PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78944)
Posted 2825 days ago by DutchDK
Please do. I actually had to set up 1.5 CPU cores to each GPU task in app_config, to max out the R9-280x's.
2) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78805)
Posted 2832 days ago by DutchDK
2.5c run

C:\Users\Dutch\Downloads\tps-cl-alpha (2)>tpsieve-cl-boinc-x86-windows.exe -p131
20716e9 -P13120725e9 -k5 -K9999 -n6000000 -N9000000 -T -M2 -c 60
tpsieve version cl-0.2.5c (testing)
nstart=6000000, nstep=41
tpsieve initialized: 5 <= k <= 9999, 6000000 <= n < 9000000
nstep changed to 32
CL setup complete.
cthread_count = 131072
13120716307705079 | 2959*2^8094354+1
13120716380747377 | 8859*2^7847166+1
13120716412325987 | 7211*2^7702451+1
13120716843190109 | 2167*2^7951358+1
13120716846657643 | 7003*2^8206351-1
13120717075819579 | 5089*2^8180582+1
13120717160591431 | 7745*2^8765661+1
13120717284827623 | 8907*2^8826378+1
13120717824685699 | 4613*2^7552946-1
13120717861537051 | 5537*2^7081032-1
13120718815920467 | 2501*2^8550910-1
13120718930066591 | 3699*2^7482137-1
13120719057141307 | 6089*2^8973032-1
13120719059517079 | 2445*2^7778847-1
13120719477216161 | 6435*2^8752766+1
13120719544063997 | 2985*2^7826085+1
13120719661679807 | 8253*2^8382530-1
13120719782159167 | 1543*2^7389775-1
13120719966492419 | 3749*2^6399636-1
13120720493876977 | 6645*2^7538049+1
13120720553340431 | 2055*2^8038331+1
13120720607106341 | 8311*2^7968572+1
13120720635031169 | 5679*2^8994161+1
13120720662833737 | 5469*2^6562478+1
13120720682005019 | 1935*2^7473932-1
13120721017491301 | 3531*2^6213937-1
13120721212413691 | 8341*2^6170513-1
13120721663769431 | 3585*2^6907848+1
13120721691198937 | 6395*2^7988954-1
13120721797389533 | 8075*2^8648682-1
13120721851239341 | 8595*2^6216665-1
13120721947519033 | 3327*2^6867165-1
13120722319646957 | 5003*2^6419493+1
13120722440108203 | 5707*2^6807554+1
13120722557491823 | 1731*2^8986587+1
13120722686753867 | 8135*2^7034866-1
13120723071996931 | 4241*2^7534559+1
13120723189240451 | 1119*2^6735109-1
13120723194075703 | 2055*2^6453532+1
13120723256598527 | 8165*2^8191313+1
13120723258653649 | 4355*2^7639969+1
13120723434822539 | 8535*2^6601016+1
13120723717933309 | 9707*2^7180267+1
13120723954016033 | 4685*2^8440045+1
13120723983397483 | 5737*2^6363638+1
13120724045408101 | 2565*2^6057669+1
13120724075161099 | 6163*2^7807643-1
13120724084494907 | 3081*2^6075501+1
13120724364906871 | 4281*2^8009816+1
13120724414854831 | 7821*2^6744179+1
13120724617851359 | 9975*2^7303461-1
13120724673389471 | 2361*2^6415555+1
13120724738634179 | 1485*2^6690363+1
Found 53 factors


stderr.txt :
19:00:59 (1948): Can't open init data file - running in standalone mode
Sieve started: 13120716000000000 <= p < 13120725000000000
Thread 0 starting
Detected 512 multiprocessors (2560 SPUs) on device 0.
Device 0 is a 'Advanced Micro Devices, Inc.' 'Tahiti'.
GCN device detected; use -m1 --vecsize=4 to undo effect

Thread 0 completed
Sieve complete: 13120716000000000 <= p < 13120725000000000
count=242515220,sum=0x2dca915341c02724
Elapsed time: 672.91 sec. (0.66 init + 672.25 sieve) at 13388222 p/sec.
Processor time: 1038.53 sec. (0.67 init + 1037.86 sieve) at 8671881 p/sec.
Average processor utilization: 1.01 (init), 1.54 (sieve)
19:12:12 (1948): called boinc_finish
3) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78651)
Posted 2837 days ago by DutchDK
2.5b output :
C:\Users\Dutch\Downloads\tps-cl-alpha (1)>tpsieve-cl-boinc-x86-windows.exe -p131
20716e9 -P13120725e9 -k5 -K9999 -n6000000 -N9000000 -T -M2 -c 60
tpsieve version cl-0.2.5b (testing)
nstart=6000000, nstep=41
tpsieve initialized: 5 <= k <= 9999, 6000000 <= n < 9000000
nstep changed to 32
CL setup complete.
cthread_count = 131072
13120716307705079 | 2959*2^8094354+1
13120716380747377 | 8859*2^7847166+1
13120716412325987 | 7211*2^7702451+1
13120716843190109 | 2167*2^7951358+1
13120716846657643 | 7003*2^8206351-1
13120717075819579 | 5089*2^8180582+1
13120717160591431 | 7745*2^8765661+1
13120717284827623 | 8907*2^8826378+1
13120717824685699 | 4613*2^7552946-1
13120717861537051 | 5537*2^7081032-1
13120718815920467 | 2501*2^8550910-1
13120718930066591 | 3699*2^7482137-1
13120719057141307 | 6089*2^8973032-1
13120719059517079 | 2445*2^7778847-1
13120719477216161 | 6435*2^8752766+1
13120719544063997 | 2985*2^7826085+1
13120719661679807 | 8253*2^8382530-1
13120719782159167 | 1543*2^7389775-1
13120719966492419 | 3749*2^6399636-1
13120720493876977 | 6645*2^7538049+1
13120720553340431 | 2055*2^8038331+1
13120720607106341 | 8311*2^7968572+1
13120720635031169 | 5679*2^8994161+1
13120720662833737 | 5469*2^6562478+1
13120720682005019 | 1935*2^7473932-1
13120721017491301 | 3531*2^6213937-1
13120721212413691 | 8341*2^6170513-1
13120721663769431 | 3585*2^6907848+1
13120721691198937 | 6395*2^7988954-1
13120721797389533 | 8075*2^8648682-1
13120721851239341 | 8595*2^6216665-1
13120721947519033 | 3327*2^6867165-1
13120722319646957 | 5003*2^6419493+1
13120722440108203 | 5707*2^6807554+1
13120722557491823 | 1731*2^8986587+1
13120722686753867 | 8135*2^7034866-1
13120723071996931 | 4241*2^7534559+1
13120723189240451 | 1119*2^6735109-1
13120723194075703 | 2055*2^6453532+1
13120723256598527 | 8165*2^8191313+1
13120723258653649 | 4355*2^7639969+1
13120723434822539 | 8535*2^6601016+1
13120723717933309 | 9707*2^7180267+1
13120723954016033 | 4685*2^8440045+1
13120723983397483 | 5737*2^6363638+1
13120724045408101 | 2565*2^6057669+1
13120724075161099 | 6163*2^7807643-1
13120724084494907 | 3081*2^6075501+1
13120724364906871 | 4281*2^8009816+1
13120724414854831 | 7821*2^6744179+1
13120724617851359 | 9975*2^7303461-1
13120724673389471 | 2361*2^6415555+1
13120724738634179 | 1485*2^6690363+1
Found 53 factors

stderr.txt :
19:25:22 (10828): Can't open init data file - running in standalone mode
Sieve started: 13120716000000000 <= p < 13120725000000000
Thread 0 starting
Detected 512 multiprocessors (2560 SPUs) on device 0.
Device 0 is a 'Advanced Micro Devices, Inc.' 'Tahiti'.
GCN device detected; use -m1 --vecsize=4 to undo effect

Thread 0 completed
Sieve complete: 13120716000000000 <= p < 13120725000000000
count=242515220,sum=0x2dca915341c02724
Elapsed time: 672.34 sec. (0.62 init + 671.72 sieve) at 13398667 p/sec.
Processor time: 1038.12 sec. (0.62 init + 1037.50 sieve) at 8674880 p/sec.
Average processor utilization: 1.01 (init), 1.54 (sieve)
19:36:34 (10828): called boinc_finish
4) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78499)
Posted 2846 days ago by DutchDK

The first thing you need to realize is that I don't trust AMD's OpenCL to compute the tricky initialization of each step right. Nvidia does it on the GPU, but AMD had failures early on. An alpha GCC also had issues with that, so I gather it's tricky for compilers to get it right. Anyway, this means doing a significant amount of work on the CPU.


Out of pure interest, the current app does the init on the GPU ? Since the results with both the current and the new was the same on the windows run.


Then, with the app making errors, I implemented an error-check for each step. But this takes about the same time as the initialization, so that nearly doubles the CPU usage per step.

Possible to have the errorcheck only enabled when running with a -debug commandline parameter ? Or is it inlined with the rest of the code for the steps its checking ?

If you can make the errorcheck optional, that would then nearly halve the cpu usage.
5) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78485)
Posted 2846 days ago by DutchDK

I'm a little concerned that DutchDK may have been CPU-limited in his test. (He has lots of cores, but only one can be used for the CPU part of the sieve.) But that's something I can work on optimizing later; it shouldn't prevent deployment of the current apps.

Ken


Not really a big issue, Ken. In Boinc, I set up one logical core for each GPU task. I tend to run 2 PPS sieve tasks on each R9-280X GPU, so two physical cores/4 logical cores are dedicated to the GPU tasks, which is fine.


What I am more curious about, is this in the new app :

cthread_count = 131072


compared to this in the current app :

cthread_count = 32768


Is this the cause of the speedup in the new app ?

And also the stderr timereports looks weird,
new:

Elapsed time: 707.11 sec. (0.47 init + 706.64 sieve) at 12736609 p/sec.
Processor time: 1037.47 sec. (0.47 init + 1037.00 sieve) at 8679056 p/sec.
Average processor utilization: 1.00 (init), 1.47 (sieve)

versus old:

Elapsed time: 1262.26 sec. (0.45 init + 1261.81 sieve) at 7132738 p/sec.
Processor time: 354.23 sec. (0.47 init + 353.76 sieve) at 25441265 p/sec.
Average processor utilization: 1.05 (init), 0.28 (sieve)


Its almost like processor and elapsed time reports have been swapped in the new vs old app.
6) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78444)
Posted 2848 days ago by DutchDK
I know its summer and holidays etc has an effect, but any eta/news on possible reinstatement of the ATI CL app for PPS Sieving ?
7) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78393)
Posted 2850 days ago by DutchDK
And just for good measure, same test with the current Boinc executable :
C:\Users\Dutch\Downloads\tps-cl-alpha>primegrid_tpsieve_1.39_windows_intelx86__a
tiPPSsieve.exe -p13120716e9 -P13120725e9 -k5 -K9999 -n6000000 -N9000000 -T -M2 -
c 60
tpsieve version cl-0.2.3e (testing)
nstart=6000000, nstep=41
tpsieve initialized: 5 <= k <= 9999, 6000000 <= n < 9000000
nstep changed to 32
CL setup complete.
cthread_count = 32768
13120716307705079 | 2959*2^8094354+1
13120716380747377 | 8859*2^7847166+1
13120716412325987 | 7211*2^7702451+1
13120716843190109 | 2167*2^7951358+1
13120716846657643 | 7003*2^8206351-1
13120717075819579 | 5089*2^8180582+1
13120717160591431 | 7745*2^8765661+1
13120717284827623 | 8907*2^8826378+1
13120717824685699 | 4613*2^7552946-1
13120717861537051 | 5537*2^7081032-1
13120718815920467 | 2501*2^8550910-1
13120718930066591 | 3699*2^7482137-1
13120719057141307 | 6089*2^8973032-1
13120719059517079 | 2445*2^7778847-1
13120719477216161 | 6435*2^8752766+1
13120719544063997 | 2985*2^7826085+1
13120719661679807 | 8253*2^8382530-1
13120719782159167 | 1543*2^7389775-1
13120719966492419 | 3749*2^6399636-1
13120720493876977 | 6645*2^7538049+1
13120720553340431 | 2055*2^8038331+1
13120720607106341 | 8311*2^7968572+1
13120720635031169 | 5679*2^8994161+1
13120720662833737 | 5469*2^6562478+1
13120720682005019 | 1935*2^7473932-1
13120721017491301 | 3531*2^6213937-1
13120721212413691 | 8341*2^6170513-1
13120721663769431 | 3585*2^6907848+1
13120721691198937 | 6395*2^7988954-1
13120721797389533 | 8075*2^8648682-1
13120721851239341 | 8595*2^6216665-1
13120721947519033 | 3327*2^6867165-1
13120722319646957 | 5003*2^6419493+1
13120722440108203 | 5707*2^6807554+1
13120722557491823 | 1731*2^8986587+1
13120722686753867 | 8135*2^7034866-1
13120723071996931 | 4241*2^7534559+1
13120723189240451 | 1119*2^6735109-1
13120723194075703 | 2055*2^6453532+1
13120723256598527 | 8165*2^8191313+1
13120723258653649 | 4355*2^7639969+1
13120723434822539 | 8535*2^6601016+1
13120723717933309 | 9707*2^7180267+1
13120723954016033 | 4685*2^8440045+1
13120723983397483 | 5737*2^6363638+1
13120724045408101 | 2565*2^6057669+1
13120724075161099 | 6163*2^7807643-1
13120724084494907 | 3081*2^6075501+1
13120724364906871 | 4281*2^8009816+1
13120724414854831 | 7821*2^6744179+1
13120724617851359 | 9975*2^7303461-1
13120724673389471 | 2361*2^6415555+1
13120724738634179 | 1485*2^6690363+1
Found 53 factors

Stderr.txt :
04:53:01 (1992): Can't open init data file - running in standalone mode
Sieve started: 13120716000000000 <= p < 13120725000000000
Thread 0 starting
Detected 512 multiprocessors (2560 SPUs) on device 0.

Thread 0 completed
Sieve complete: 13120716000000000 <= p < 13120725000000000
count=242515220,sum=0x2dca915341c02724
Elapsed time: 1262.26 sec. (0.45 init + 1261.81 sieve) at 7132738 p/sec.
Processor time: 354.23 sec. (0.47 init + 353.76 sieve) at 25441265 p/sec.
Average processor utilization: 1.05 (init), 0.28 (sieve)
05:14:03 (1992): called boinc_finish
8) Message boards : Proth Prime Search : ATI sieve app turned off; Help needed. (Message 78383)
Posted 2850 days ago by DutchDK
Win7-x64 box, Dual Xeon X5570, 48 GB DDR3 ECC, Dual R9-280X cards.

C:\Users\Dutch\Downloads\tps-cl-alpha>tpsieve-cl-boinc-x86-windows.exe -p1312071
6e9 -P13120725e9 -k5 -K9999 -n6000000 -N9000000 -T -M2 -c 60
tpsieve version cl-0.2.5a (testing)
nstart=6000000, nstep=41
tpsieve initialized: 5 <= k <= 9999, 6000000 <= n < 9000000
nstep changed to 32
CL setup complete.
cthread_count = 131072
13120716307705079 | 2959*2^8094354+1
13120716380747377 | 8859*2^7847166+1
13120716412325987 | 7211*2^7702451+1
13120716843190109 | 2167*2^7951358+1
13120716846657643 | 7003*2^8206351-1
13120717075819579 | 5089*2^8180582+1
13120717160591431 | 7745*2^8765661+1
13120717284827623 | 8907*2^8826378+1
13120717824685699 | 4613*2^7552946-1
13120717861537051 | 5537*2^7081032-1
13120718815920467 | 2501*2^8550910-1
13120718930066591 | 3699*2^7482137-1
13120719057141307 | 6089*2^8973032-1
13120719059517079 | 2445*2^7778847-1
13120719477216161 | 6435*2^8752766+1
13120719544063997 | 2985*2^7826085+1
13120719661679807 | 8253*2^8382530-1
13120719782159167 | 1543*2^7389775-1
13120719966492419 | 3749*2^6399636-1
13120720493876977 | 6645*2^7538049+1
13120720553340431 | 2055*2^8038331+1
13120720607106341 | 8311*2^7968572+1
13120720635031169 | 5679*2^8994161+1
13120720662833737 | 5469*2^6562478+1
13120720682005019 | 1935*2^7473932-1
13120721017491301 | 3531*2^6213937-1
13120721212413691 | 8341*2^6170513-1
13120721663769431 | 3585*2^6907848+1
13120721691198937 | 6395*2^7988954-1
13120721797389533 | 8075*2^8648682-1
13120721851239341 | 8595*2^6216665-1
13120721947519033 | 3327*2^6867165-1
13120722319646957 | 5003*2^6419493+1
13120722440108203 | 5707*2^6807554+1
13120722557491823 | 1731*2^8986587+1
13120722686753867 | 8135*2^7034866-1
13120723071996931 | 4241*2^7534559+1
13120723189240451 | 1119*2^6735109-1
13120723194075703 | 2055*2^6453532+1
13120723256598527 | 8165*2^8191313+1
13120723258653649 | 4355*2^7639969+1
13120723434822539 | 8535*2^6601016+1
13120723717933309 | 9707*2^7180267+1
13120723954016033 | 4685*2^8440045+1
13120723983397483 | 5737*2^6363638+1
13120724045408101 | 2565*2^6057669+1
13120724075161099 | 6163*2^7807643-1
13120724084494907 | 3081*2^6075501+1
13120724364906871 | 4281*2^8009816+1
13120724414854831 | 7821*2^6744179+1
13120724617851359 | 9975*2^7303461-1
13120724673389471 | 2361*2^6415555+1
13120724738634179 | 1485*2^6690363+1
Found 53 factors

stderr.txt :
23:46:40 (3872): Can't set up shared mem: -1. Will run in standalone mode.
Sieve started: 13120716000000000 <= p < 13120725000000000
Thread 0 starting
Detected 512 multiprocessors (2560 SPUs) on device 0.
Device 0 is a Advanced Micro Devices, Inc. Tahiti.
GCN device detected; use -m1 --vecsize=4 to undo effect

Thread 0 completed
Sieve complete: 13120716000000000 <= p < 13120725000000000
count=242515220,sum=0x2dca915341c02724
Elapsed time: 707.11 sec. (0.47 init + 706.64 sieve) at 12736609 p/sec.
Processor time: 1037.47 sec. (0.47 init + 1037.00 sieve) at 8679056 p/sec.
Average processor utilization: 1.00 (init), 1.47 (sieve)
23:58:27 (3872): called boinc_finish

9) Message boards : Problems and Help : problem with 7790 Bonaire (Message 76462)
Posted 2921 days ago by DutchDK
CGN 1.1 vs CGN 1.0 cards. R7-260X = 7790, both Bonaire based, both are CGN 1.1, both have CAL deprecated in any catalyst driver later than 13.4.

R9-280X = 7970, both are Tahiti based, both are CGN 1.0, both have CAL enabled in later drivers.

Currently running 2 r9-280X's, and using 14.4 drivers. Used to run a 7790, and had to use 13.4, in order to not fiddle with app_info.xml.
Not a Primegrid problem, nor a Boinc problem. AMD has decided to deprecate CAL, and has done so for their CGN 1.1 cards, and probably also will do so for the CGN 1.0 cards later.

10) Message boards : Number crunching : New power per watt GPU crunching winner? (Message 74998)
Posted 2971 days ago by DutchDK
Aaannd the results are in, and they're REALLY DISAPPOINTING!


Those results seem to be quite realistic considering following:

Nvidia's mainstream Kepler-based GPUs offered double-precision compute performance that was 1/24 as fast as its FP32 math. Maxwell is ever worse at 1/32.

Source

but the power per watt performance is still better on your new 750ti


I'm running an ASUS AMD Radeon HD 7790, bought for the equivalent of 120USD, pulling 85 watt at max, and the genefer WU finishes in 20 hours.
I'm using dual XEON X5570's and have hyper treading enabled, allocating 0.33 CPU to the GPU processes.

So power/performance wise with most bang for the buck, I still think the mid/low cost DP capable AMD cards holds the edge.

Planning on getting one more HD7790 or the equivalent R7 260x, to fill up the second pci-e slot in my workstation.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 4.44, 3.77, 3.63
Generated 17 May 2022 | 14:43:56 UTC