AP V7

Message boards : Number crunching : AP V7
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 20 · Next

AuthorMessage
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1587081 - Posted: 15 Oct 2014, 10:24:54 UTC - in response to Message 1586961.  
Last modified: 15 Oct 2014, 10:48:42 UTC

In my case a v6 AP task was taking circa 1 hour to do, so with 2xGPU along with 2 instances running per GPU I was pumping through nearly 100 v6 AP per day per i7 puter. The Qs run a bit less than that with their GPUs.

Parity with MB occurs around 369 credits per v7AP WU. Parity with v6AP occurs around 484 credits per v7AP WU.

The average credits for a v7AP WU are near parity with MB but just slightly under. The average credits for a v7AP WU are well off parity with v6AP.

To look at it another way, average credits for v6 were circa 650. Average credits for v7 are circa 350 (at present). This is represents about 54% of what they were previously per WU. At 30 tasks per day, you would need to increase throughput to 55.7 tasks per day to breakeven. At 48 tasks per day, you are still under water.

cheers

L.


But for my computer since 22:00 UTC the average granted has been 402.67, and average run time is 1806 sec. Which will give about 19,300 every 24 hours, which compares quite well with the 650 * 30 V6 tasks, 19,500.



@WinterKnight: I haven't looked at your run times yet but I am not seeing what you appear to be seeing.

Looking at your machine 7006214. Taking the last 20 results (no outliers) gives an average of 380.1635.

Looking at your machine 6714690. Taking the last 20 results (no outliers) gives an average of 349.6995.

Average across both your machines, 364.9315.

cheers

L.

Edit:
I just looked at your run times.
7006214: 1761.754 seconds
6714690: 1788.103 seconds
Average across both: 1774.928 seconds

cheers

I was only talking about 7006214, and about tasks done since 22:00 UTC, which was less than 20 tasks. If you had looked closely you would have seen the ones before 22:00 were all reported at the same time as a result of the maint period.

Since 22:00 the average is now 407.155 at average run time of 1740 sec, so 842/hr.

edit] As you brought up my sons computer, that only runs BOINC when he's up and not playing games or watching vids, the three results reported Today got 413 (1823s), 428 (1799s) and 424 (1814s) credits.
ID: 1587081 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1587158 - Posted: 15 Oct 2014, 14:32:28 UTC

Update to my last.

The results of last 20 AP V7 tasks is

Ave Runtime	Ave Credit	Credit/Hr
1739.323	413.615 	856.1


My last 15 AP V6 (thats all I have)

3362.705	744.499 	797.0
ID: 1587158 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1587228 - Posted: 15 Oct 2014, 17:59:04 UTC - in response to Message 1587158.  
Last modified: 15 Oct 2014, 18:07:38 UTC

Looking over 2 of my crunchers, looks like they just past thousand combined for APv7 validated. Credit seems to stabilized between 380-425 per AP. There are occasionally much higher and much lower credits but they are very few and not that often.

All AstroPulse v7 Task for
State: All (1041) · In progress (220) · Validation pending (264) · Validation inconclusive (7) · Valid (550) · Invalid (0) · Error (0)

I took this from the applications page of 1 computer

AstroPulse v7 (anonymous platform, NVIDIA GPU)
Number of tasks completed 87
Max tasks per day 150
Number of tasks today 58
Consecutive valid tasks 117
Average processing rate 522.27 GFLOPS
Average turnaround time 0.53 days

Here the other one

State: All (854) · In progress (183) · Validation pending (186) · Validation inconclusive (4) · Valid (481) · Invalid (0) · Error (0)

AstroPulse v7 (anonymous platform, NVIDIA GPU)
Number of tasks completed 21
Max tasks per day 57
Number of tasks today 121
Consecutive valid tasks 24
Average processing rate 557.53 GFLOPS
Average turnaround time 0.53 days

I haven't figured out how to convert GFLOPS to credits (any help would be appreciated) but this appears to be where it has stabilized out at.

However, on 1 machine I've noticed the results of the new APv7 dealing with highly blanked. Usually they only take about 4-6 secs and give 0.8-1 credits but then there are a few that run for 10-15 minutes with result of 80-90 credits. I've seen at least 100 so far. Interesting.

Happy Crunching...


Zalster

Edit

Thought I would compare stock AP v7 to Lunatics


AstroPulse v7 7.05 windows_intelx86 (opencl_nvidia_100)
Number of tasks completed 325
Max tasks per day 555
Number of tasks today 0
Consecutive valid tasks 522
Average processing rate 1,102.21 GFLOPS
Average turnaround time 0.92 days


AstroPulse v7 (anonymous platform, NVIDIA GPU)
Number of tasks completed 87
Max tasks per day 150
Number of tasks today 58
Consecutive valid tasks 117
Average processing rate 522.27 GFLOPS
Average turnaround time 0.53 days


other machine

AstroPulse v7 7.05 windows_intelx86 (opencl_nvidia_100)
Number of tasks completed 347
Max tasks per day 561
Number of tasks today 0
Consecutive valid tasks 454
Average processing rate 611.39 GFLOPS
Average turnaround time 1.33 days

AstroPulse v7 (anonymous platform, NVIDIA GPU)
Number of tasks completed 21
Max tasks per day 57
Number of tasks today 121
Consecutive valid tasks 24
Average processing rate 557.53 GFLOPS
Average turnaround time 0.53 days

Well, we all knew the lunatics would speed up the process.. Ok back to work...
ID: 1587228 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1587251 - Posted: 15 Oct 2014, 18:25:10 UTC - in response to Message 1587228.  
Last modified: 15 Oct 2014, 18:26:29 UTC

And taking into account that OpenCL AP7 apps are the same for "Lunatics" and "stock" (cause GPU stock are Lunatics) that comparison compares something another instead of application speed (maybe, selection of tuning options chosen?)
ID: 1587251 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1587307 - Posted: 15 Oct 2014, 19:23:04 UTC - in response to Message 1586928.  

My very first v7 AP WU had a credit of 1400, after that they all cratered. In looking at various peoples results, you occasionally get a result at 400+ however, the average appears to be hovering around 350. I can't really see any indication in anyone's data at present that it is going to get any better. The best it appears to be doing is heading towards equalibrium with MB.

cheers

L.

Many 400+ scores now, http://setiathome.berkeley.edu/results.php?hostid=6796479&offset=0&show_names=0&state=4&appid=

APv7 continues it's consistent march toward APv6. Considering v7 has close to the same APR & RunTime as v6, there isn't any reason it would not obtain parity...
ID: 1587307 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1587324 - Posted: 15 Oct 2014, 19:43:02 UTC - in response to Message 1587251.  

And taking into account that OpenCL AP7 apps are the same for "Lunatics" and "stock" (cause GPU stock are Lunatics) that comparison compares something another instead of application speed (maybe, selection of tuning options chosen?)


Opps, you are right about that as well Raistmer. I forgot about the commandline that I installed. I blame it on lack of sleep, lol.
ID: 1587324 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1639
Credit: 12,921,799
RAC: 89
New Zealand
Message 1587334 - Posted: 15 Oct 2014, 19:54:53 UTC - in response to Message 1587057.  

Are non-outlier tasks classed as tasks that run for less than 30 seconds because they are 100% blanked? Sorry if this is a very basic question I have no idea

Those 100% blanked / 30 second tasks are certainly classed as outliers, but so are less extreme cases with lower blanking, or which exit early because 30 pulses are found.

Thanks Richard.
ID: 1587334 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 680
Credit: 563,640,304
RAC: 597
Australia
Message 1587435 - Posted: 15 Oct 2014, 21:48:58 UTC - in response to Message 1587158.  
Last modified: 15 Oct 2014, 21:49:16 UTC

Update to my last.

The results of last 20 AP V7 tasks is

Ave Runtime	Ave Credit	Credit/Hr
1739.323	413.615 	856.1


My last 15 AP V6 (thats all I have)

3362.705	744.499 	797.0


---------------------------------------------------


In looking at your results, I see that in your last 20 they are mainly high 300s low-mid 400s, with a couple peaking over 500.

In looking at Ulrich Metzner's results I see (last 20) there are many in the 200s, so filtering these out and looking at the last 9, they range from 320+ to ~447.

In looking at TBar's busiest machine (and neglecting the high numbers of 2 second WUs), credit ranges from 335 to 486.

In looking at ExchangeMan's busiest machine I see (over the last 20) credit ranging from 393 to 670. The run time on the 670 credit was 3474 seconds.

Whilst I haven't compiled all the numbers related to the above in a spreadsheet (run times, credit, user, etc), it would appear that parity (circa 369 credits average) with MBv7 may have been achieved or that it is near parity at the moment. It also depends on whether the increase that is seen to date is sustained rather than a peaking.

cheers

L.
ID: 1587435 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 680
Credit: 563,640,304
RAC: 597
Australia
Message 1587438 - Posted: 15 Oct 2014, 21:53:55 UTC - in response to Message 1587435.  
Last modified: 15 Oct 2014, 21:54:20 UTC

I have just flipped MB off and APv7 on, on both the i7s to observe impact.

cheers

L.
ID: 1587438 · Report as offensive
Profile cliff
Avatar

Send message
Joined: 16 Dec 07
Posts: 625
Credit: 3,590,440
RAC: 0
United Kingdom
Message 1587472 - Posted: 15 Oct 2014, 23:14:16 UTC

Hi Folks,
Anyone have any idea why when running an AP GPU task with a Cuda50 WU on the same GPU the CUDA50 WU would take 3 times its normal running time to complete?

I've had this occur 3 times so far and a 4th instance is currently running the same way.

Regards,
Cliff,
Been there, Done that, Still no damm T shirt!
ID: 1587472 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1587481 - Posted: 15 Oct 2014, 23:21:42 UTC - in response to Message 1587435.  
Last modified: 15 Oct 2014, 23:32:34 UTC

I don't know where you are looking, the only computer to look at is Host 700621
I told you before the other computer is my sons and used infrequently for BOINC/Seti, it still has tasks sent on the 11th to process. My computers oldest in the "in progress" selection is from the 14th

1793.71	1762.69	441.02	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1788.93	1760.44	433.77	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1767.97	1747.41	411.62	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1756.2	1734.47	398.39	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2010.36	1946.03	507.85	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2078.84	1751	567.42	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1795.16	1742.11	448.19	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1776.35	1728.13	447.58	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1790.92	1744.93	383.78	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764.13	1743.67	432.65	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.11	1747.63	400.07	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.14	1743.37	445.37	AstroPulse v7 Anonymous platform (NVIDIA GPU)
6.24	1.31	0.68	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1754.33	1736.88	398.55	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1753.16	1736.74	374.29	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1763.1	1746.68	415.11	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1761.77	1741.85	413.22	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.91	1752.19	480.57	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764	1746.79	477.9	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.58	1741.46	412.41	AstroPulse v7 Anonymous platform (NVIDIA GPU)


These are the last 20 tasks as at time of posting. there are only 4 tasks below 400 credits (just) and 4 over 470.
ID: 1587481 · Report as offensive
Speedy
Volunteer tester
Avatar

Send message
Joined: 26 Jun 04
Posts: 1639
Credit: 12,921,799
RAC: 89
New Zealand
Message 1587482 - Posted: 15 Oct 2014, 23:22:54 UTC - in response to Message 1587472.  

Hi Folks,
Anyone have any idea why when running an AP GPU task with a Cuda50 WU on the same GPU the CUDA50 WU would take 3 times its normal running time to complete?

I've had this occur 3 times so far and a 4th instance is currently running the same way.

Regards,

I am completely guessing perhaps it could be the fact that the MB work unit is getting starved of CPU time.
ID: 1587482 · Report as offensive
Profile cliff
Avatar

Send message
Joined: 16 Dec 07
Posts: 625
Credit: 3,590,440
RAC: 0
United Kingdom
Message 1587486 - Posted: 15 Oct 2014, 23:34:40 UTC
Last modified: 15 Oct 2014, 23:37:57 UTC

Hi Folks,

Ref running times:- The AP task ran 27 mins 11 secs and the CUDA50 task ran for
27 mins 54 secs

The CUDA50 would normally take sub 17 min to run.

Both on a GTX760.
AP:- ap_08jl11ad_B0_P1_00279_20141014_19199.wu_3
C50:- 30au13ac.25621.305975.438086664206.12.146_2

Regards,
Cliff,
Been there, Done that, Still no damm T shirt!
ID: 1587486 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1587519 - Posted: 16 Oct 2014, 0:41:41 UTC - in response to Message 1587481.  

I don't know where you are looking, the only computer to look at is Host 700621
I told you before the other computer is my sons and used infrequently for BOINC/Seti, it still has tasks sent on the 11th to process. My computers oldest in the "in progress" selection is from the 14th

1793.71	1762.69	441.02	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1788.93	1760.44	433.77	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1767.97	1747.41	411.62	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1756.2	1734.47	398.39	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2010.36	1946.03	507.85	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2078.84	1751	567.42	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1795.16	1742.11	448.19	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1776.35	1728.13	447.58	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1790.92	1744.93	383.78	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764.13	1743.67	432.65	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.11	1747.63	400.07	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.14	1743.37	445.37	AstroPulse v7 Anonymous platform (NVIDIA GPU)
6.24	1.31	0.68	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1754.33	1736.88	398.55	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1753.16	1736.74	374.29	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1763.1	1746.68	415.11	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1761.77	1741.85	413.22	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.91	1752.19	480.57	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764	1746.79	477.9	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.58	1741.46	412.41	AstroPulse v7 Anonymous platform (NVIDIA GPU)


These are the last 20 tasks as at time of posting. there are only 4 tasks below 400 credits (just) and 4 over 470.

It looks like my average credit for normal tasks goes up about 40 each day since we started doing AP v7. Even the highly blanked tasks are tracking up as well.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1587519 · Report as offensive
woohoo
Volunteer tester

Send message
Joined: 30 Oct 13
Posts: 972
Credit: 165,671,404
RAC: 5
United States
Message 1587521 - Posted: 16 Oct 2014, 0:43:15 UTC

maybe nvidia pays better than ati
ID: 1587521 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 680
Credit: 563,640,304
RAC: 597
Australia
Message 1587531 - Posted: 16 Oct 2014, 1:13:28 UTC - in response to Message 1587481.  

I don't know where you are looking, the only computer to look at is Host 700621
I told you before the other computer is my sons and used infrequently for BOINC/Seti, it still has tasks sent on the 11th to process. My computers oldest in the "in progress" selection is from the 14th

1793.71	1762.69	441.02	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1788.93	1760.44	433.77	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1767.97	1747.41	411.62	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1756.2	1734.47	398.39	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2010.36	1946.03	507.85	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2078.84	1751	567.42	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1795.16	1742.11	448.19	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1776.35	1728.13	447.58	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1790.92	1744.93	383.78	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764.13	1743.67	432.65	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.11	1747.63	400.07	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.14	1743.37	445.37	AstroPulse v7 Anonymous platform (NVIDIA GPU)
6.24	1.31	0.68	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1754.33	1736.88	398.55	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1753.16	1736.74	374.29	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1763.1	1746.68	415.11	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1761.77	1741.85	413.22	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.91	1752.19	480.57	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764	1746.79	477.9	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.58	1741.46	412.41	AstroPulse v7 Anonymous platform (NVIDIA GPU)


These are the last 20 tasks as at time of posting. there are only 4 tasks below 400 credits (just) and 4 over 470.



As I said ... and here is one of the WUs that I was referring to on that machine (and only that machine), you know, one of the over 500s:
http://setiathome.berkeley.edu/workunit.php?wuid=1615669580
ID: 1587531 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 680
Credit: 563,640,304
RAC: 597
Australia
Message 1587533 - Posted: 16 Oct 2014, 1:15:29 UTC - in response to Message 1587472.  
Last modified: 16 Oct 2014, 1:26:08 UTC

Hi Folks,
Anyone have any idea why when running an AP GPU task with a Cuda50 WU on the same GPU the CUDA50 WU would take 3 times its normal running time to complete?

I've had this occur 3 times so far and a 4th instance is currently running the same way.

Regards,


Yes, it happens with me as well but not by 3x I think. I suspect it is a resourcing issue. I've never worried about it myself.

cheers

L.
ID: 1587533 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 680
Credit: 563,640,304
RAC: 597
Australia
Message 1587534 - Posted: 16 Oct 2014, 1:18:18 UTC - in response to Message 1587519.  

I don't know where you are looking, the only computer to look at is Host 700621
I told you before the other computer is my sons and used infrequently for BOINC/Seti, it still has tasks sent on the 11th to process. My computers oldest in the "in progress" selection is from the 14th

1793.71	1762.69	441.02	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1788.93	1760.44	433.77	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1767.97	1747.41	411.62	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1756.2	1734.47	398.39	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2010.36	1946.03	507.85	AstroPulse v7 Anonymous platform (NVIDIA GPU)
2078.84	1751	567.42	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1795.16	1742.11	448.19	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1776.35	1728.13	447.58	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1790.92	1744.93	383.78	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764.13	1743.67	432.65	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.11	1747.63	400.07	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.14	1743.37	445.37	AstroPulse v7 Anonymous platform (NVIDIA GPU)
6.24	1.31	0.68	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1754.33	1736.88	398.55	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1753.16	1736.74	374.29	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1763.1	1746.68	415.11	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1761.77	1741.85	413.22	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1771.91	1752.19	480.57	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1764	1746.79	477.9	AstroPulse v7 Anonymous platform (NVIDIA GPU)
1766.58	1741.46	412.41	AstroPulse v7 Anonymous platform (NVIDIA GPU)


These are the last 20 tasks as at time of posting. there are only 4 tasks below 400 credits (just) and 4 over 470.

It looks like my average credit for normal tasks goes up about 40 each day since we started doing AP v7. Even the highly blanked tasks are tracking up as well.



Just had a look at your machine 5255585. Good numbers.

cheers

L.
ID: 1587534 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1587613 - Posted: 16 Oct 2014, 5:01:36 UTC - in response to Message 1587472.  
Last modified: 16 Oct 2014, 5:02:01 UTC

Hi Folks,
Anyone have any idea why when running an AP GPU task with a Cuda50 WU on the same GPU the CUDA50 WU would take 3 times its normal running time to complete?

I've had this occur 3 times so far and a 4th instance is currently running the same way.

Regards,


One possibility is the nature of GPU. GPU is not preemptive system. kernel will run to completion. If one app (AP in this case) has bigger kernels and less gaps between kernel launches it will occupy more GPU and often. And GPU capabilities are limited hence to give more to one is to give less to other.
ID: 1587613 · Report as offensive
Profile cliff
Avatar

Send message
Joined: 16 Dec 07
Posts: 625
Credit: 3,590,440
RAC: 0
United Kingdom
Message 1587617 - Posted: 16 Oct 2014, 5:13:33 UTC - in response to Message 1587613.  

Hi,
OK, so AP is a greedy hog:-) So now in order to prevent it suffocating other GPU tasks a way to ensure that AP tasks work by themselves on the GPU or only with other AP tasks in rigs with multi GPU and more than one task per GPU

But I wont hold my breath waiting for THAT to happen:-)

Regards,
Cliff,
Been there, Done that, Still no damm T shirt!
ID: 1587617 · Report as offensive
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 20 · Next

Message boards : Number crunching : AP V7


 
©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.