Intel ARC GPUs
log in

Advanced search

Message boards : Number crunching : Intel ARC GPUs

Previous · 1 · 2
Author Message
Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8573 - Posted: 27 Dec 2022, 8:04:40 UTC - in response to Message 8566.
Last modified: 27 Dec 2022, 8:05:11 UTC

I have given up on the Intel ARC 770 and replaced it with a Gigabyte GTX1660 Super 6GB card.




How many seconds does it take to complete a TF on average?
If you got that info before giving up on it.
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8574 - Posted: 27 Dec 2022, 8:19:35 UTC - in response to Message 8563.
Last modified: 27 Dec 2022, 8:55:01 UTC



It's at 8 hours and still crunching on the UHD. Most will be aborted for deadline.



It took 118,000 sec for that 1st WU.
The next one finished in 3680 sec though. EDIT: That 118k is a guess. I do not see 2 TF Valid WU's. The valid WU's usually purge after 24 hours so maybe the 3680sec reported IS the long WU and only the last hour of CPU time after a BOINC suspension is showing. See my question below...
The current is at 12 hours and still going.

My electric company has put me on a peak/off peak plan where it's 4 cents/KWh offpeak and 31 cents peak (6am-8am and 6pm- 8pm weekdays).

So I've had to pause all electronic devices, including BOINC, 6-8am and 6-8pm (which is it's own issue since BOINC doesn't support 2 pause periods per day; have to manually pause the GPU hosts daily until my lazy butt gets around to doing 2nd BOINC data folders).

1) Is the vast discrepancy in run times related to TF restarting from the beginning after a suspension period?
2) Are the WUs TF CPU times only showing the amount of time after a BOINC Time Of Day suspension? (As in 3600sec of an actual 34 hour run)
3) Should my NVidia 1060 run only 1 WU at a time instead of 4 if the suspensions are an issue?
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8575 - Posted: 27 Dec 2022, 11:14:20 UTC - in response to Message 8574.



It's at 8 hours and still crunching on the UHD. Most will be aborted for deadline.



It took 118,000 sec for that 1st WU.
The next one finished in 3680 sec though. EDIT: That 118k is a guess. I do not see 2 TF Valid WU's. The valid WU's usually purge after 24 hours so maybe the 3680sec reported IS the long WU and only the last hour of CPU time after a BOINC suspension is showing. See my question below...
The current is at 12 hours and still going.

My electric company has put me on a peak/off peak plan where it's 4 cents/KWh offpeak and 31 cents peak (6am-8am and 6pm- 8pm weekdays).

So I've had to pause all electronic devices, including BOINC, 6-8am and 6-8pm (which is it's own issue since BOINC doesn't support 2 pause periods per day; have to manually pause the GPU hosts daily until my lazy butt gets around to doing 2nd BOINC data folders).

1) Is the vast discrepancy in run times related to TF restarting from the beginning after a suspension period?
2) Are the WUs TF CPU times only showing the amount of time after a BOINC Time Of Day suspension? (As in 3600sec of an actual 34 hour run)
3) Should my NVidia 1060 run only 1 WU at a time instead of 4 if the suspensions are an issue?


1. there is a checkpointing every 120s then its minimal
2. CPU time is overall
3. yes

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8576 - Posted: 27 Dec 2022, 11:15:14 UTC - in response to Message 8573.

I have given up on the Intel ARC 770 and replaced it with a Gigabyte GTX1660 Super 6GB card.




How many seconds does it take to complete a TF on average?
If you got that info before giving up on it.


1h23min on my RTX500XT, in the current range a GTX1660 should take 25min

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8580 - Posted: 28 Dec 2022, 11:27:16 UTC - in response to Message 8575.


2) Are the WUs TF CPU times only showing the amount of time after a BOINC Time Of Day suspension? (As in 3600sec of an actual 34 hour run)
3) Should my NVidia 1060 run only 1 WU at a time instead of 4 if the suspensions are an issue?


(snip)
2. CPU time is overall
3. yes


2. So, how did the UHD on the 8250U complete a TF in 36xx secs (normal is 100k+)?
There is a checkpoint in the algorithm that can determine that there is no possible prime in the remaining data set and the WU ends prematurely? Or I found a prime on the UHD! (lol)

3. I'm compiling a data set of 4 TF and 1 TF at once. Will report back the results in couple days. Sorry, I have no ARC to test upon.

Currently the 1060 at 4x WU is putting out 4.45 credit/(CPU+GPU sec).
I control the GPU clock/power and the room temp for these tests.

(BTW, running the TF on the UHD is for WUProp hours.
At least it does beat the deadline and get credit.
It's only other project is Einstein, AFAIK.)
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8581 - Posted: 28 Dec 2022, 11:38:47 UTC - in response to Message 8580.
Last modified: 28 Dec 2022, 11:39:14 UTC


2) Are the WUs TF CPU times only showing the amount of time after a BOINC Time Of Day suspension? (As in 3600sec of an actual 34 hour run)
3) Should my NVidia 1060 run only 1 WU at a time instead of 4 if the suspensions are an issue?


(snip)
2. CPU time is overall
3. yes


2. So, how did the UHD on the 8250U complete a TF in 36xx secs (normal is 100k+)?
There is a checkpoint in the algorithm that can determine that there is no possible prime in the remaining data set and the WU ends prematurely? Or I found a prime on the UHD! (lol)

3. I'm compiling a data set of 4 TF and 1 TF at once. Will report back the results in couple days. Sorry, I have no ARC to test upon.

Currently the 1060 at 4x WU is putting out 4.45 credit/(CPU+GPU sec).
I control the GPU clock/power and the room temp for these tests.

(BTW, running the TF on the UHD is for WUProp hours.
At least it does beat the deadline and get credit.
It's only other project is Einstein, AFAIK.)


for point 2 this is a wrapper issue where the runtime after restart starting from 0, you cant find a prime, TF is only sieving

kotenok2000
Avatar
Send message
Joined: 4 Oct 20
Posts: 13
Credit: 33,595,257
RAC: 93
Message 8582 - Posted: 29 Dec 2022, 7:47:41 UTC - in response to Message 8581.
Last modified: 29 Dec 2022, 7:47:52 UTC

Maybe it was opposite? It found factor at 36xx seconds?

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8583 - Posted: 29 Dec 2022, 8:24:02 UTC - in response to Message 8582.
Last modified: 29 Dec 2022, 8:28:36 UTC

Maybe it was opposite? It found factor at 36xx seconds?


no, we have only one bitlvl per test. The BOINC wrapper is reporting the runtime, after a restart it is counting from zero.

kotenok2000
Avatar
Send message
Joined: 4 Oct 20
Posts: 13
Credit: 33,595,257
RAC: 93
Message 8584 - Posted: 29 Dec 2022, 12:03:53 UTC - in response to Message 8583.

I just thought it found factor and decided not to factor further and report at once.

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8628 - Posted: 7 Jan 2023, 16:20:37 UTC - in response to Message 8584.

I just thought it found factor and decided not to factor further and report at once.


Yeah, if this is the logic then you would find TF's that end much more quickly than others.

Maybe that explains my data set so far, unless these are shorter because I restarted the BOINC client. ThrottleStop won't adjust AMD clock frequencies so I've been rebooting to adjust TDP wattages on the CPU. Ryzen Master won't work on my severely stripped down Windows 10 OS.
We had drastic temperature shifts the last 2 weeks (-15C back to 12C).

92207740 88474001 23 Dec 2022, 4:05:08 UTC 28 Dec 2022, 5:58:22 UTC Completed and validated 11788.17 21.27 14000 TF v0.12 (cuda100) 92208081 88474342 23 Dec 2022, 2:46:03 UTC 28 Dec 2022, 5:53:03 UTC Completed and validated 11503.82 21.2 14000 TF v0.12 (cuda100) 92207948 88474209 23 Dec 2022, 2:32:35 UTC 28 Dec 2022, 5:26:11 UTC Completed and validated 9958.87 19.86 14000 TF v0.12 (cuda100) 92207038 88473299 22 Dec 2022, 22:59:13 UTC 28 Dec 2022, 5:28:12 UTC Completed and validated 10067.67 18.89 14000 TF v0.12 (cuda100) 92206586 88472847 22 Dec 2022, 21:39:10 UTC 27 Dec 2022, 22:08:58 UTC Completed and validated 15354.27 36.41 14000 TF v0.12 (cuda100) 92206262 88472523 22 Dec 2022, 20:19:05 UTC 27 Dec 2022, 22:04:50 UTC Completed and validated 15309.16 35.41 14000 TF v0.12 (cuda100) 92205654 88471915 22 Dec 2022, 18:58:21 UTC 27 Dec 2022, 21:59:16 UTC Completed and validated 15301.56 36.69 14000 TF v0.12 (cuda100) 92204936 88471197 22 Dec 2022, 17:39:26 UTC 27 Dec 2022, 21:37:44 UTC Completed and validated 15086.4 34.75 14000 TF v0.12 (cuda100) 92204702 88470963 22 Dec 2022, 16:18:33 UTC 27 Dec 2022, 16:42:44 UTC Completed and validated 7395.35 18.67 14000 TF v0.12 (cuda100) 92203730 88469991 22 Dec 2022, 14:59:05 UTC 27 Dec 2022, 16:38:33 UTC Completed and validated 7212.77 17.42 14000 TF v0.12 (cuda100) 92202885 88469146 22 Dec 2022, 11:37:02 UTC 27 Dec 2022, 16:32:23 UTC Completed and validated 6952.6 16.97 14000 TF v0.12 (cuda100) 92202790 88469051 22 Dec 2022, 10:17:17 UTC 27 Dec 2022, 16:12:36 UTC Completed and validated 6095.67 14.05 14000 TF v0.12 (cuda100) 92202335 88468596 22 Dec 2022, 8:58:02 UTC 27 Dec 2022, 9:13:32 UTC Completed and validated 17284.71 36.39 14000 TF v0.12 (cuda100) 92201866 88468127 22 Dec 2022, 7:38:42 UTC 27 Dec 2022, 9:06:25 UTC Completed and validated 17268.87 36.86 14000 TF v0.12 (cuda100) 92201549 88467810 22 Dec 2022, 6:20:40 UTC 27 Dec 2022, 9:02:18 UTC Completed and validated 17283.26 36.09 14000 TF v0.12 (cuda100) 92200970 88467231 22 Dec 2022, 5:02:27 UTC 27 Dec 2022, 8:42:04 UTC Completed and validated 17116.18 37.14 14000 TF v0.12 (cuda100) 93642926 89833820 6 Jan 2023, 7:31:31 UTC 7 Jan 2023, 14:05:03 UTC Completed and validated 15996.08 137.91 14000 TF v0.19 (cuda120) 93642629 89833523 6 Jan 2023, 7:31:03 UTC 7 Jan 2023, 13:00:32 UTC Completed and validated 15698.36 139.28 14000 TF v0.19 (cuda120) 93642584 89833478 6 Jan 2023, 7:30:33 UTC 7 Jan 2023, 12:48:38 UTC Completed and validated 15814.41 138.52 14000 TF v0.19 (cuda120) 93642649 89833543 6 Jan 2023, 7:30:03 UTC 7 Jan 2023, 12:26:14 UTC Completed and validated 15666.06 136.45 14000 TF v0.19 (cuda120)


These were all run on the GTX 1060, 4 at a time, same room temperature (within 3 f) and MSI Afterburner settings.
So either the shorter runs (6 to 7k secs) are the WU ending when it found a factor or I reset my BOINC client. The runs at 172xx seconds might have been when the CPU was running at a lower TDP/frequency. I'm not sure how much CPU performance effects the TF runs.
I'll start the data set from the 4 that completed today and not restart the BOINC client and hold the CPU at current TDP of 40 watts (if the weather doesn't get hot in January again...)
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8629 - Posted: 7 Jan 2023, 16:36:47 UTC - in response to Message 8628.

There is nearly no CPU time to run TF on GPU.

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8631 - Posted: 7 Jan 2023, 16:54:04 UTC - in response to Message 8629.

There is nearly no CPU time to run TF on GPU.


If the CPU is starved from 3300 Ghz to 1800 Ghz, I'd think it might have some 5 to 10% effect. I'll test that hypothesis some weeks from now.

Of course, it's possible the GPU downclocking from the room getting hotter, caused the 17k run times when 15k is normal today.

The main question above was:
Do TF's end early when they find a factor so we will see runtimes ending as low as a few seconds to maybe half the normal run time?
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 8641 - Posted: 8 Jan 2023, 9:40:23 UTC - in response to Message 8631.
Last modified: 8 Jan 2023, 9:44:02 UTC

There is nearly no CPU time to run TF on GPU.


If the CPU is starved from 3300 Ghz to 1800 Ghz, I'd think it might have some 5 to 10% effect. I'll test that hypothesis some weeks from now.

Of course, it's possible the GPU downclocking from the room getting hotter, caused the 17k run times when 15k is normal today.

The main question above was:
Do TF's end early when they find a factor so we will see runtimes ending as low as a few seconds to maybe half the normal run time?


I have removed my last post and checked a found factor result, you are right, the test ended earlier if a factor is found. Thats a bonus too in credits.

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 8642 - Posted: 10 Jan 2023, 13:35:09 UTC - in response to Message 8641.



The main question above was:
Do TF's end early when they find a factor so we will see runtimes ending as low as a few seconds to maybe half the normal run time?


I have removed my last post and checked a found factor result, you are right, the test ended earlier if a factor is found. Thats a bonus too in credits.


Maybe I should start a new thread about this since it's not ARC GPU.

The 3 WU at once is getting interesting results with several ending quite quickly:
89848291 7 Jan 2023, 14:49:44 UTC 9 Jan 2023, 17:01:16 UTC Completed and validated 641.52 1.27 14000 TF v0.21 (cuda120)

The typical appears to be:
89847929 7 Jan 2023, 14:57:16 UTC 10 Jan 2023, 11:44:51 UTC Completed and validated 10447.65 3.31 14000 TF v0.21 (cuda120)



The 4 WU at once had 10x longer CPU time when Parlea@Home was on the CPU's vs LODA WU's
Typical run (4 WU at once) Parlea@Home on CPU's:
89834340 6 Jan 2023, 8:43:36 UTC 8 Jan 2023, 13:56:58 UTC Completed and validated 15232.17 128.64 14000 TF v0.19 (cuda120)


Typical with LODA on CPU's:
89848299 7 Jan 2023, 14:45:49 UTC 9 Jan 2023, 2:41:13 UTC Completed and validated 15036.94 6.58 14000 TF v0.21 (cuda120)


I'm not sure what happened to this one; it's CPU usage is severe:
89833841 6 Jan 2023, 7:44:22 UTC 8 Jan 2023, 11:19:22 UTC Completed and validated 15971.15 2114.92 14000 TF v0.19 (cuda120)


Since the runs can end soon with early factor found... that means my data sets will need to be larger. ::sigh::
____________
My primes found at SRBase:
40*1017^215605+1 (Top 5000)
18922*111^383954+1 (Top 5000)
4281*880^27069+1

Previous · 1 · 2
Post to thread

Message boards : Number crunching : Intel ARC GPUs


Main page · Your account · Message boards


Copyright © 2014-2024 BOINC Confederation / rebirther