rest of the code, while both clearly did improve.
Thus people relying on single-point estimates based on a result near the trough will somewhat over-estimate the overall improvement, while those relying on results near a peak will somewhat under-estimate the overall improvement.
Not a big problem but I thought I should report it anyway.
I switched to 4.32 from 4.26 in the middle of a task and after switching the screen saver would not work. I clicked on the show graphics button but nothing happened, checked task manager and the graphics app did not run. The task did run and completed as it should. Now that task is completed and a new has started and everything is as it should, the screen saver is now working as expected.
I rarely use the screen saver but I thought I should test it as this is a new app.
By the way, my system is a P4 2.4 Ghz running Windows XP Home, SP2, 1 Gb ram and a Nvidia GeForce FX 5200 video card.
Not a big problem but I thought I should report it anyway.
I switched to 4.32 from 4.26 in the middle of a task and after switching the screen saver would not work. I clicked on the show graphics button but nothing happened, checked task manager and the graphics app did not run. The task did run and completed as it should. Now that task is completed and a new has started and everything is as it should, the screen saver is now working as expected.
I rarely use the screen saver but I thought I should test it as this is a new app.
By the way, my system is a P4 2.4 Ghz running Windows XP Home, SP2, 1 Gb ram and a Nvidia GeForce FX 5200 video card.
I don't have a mixed result to check this against now, so I guess the question I have is did you reboot between when you saw that it didn't work and when it resumed working properly?
Thus people relying on single-point estimates based on a result near the trough will somewhat over-estimate the overall improvement, while those relying on results near a peak will somewhat under-estimate the overall improvement.
There was a reason I said "at least" when I gave my 25-30%... ;-) I was cautious with the estimate. Turns out I'm processing downwards toward a trough, but still much closer to the peak than the trough, if the RRV5A is accurate for tasks > 800...
I don't have a mixed result to check this against now, so I guess the question I have is did you reboot between when you saw that it didn't work and when it resumed working properly?
No I did not reboot, rearly do. I switched apps and checked that it was working, except for the screen saver that is, and then went to bed. The next day the task finished and Boinc downloaded a new task. I tried the screen saver again with the new task and everything seemed to work as it should.
Could it have something to do with the task being run with 4.32 and Boinc trying to run the grapichs part form 4.26???
On both my 2.4 GHz P-IV machine, and my 3500+ machine, there's been a dramatic speed-up, even though they're now both crunching on the "b" workunits. (In fact, the 2.2 GHz 3500+ with this Windows app isn't that much slower than the 3.0 GHz 6000+ with the Linux power-users app.)
i can't get new client to work. i follow installation instruction as described here: http://einstein.phys.uwm.edu/app_test.php, but boinc start again 4.26 instead of 4.32. what i do wrong? (boinc istalled as a service, btw)
i can't get new client to work. i follow installation instruction as described here: http://einstein.phys.uwm.edu/app_test.php, but boinc start again 4.26 instead of 4.32. what i do wrong? (boinc istalled as a service, btw)
It's probably okay. If you've had uncompleted workunits at the time that you installed 4.32, BOINC will show that it's still running 4.26, but you actually are running 4.32.
I've got 4.32 results from
)
I've got 4.32 results from two Conroe-class hosts with enough difference in phase to permit a decent estimate of both the peak and the variance parameters, thus allowing an overall performance improvement estimate.
For my Q6600 3.006 GHz Quad, running 794.65 frequency, I see 18934 CPU seconds at sequence 93, and 24201 at sequence 129. This gives an estimate of 24384 peak time and .286 variance, for an overall estimated runtime average of 19939.
On 4.26, my peak and variance estimates were 31,000 and .21. Those correspond to a runtime average of 26856. The 4.32 productivity is thus estimated at 35% higher than on 4.26
For my E6600 3.006 GHz Duo, running 792.15 frequency, I see 17931 CPU seconds at sequence 90, and 22447.73 at sequence 0. This gives an estimate of 22448 peak time and .286 variance, for an over all estimated runtime average of 18923.
On 4.26 for the Duo, my peak and variance estimates were 29453 and 0.200. Those correspond to a runtime average of 25703. The 4.32 productivity is thus estimated at 36% higher than 4.26
Conclusions: For Conroe-class hosts, 4.32 is probably between 30 and 40 % more productive than 4.26, and likely pretty near the middle of that range. As my variance estimate has increased, it appears that there was proportionally more improvement in the code which executes a varying amount between peak and valley than in the rest of the code, while both clearly did improve.
Thus people relying on single-point estimates based on a result near the trough will somewhat over-estimate the overall improvement, while those relying on results near a peak will somewhat under-estimate the overall improvement.
Not a big problem but I
)
Not a big problem but I thought I should report it anyway.
I switched to 4.32 from 4.26 in the middle of a task and after switching the screen saver would not work. I clicked on the show graphics button but nothing happened, checked task manager and the graphics app did not run. The task did run and completed as it should. Now that task is completed and a new has started and everything is as it should, the screen saver is now working as expected.
I rarely use the screen saver but I thought I should test it as this is a new app.
By the way, my system is a P4 2.4 Ghz running Windows XP Home, SP2, 1 Gb ram and a Nvidia GeForce FX 5200 video card.
RE: Not a big problem but I
)
I don't have a mixed result to check this against now, so I guess the question I have is did you reboot between when you saw that it didn't work and when it resumed working properly?
RE: Thus people relying on
)
There was a reason I said "at least" when I gave my 25-30%... ;-) I was cautious with the estimate. Turns out I'm processing downwards toward a trough, but still much closer to the peak than the trough, if the RRV5A is accurate for tasks > 800...
RE: I don't have a mixed
)
No I did not reboot, rearly do. I switched apps and checked that it was working, except for the screen saver that is, and then went to bed. The next day the task finished and Boinc downloaded a new task. I tried the screen saver again with the new task and everything seemed to work as it should.
Could it have something to do with the task being run with 4.32 and Boinc trying to run the grapichs part form 4.26???
Just finished this result
)
Just finished this result with v4.32: http://einsteinathome.org/task/92241712
This is my only other result still in the DB from the same host and it was done with v4.26: http://einsteinathome.org/task/92143465
Both claimed the 237.67; but, the v4.32 result only took 47,846.40 seconds vs. 64,308.22 seconds for v4.26.
This must be doing me some
)
This must be doing me some good.
On both my 2.4 GHz P-IV machine, and my 3500+ machine, there's been a dramatic speed-up, even though they're now both crunching on the "b" workunits. (In fact, the 2.2 GHz 3500+ with this Windows app isn't that much slower than the 3.0 GHz 6000+ with the Linux power-users app.)
Another one finished on the
)
Another one finished on the same host: http://einsteinathome.org/task/92348954 - again, claimed credit of 237.67 (and it validated), this one only took 41,556.08 seconds.
i can't get new client to
)
i can't get new client to work. i follow installation instruction as described here: http://einstein.phys.uwm.edu/app_test.php, but boinc start again 4.26 instead of 4.32. what i do wrong? (boinc istalled as a service, btw)
RE: i can't get new client
)
It's probably okay. If you've had uncompleted workunits at the time that you installed 4.32, BOINC will show that it's still running 4.26, but you actually are running 4.32.