Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Generalized Fermat Prime Search :
GFN21 GFN22 Scoreboard, post your times/hardware
| Author |
Message |
|---|
|
|
|
I've been searching and have not yet found a similar thread. I was hoping to see some reports (be honest) on how quickly your hardware can crunch GFN21 and GFN22 tasks.
From Michael Goetz-One thing to remember is that ONLY GFN 21 and GFN 22, amongst all the apps at PrimeGrid, still use double precision. All the other GFN apps, as well as AP27 and PPS-Sieve, do not use double precision at all. Not sure how long ago that was, hope it's still true...
Frankly, I've acquired what is supposed to be some serious DP hardware, and it doesn't seem to be living up to it's potential, but it does have what seems to be a good start on the first of a batch of GFN21.
So I was hoping to get times from others to see if this card needs to go back to the seller or not. | |
|
|
|
|
Recently ran my first GFN21 and 22 tasks on my GTX 980 Ti:
GFN22 - Run time (sec): 187,099.15 - CPU time (sec): 658.23 - Genefer 22 v3.13 (OCLcudaGFNWR)
GFN21 - 95,441.37 - 289.42 - Genefer 21 v3.13 (OCLcudaGFN)
GFN21 - 96,064.81 - 293.05 - Genefer 21 v3.13 (OCLcudaGFN)
Nothing else than BOINC was running on the computer.
| |
|
|
|
|
I've been trying to go through the list of computers from the more productive members, to answer my own question.
So far I've seen a few GFN21's in the 50,000 second range from a GTX1070 or two, and 44,000 secs from a GTX1080.
For the GFN22's, I've seen 195k secs from a 1070 and 208k secs from a Titan(X?). | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14044
ID: 53948
Credit: 482,370,180
RAC: 568,853
                               
|
From Michael Goetz-One thing to remember is that ONLY GFN 21 and GFN 22, amongst all the apps at PrimeGrid, still use double precision. All the other GFN apps, as well as AP27 and PPS-Sieve, do not use double precision at all. Not sure how long ago that was, hope it's still true...
While what I said is true... there's a caveat that is important. Recent generations of Nvidia GPUs, while very awesome, just plain suck at double precision. The last generation of really powerful double precision Nvidia GPus were the GTX 5xx series, unless you count some TITAN models.
Because of this, Maxwell and newer GPUs will generally choose to skip the double precision version of the algorithm (i.e., is will skip OCL in favor of OCL4). So double precision, at the moment, doesn't matter that much with Nvidia GPUs unless you're running an old GPU, or unless future generations have improved double precision performance. I consider that unlikely since double precision isn't very useful for gaming, so there's not much of a reason to build hardware that excels at DP math.
Even though the double precision performance of new GPUs is awful, GFN performance is still way, way up because the GPUs are so much faster and larger. My mid-range GTX 1060 is about twice as fast as the top-of-the-line GTX 580 that it replaced.
So I was hoping to get times from others...
Have you seen this page: https://www.primegrid.com/gpu_list.php#GFN22 ?
The server collects run time information on GPU tasks. The results can be distorted for various reasons, so take the information with a grain of salt, but it should give you a decent idea of the ballpark speeds of various GPUs.
My own times with a GTX 1060 (6GB):
GFN20: About 20,500 seconds
GFN21: About 67,500 seconds
GFN22: About 260,500 seconds
(IIRC the GTX 580 could do a GFN22 in about 500K seconds, give or take a day or two. Or at least it could if it was stable enough to do so. It had some reliability problems.)
____________
My lucky number is 75898524288+1 | |
|
|
|
|
MSI GeForce GTX 1080 Gaming X (no OC, fan @ 60%)
AP 27 : 18.80 min (1128 seconds) -> 1 x 4043 credits/unit = 309,677 credits/day**
GFN 22 : 45.27 hrs (163k seconds) -> 1 x 673,033 credits/unit = 356,810 credits/day****
PPS Sieve: 7.70 min ( 462 seconds) -> 2 x 3371 credits/unit = 1,260,842 credits/day***
PPR12M : throughput @ 36.07M p/sec, 0.25 CPU cores *
* with all eight threads on i7 6700k running TRP Sieve
** with all eight threads on i7 6700k running TRP Sieve and one GPU unit running simultaneously
*** with all eight threads on i7 6700k running TRP Sieve and two GPU units running simultaneously
**** with all four cores on i7 6700k running PPS MEGA/SGS and one GPU unit running simultaneously
Gigabyte GTX 1060 G1 3 Gb (no OC, fan @ 60%)
AP27 : 40.7 min (2442 seconds) -> 4043 credits/unit = 143,045 credits/day**
GFN 22 : 74.69 hrs (268.9k seconds) -> 1 x 673,075 credits/unit = 216,267 credits/day****
PPS Sieve: 9.28 min ( 557 seconds) -> 3371 credits/unit = 523,086 credits/day***
PPR12M : throughput @ 16.85M p/sec, 0.25 CPU cores *
* with all eight threads on i7 6700k running TRP Sieve
** with all eight threads on i7 6700k running TRP Sieve and one GPU unit running simultaneously
*** with all eight threads on i7 6700k running TRP Sieve and two GPU units running simultaneously
**** with both cores on i3 6300 running PPS MEGA/SGS and one GPU unit running simultaneously
i7 6700k Skylake (4.0 GHz) with 2x4GB DDR4 2666 MHz running at 2133 MHz (due to the mobo)
GFN 20: 25 hrs with four physical cores running GFN 20
i3 6300 Skylake (3.8 GHz) with 2x4 Gb DDR4 2666 MHz running at 2133 MHz (due to the mobo)
GFN 20: 31.91 hrs with both physical cores running GFN 20
i3 6100 Skylake (3.7 Ghz) with 2x4 Gb DDR4 2666 MHz running at 2133 MHz (due to the mobo)
GFN 20: 31.91 hrs with both physical cores running GFN 20
GFN 21: 140.84 hrs with both physical cores running GFN 20/21 | |
|
|
|
|
I'm running a AMD FirePro S9150, with a top DP of 2.53 TFLOPS, single precision in the 5.1 TFLOPS ballpark. So far it is sucking at every DC project I toss to it, no better and often worse than a R9-280X. | |
|
|
|
|
AMD A10-7800 APU 3,183,434 seconds GFN22 :(
Will have my new 1080ti running one before May 13th though so that'll be interesting | |
|
tngSend message
Joined: 29 Aug 10
Posts: 500
ID: 66603
Credit: 50,926,324,921
RAC: 26,135,723
                                                   
|
I've been trying to go through the list of computers from the more productive members, to answer my own question.
So far I've seen a few GFN21's in the 50,000 second range from a GTX1070 or two, and 44,000 secs from a GTX1080.
For the GFN22's, I've seen 195k secs from a 1070 and 208k secs from a Titan(X?).
Take a look at my systems for some GTX1070/1080 results.
____________
 | |
|
tngSend message
Joined: 29 Aug 10
Posts: 500
ID: 66603
Credit: 50,926,324,921
RAC: 26,135,723
|
My own times with a GTX 1060 (6GB):
GFN20: About 20,500 seconds
GFN21: About 67,500 seconds
GFN22: About 260,500 seconds
(IIRC the GTX 580 could do a GFN22 in about 500K seconds, give or take a day or two. Or at least it could if it was stable enough to do so. It had some reliability problems.)
For a GTX1050Ti:
GFN21: about 108,000 seconds
GFN22: about 425,000 seconds
This from a card that requires no supplemental power connections and is available in low profile versions. Until the GTX10xx cards came out, I was running a number of GTX580s (I didn't encounter any reliability problems). I no longer own any.
____________
| |
|
MonkeydeeVolunteer tester
Send message
Joined: 8 Dec 13
Posts: 548
ID: 284516
Credit: 1,725,565,773
RAC: 3,187,175
                    
|
For a GTX1050Ti:
GFN21: about 108,000 seconds
GFN22: about 425,000 seconds
Thanks! Now I know what I'll be facing should I decide to move up from GFN16.
The 1050ti's are very impressive for the price. Snappy at all Primegrid tasks, great for 1080p gaming, use very little electricity, and make very little heat.
____________
My Primes
Badge Score: 4*2 + 6*2 + 7*1 + 8*11 + 9*1 + 11*3 + 12*1 = 169
 | |
|
|
|
I'm running a AMD FirePro S9150, with a top DP of 2.53 TFLOPS, single precision in the 5.1 TFLOPS ballpark. So far it is sucking at every DC project I toss to it, no better and often worse than a R9-280X.
I feel your pain.
I have a pair of FirePro W9100s. They should also have 2.5 TFLOPS DP. They don't, at least not for GFN. They are roughly equivalent to my 290x in both DP and SP for tasks here, and draw a whole lot of power in the process.
My 280x and individual cores of my 7990s are (a lot) faster. Without going digging, IIRC GFN22 ran around 275k seconds on my 290x and 9100s (Hawaii cores), 210-220k on my Tahiti cores, depending on how they are clocked at the time.
| |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
                   
|
|
GTX 760:
GFN 21 - about 121,000 seconds
GFN 22 - about 508,500 seconds
| |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
|
GTX 760:
GFN 21 - about 121,000 seconds
GFN 22 - about 508,500 seconds
That was (mainly GPU) runtime. Add in CPU time (significant for OCL on NVidia on Linux):
GFN 21 - about 107,000 seconds
GFN 22 - about 466,700 seconds | |
|
|
|
|
I have a GTX 1080 Ti Founder's Edition (reference blower type), no manual overclock, set to 75c degree limit with fan curve. The card may have ramped up the core clock speed automatically but I did not make a note of it. I only cared about stability and error free task completion.
It completed and validated two GFN22 tasks a couple of weeks ago at 114,044.27 and 113,403.93 seconds. | |
|
|
|
|
1080ti FE (running +209MHz GPU clock +500MHz MEM clock) It can be overclocked alot more but I just am not seeing a performance increase enough to justify.
GFN19 3,057 seconds
GFN20 9,904 seconds
GFN21 35,120 seconds
GFN22 112,714 seconds | |
|
|
|
|
Just to round out my last post
GFN 18 837.87 seconds
GFN 17MEGA 396.94 Seconds
GFN 17low 300.01 Seconds
GFN 16 112.34 Seconds
GFN 15 48.28 Seconds | |
|
GDBSend message
Joined: 15 Nov 11
Posts: 304
ID: 119185
Credit: 4,291,416,653
RAC: 1,774,558
       
|
|
1080 OC
GFN 19 91 sec.
GFN 20 102 sec.
GFN 21 171 sec.
GFN 22 209 sec. | |
|
|
|
|
Running on platform 'NVIDIA CUDA', device 'GeForce GTX 1070 Ti', vendor 'NVIDIA Corporation', version 'OpenCL 1.2 CUDA' and driver '398.11'.
19 computeUnits @ 1683MHz, memSize=8192MB, cacheSize=304kB, cacheLineSize=128B, localMemSize=48kB, maxWorkGroupSize=1024.
Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5
Command line: projects/www.primegrid.com/primegrid_genefer_3_3_3_3.19_windows_intelx86__OCLcudaGFNWR.exe -boinc -q 128830^4194304+1
Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz [Family 6 Model 63 Stepping 2]
GPU 183,045.07
CPU 1,931.95
Credits 697,525.44
GFN 22 | |
|
|
|
|
First GFN 21 task on my NVIDIA GeForce GTX 1060 (6GB):
GFN 21: 69,952.19 sec - 147,566.26 credits | |
|
|
|
|
TITAN V (OC: GPU+65MHz , MEM+175MHz)
GFN21: 13,420 seconds
| |
|
Yves GallotVolunteer developer
Project scientist
Send message
Joined: 19 Aug 12
Posts: 843
ID: 164101
Credit: 306,554,426
RAC: 5,458
|
TITAN V (OC: GPU+65MHz , MEM+175MHz)
GFN21: 13,420 seconds
It's very impressive, this is more than twice the speed of a 1080Ti!
We have about (GFN21)
GTX 1060: 20 h
GTX 1070: 14 h
GTX 1080: 12 h
GTX 1080Ti: 8 h 30
TITAN V: 4 h
Note that the GeForce GTX use ocl4 (int32) but the TITAN V selects ocl (fp64). Then we can't try to predict the Turing performance with Volta running time. | |
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1963
ID: 352
Credit: 6,420,056,055
RAC: 2,653,228
                        
|
|
RTX 2080: 7 h
____________
My stats | |
|
Yves GallotVolunteer developer
Project scientist
Send message
Joined: 19 Aug 12
Posts: 843
ID: 164101
Credit: 306,554,426
RAC: 5,458
|
RTX 2080: 7 h
Very good news! Faster than the GTX 1080Ti :-) and for the same price.
The RTX 2080Ti should take about 5 hours.
| |
|
Yves GallotVolunteer developer
Project scientist
Send message
Joined: 19 Aug 12
Posts: 843
ID: 164101
Credit: 306,554,426
RAC: 5,458
|
|
I found in the task list:
Tesla V100-SXM2-16GB: 3 h 07
http://www.primegrid.com/results.php?hostid=937863&offset=0&show_names=0&state=0&appid=16
Congratulations to srowen!
Who says better? | |
|
|
|
|
Ancient Titan Black: 19 hours, but you have to force the software to use OCL instead of OCL4 (runtime prediction at the beginnig is false)
1070 TI: 13 hours.
____________
DeleteNull | |
|
robishVolunteer moderator
Volunteer tester
Send message
Joined: 7 Jan 12
Posts: 2223
ID: 126266
Credit: 7,973,124,568
RAC: 5,430,423
        
|
|
Titan X is 13.6 hours for GFN21.
____________
My lucky number 10590941048576+1 | |
|
LookASVolunteer tester
Send message
Joined: 19 Apr 08
Posts: 38
ID: 21649
Credit: 354,890,618
RAC: 0
      
|
I found in the task list:
Tesla V100-SXM2-16GB: 3 h 07
http://www.primegrid.com/results.php?hostid=937863&offset=0&show_names=0&state=0&appid=16
Congratulations to srowen!
Who says better?
and same user same card GFN22 - 12h 5m :-o
http://www.primegrid.com/results.php?hostid=937863&offset=0&show_names=0&state=0&appid=17 | |
|
|
|
TITAN V: 4 h
Tesla V100-SXM2-16GB: 3 h 07
Titan V may be limited by memory-bandwidth for ocl in GFN21.
GPU-Z
GPU-Z shows that PrefCap-Reason is nothing and Memory-Controller-Load is maxed out.
Power-Consumption never goes above 60%(=150W).
Tesla(900GB/sec) vs Titan V(653GB/sec) = 3h vs 4h
| |
|
|
|
|
Looks like I have been running into that Tesla:
http://www.primegrid.com/workunit.php?wuid=582194078
My old 7990s are taking ~65500 sec, but at least they are doing 2 at a time. | |
|
AzmodesVolunteer tester
Send message
Joined: 30 Dec 16
Posts: 184
ID: 479275
Credit: 2,203,435,344
RAC: 1,751
        
|
|
GFN22 on an RTX 2070 FE: 107,513.65 seconds
____________
Long live the sievers.
+ Encyclopaedia Metallum: The Metal Archives + | |
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2420
ID: 1178
Credit: 20,175,556,434
RAC: 23,180,039
                                    
|
|
Some GFN21 times on CPU (all running maximum MT count):
Xeon E3-1270 (basically same as an i7-2600) - about 250,000 seconds or 70 hours
Xeon E5-1620 (similar to i7-4770 or i7-4790) - about 150,000 seconds or 42 hours
Xeon E5-1650 (similar to 6 core Haswell i7) - about 140,000 seconds or 39 hours
| |
|
Yves GallotVolunteer developer
Project scientist
Send message
Joined: 19 Aug 12
Posts: 843
ID: 164101
Credit: 306,554,426
RAC: 5,458
|
Some GFN21 times on CPU (all running maximum MT count):
Xeon E3-1270 (basically same as an i7-2600) - about 250,000 seconds or 70 hours
Xeon E5-1620 (similar to i7-4770 or i7-4790) - about 150,000 seconds or 42 hours
Xeon E5-1650 (similar to 6 core Haswell i7) - about 140,000 seconds or 39 hours
Xeon W-2125 (4-core Skylake-W @ 4 GHz) - about 87,000 seconds or 24 hours
| |
|
|
|
Some GFN21 times on CPU (all running maximum MT count):
Xeon E3-1270 (basically same as an i7-2600) - about 250,000 seconds or 70 hours
Xeon E5-1620 (similar to i7-4770 or i7-4790) - about 150,000 seconds or 42 hours
Xeon E5-1650 (similar to 6 core Haswell i7) - about 140,000 seconds or 39 hours
Xeon W-2125 (4-core Skylake-W @ 4 GHz) - about 87,000 seconds or 24 hours
OCL4 Transform
25000 seconds = (225W) RTX 2080 @ 2GHz (PCIe 3.0 x4)
40000 seconds = (220W) GTX 1080 @ 2GHz (PCIe 3.0 x8)
53000 seconds = (145W) GTX 1070 @ 2GHz (PCIe 2.0 x1)
72500 seconds = (125W) GTX 1060 3GB @ 2GHz (PCIe 3.0 x4)
78000 seconds = (170W) GTX 970 @ 1450MHz (PCIe 2.0 x1) | |
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2652
ID: 29980
Credit: 570,442,335
RAC: 5,621
        
|
|
GFN21 RTX 2070 29415s (8h10m) | |
|
|
|
|
Gigabyte Windforce 2080 OC card with FX 8350, 1866mhz RAM. GPU running at 1950Mhz, GPU memory at 1775Mhz both according to GPUZ. MSI afterburner reports the same GPU Mhz but the memory is reported at 7099Mhz.
GFN21
23,600 - 25,100 seconds.
6.56 - 6.97 hrs. | |
|
axnVolunteer developer
Send message
Joined: 29 Dec 07
Posts: 285
ID: 16874
Credit: 28,027,106
RAC: 0
           
|
|
Gigabyte 1050 Ti OC edition / Core i3-8100 CPU / Ubuntu 18.04
With GFN21 running on the GPU and 3 cores on the CPU,
GPU = 32 hr
CPU = 34 hr
Running only on CPU with all 4 cores,
CPU = 30 hr
I will be doing a testing with CPU = 4 cores and GPU running simultaneously. Will post that once I have the results. | |
|
|
|
GPU memory at 1775Mhz both according to GPUZ. MSI afterburner reports the same GPU Mhz but the memory is reported at 7099Mhz.
GPU-Z displays the real clock speed. EVGA Precision, MSI Afterburner, GPU Shark display the effective DDR speed.
1775 is the real speed, 7099 is the effective speed. Note 1775 x 4 = 7100
So it's QDR (Quad Data Rate) memory rather than DDR (Double Data Rate).
This gets confusing since GDDR6 = Graphics Double Data Rate type six synchronous dynamic random-access memory.
Maybe someone else can explain that, I might look into it later myself. | |
|
|
|
GPU memory at 1775Mhz both according to GPUZ. MSI afterburner reports the same GPU Mhz but the memory is reported at 7099Mhz.
GPU-Z displays the real clock speed. EVGA Precision, MSI Afterburner, GPU Shark display the effective DDR speed.
1775 is the real speed, 7099 is the effective speed. Note 1775 x 4 = 7100
So it's QDR (Quad Data Rate) memory rather than DDR (Double Data Rate).
This gets confusing since GDDR6 = Graphics Double Data Rate type six synchronous dynamic random-access memory.
Maybe someone else can explain that, I might look into it later myself.
Thanks. I figured it was some sort of quad memory rate being reported but wasn't sure. An older version of Afterburner reported it wrong and also didn't adjust the fans properly but upgrading to the latest beta solved those problems.
| |
|
|
|
|
GFN21 (cpu)
Macbook Pro Intel(R) Core(TM) i7-4870HQ (multi-thread 4 cores) - 39.96 hrs | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
|
|
GFN21
GTX 760 - 141,248 seconds (39 h 14 m 8 s) | |
|
mfl0pSend message
Joined: 5 Apr 09
Posts: 251
ID: 38042
Credit: 2,757,874,746
RAC: 23,158
        
|
|
GFN21, GTX 1060 6GB on a PCIe x1 link 76,000 sec | |
|
|
|
|
GFN21 CPU
i5-2500K ~203,000 seconds (2 GPU)
i7-3615QM ~203,000-210,000 (4t)
i7-3770 ~226,000 (4t) (1 GPU)
i7-3770K ~194,000 (4t) (4 GPU)
i3-4130 ~292,000 (4t) (need to check 2t no HT)
i5-4690k ~184,000 (4t) (2 GPU)
i7-4670k ~206,000 (4t) (2 GPU) (suspected memory issue)
i7-4790k ~124,000 (4t) (1 GPU)
i7-5930k ~80,000 (6t) (2 GPU)
i5-6500 ~146,000 (4t) (1 GPU)
i7-6700K ~94,000 (4t) (1 GPU)
i7-7700K ~112,000 (4t) (2 GPU)
Edit:These are times for live WUs at or near leading edge. | |
|
|
|
|
GTX 1080 @ 1.9GHz on Celeron G4900 3.1GHz free:
GFN19 ~1 h (3525 sec) Using OCL3 transform
GFN21 ~10.5 h (37690 sec) Using OCL4 transform
| |
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2652
ID: 29980
Credit: 570,442,335
RAC: 5,621
|
|
h w kWh/task
RTX 2070 OCL5 8.3 174 1.4
GTX 1080 Ti OCL4 8.5 274 2.3
GTX 1070 OCL4 13.8 150 2.1
GTX 1060 3GB OCL4 21 119 2.5
GTX 980 Ti OCL4 13 258 3.4
GTX 970 OCL4 20 164 3.3
Vega 56 OCL 13 150 2.0
RX 580 OCL 24 98 2.4
R7 280X OCL 20
h w kWh/task
i5 4570s 1333dcdr 49 49 2.4
i7 5930k 2666qcsr 24.6 117 2.9
i5 5675C (65W limit) 1866dcsr 34 65 2.2
i5 5675C (unlimited) 1866dcsr 32 79 2.5
i7 5775C 2400dcdr 30 64 1.9
i3 6100 2666dcsr 51 29 1.5
i5 6600k 2666dcdr 32.5 59 1.9
i7 6700k 3200dcdr 24.3 68 1.7
i7 6700k 3000dcsr 32 65 2.1
i7 8086k 3000dcsr 26 69 1.8
R5 1600 2666dcsr 36.2 59 2.1
R7 1800 2666dcsr 31 66 2.0
R5 2600 3000dcsr 29.7 61 1.8
Hope the formatting carries over... but anyway, done a bunch of testing. CPUs are stock, ram is probably over stock in many cases.
dc = dual channel
qc = quad channel
sr = single rank
dr = dual rank
Times are in hours estimated for -q "304208^2097152+1" which was leading edge at the time I looked. For GPU tests I left them a short time to warm up before taking the estimate as they all downclock as they get hotter. I can't say for sure I left it long enough for it to stop slowing.
Power is as reported in software. It should be noted the different scenarios are NOT comparable with each other. That is, we have 4 cases: nvidia GPU, AMD GPU, AMD CPU, Intel CPU.
nvidia GPU - I understand this is the total power taken by the card
AMD GPU - I understand this is only the power taken by the GPU, so doesn't include ram, VRM efficiency etc.
AMD CPU - this is the reported core+SoC total power
Intel CPU - this is the reported socket power
I also worked out kWh/unit as a kind of performance per watt metric, which will obviously still be limited by how those watts are reported.
General observations:
On nvidia GPUs, Turing is more efficient than Pascal, which is more efficient than Maxwell. Not really any surprise there. The Turing test (pun not intended) picked OCL5, whereas Pascal/Maxwell picked OCL4.
On AMD GPUs, the ones I have at least all picked OCL transform. 280X is faster than 580X, but it didn't report power so I don't know how well it does there. Vega did well in general.
On Intel CPUs, it isn't a big surprise but it seems many of these CPUs are held back by ram bandwidth. Look at the two 6700ks for example. One takes about 30% longer due to the slightly slower ram, and not having dual rank. It is unclear if Broadwell L4 is helping here, but I'm leaning towards it is comparing against the 6600k. Note the 5675C was tested twice, as the mobo used has a bug that means it hard limits to 65W TDP and I can only relax it via software after boot.
On AMD CPUs, feels like a similar story to Intel, in that I could use better ram in those systems. | |
|
|
|
|
Thanks!
P.S. GTX 1080 OCL4 10.5 208 2.2 | |
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2652
ID: 29980
Credit: 570,442,335
RAC: 5,621
|
|
7800X 18.0h 3000qcsr
CPU software power reporting was broken. From wall, system was 95W idle, 225W under load. | |
|
|
|
|
32% is done. Gfn 22 gpu load jump to 0 is too many times on 5 minutes. My 12 volt is not absolutely stable and jumping between 12.1 and 12. This is not very healthy for hdd. Only that I don't like in gfn 22. | |
|
|
|
|
I'm letting a gfn-21 run an a RX570. lets look how llong it will takes
____________
 | |
|
|
|
|
Gigabyte RTX 2080 for GFN22
Clock: 1965MHz
Memory: 7099MHz
Temps hovered around 79C with 90%+ fan speed
GPU Usage 98% as shown by GPU-Z
Memory Controller in the 80-85% range
TDP a little less than 100% then slightly over 100% so close to 100% with fluctuation. Around 240 watts
CPU FX-8350 while running other stuff
GFN22 time first run on the 2080 - 87,864.97 seconds or 24.4 hrs | |
|
|
|
|
Van Zimmerman
i7-5930k ~80,000 (6t) (2 GPU)
That's faster than my RTX 2080 | |
|
AzmodesVolunteer tester
Send message
Joined: 30 Dec 16
Posts: 184
ID: 479275
Credit: 2,203,435,344
RAC: 1,751
|
|
Can't be. :P Those are GFN21 times, by the way.
____________
Long live the sievers.
+ Encyclopaedia Metallum: The Metal Archives + | |
|
|
|
Can't be. :P Those are GFN21 times, by the way.
ah hah! Makes sense now. I was wondering how in the world that could be.... I forgot you can't multithread or run GFN22 on CPU even.
Thanks for clearing that up Azmodes :) | |
|
|
|
I'm letting a gfn-21 run an a RX570. lets look how llong it will takes
done.
runtime of GFN 21 on a ASUS RX570 Strix: 94,473.55 sec
____________
| |
|
|
|
|
The upcoming Radeon VII should be beastly at GFN21 and GFN22. Nearly 7tflops of FP64 for $699. Might be able to beat a Titan V since it has over 50% more memory bandwidth. | |
|
|
|
|
Some AMD numbers...
RX480
GFN 21: 85,636.77 sec
GFN 22: 328,450.69 sec
Vega 64 LC
GFN 21: 38,716.65 sec
GFN 22: 157,553.05 sec
Waiting for Radeon VII...
| |
|
|
|
Some AMD numbers...
RX480
GFN 21: 85,636.77 sec
GFN 22: 328,450.69 sec
Vega 64 LC
GFN 21: 38,716.65 sec
GFN 22: 157,553.05 sec
Waiting for Radeon VII...
Looks like the rumor that the VII would have an uncapped FP64 was incorrect. RIP... | |
|
|
|
Looks like the rumor that the VII would have an uncapped FP64 was incorrect. RIP...
I just read that :(
But I’ll probably buy one any way while waiting for Zen 3....
| |
|
|
|
|
I think I may have found the slowest system out there processing GFN 21.
Take a look at my wingman's GPU. Any bets on how long this will take? Haha.
https://www.primegrid.com/workunit.php?wuid=590986499
To make matters worse, he's also Sieving 56 tasks too. | |
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 990,114,734
RAC: 55,798
   
|
I think I may have found the slowest system out there processing GFN 21.
Take a look at my wingman's GPU. Any bets on how long this will take? Haha.
https://www.primegrid.com/workunit.php?wuid=590986499
To make matters worse, he's also Sieving 56 tasks too.
He'll finish tomorrow (January 17). If we didn't want CPUs running GFN21, we wouldn't have configured the system to send them out. | |
|
Yeti
Send message
Joined: 2 Nov 05
Posts: 6
ID: 1294
Credit: 1,205,499,000
RAC: 50,390
       
|
Gigabyte RTX 2080 for GFN22
Clock: 1965MHz
Memory: 7099MHz
Temps hovered around 79C with 90%+ fan speed
GPU Usage 98% as shown by GPU-Z
Memory Controller in the 80-85% range
TDP a little less than 100% then slightly over 100% so close to 100% with fluctuation. Around 240 watts
CPU FX-8350 while running other stuff
GFN22 time first run on the 2080 - 87,864.97 seconds or 24.4 hrs
Mine are not so hot: 74° with 69% FanSpeed
85.474 - 89.300, with most of them at 88.xxx seconds
Palit 2080 Gamerock Premium
____________
Supporting BOINC, a great concept ! | |
|
|
|
|
GNF22 by GTX 1060 3GB
Running on platform 'NVIDIA CUDA', device 'GeForce GTX 1060 3GB', vendor 'NVIDIA Corporation', version 'OpenCL 1.2 CUDA' and driver '396.54'.
9 computeUnits @ 1835MHz, memSize=3018MB, cacheSize=144kB, cacheLineSize=128B, localMemSize=48kB, maxWorkGroupSize=1024.
Time: 288,834s ~ 80h14m
Sixth to receive, first to complete. :)
https://postimg.cc/tnHVW4tR
____________
My DC mathematical side :)
 | |
|
|
|
|
@Luigi, a tip regarding image links.
PG doesn't like to embed some imges links with https.
You can try removing the s from https in the image link.
It works for images hosted at postimage.org.
I used this:
[img]http://i.postimg.cc/sjcnh7yT/my-gfn22.png[/img]
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
|
I didn't want to show image directly because it's very large.
____________
My DC mathematical side :)
| |
|
|
|
|
Luigi R. wrote: I didn't want to show image directly because it's very large.
Oh, OK then.
____________
"Accidit in puncto, quod non contingit in anno."
Something that does not occur in a year may, perchance, happen in a moment. | |
|
|
|
|
GNF21 by GTX 1060 3GB
Running on platform 'NVIDIA CUDA', device 'GeForce GTX 1060 3GB', vendor 'NVIDIA Corporation', version 'OpenCL 1.2 CUDA' and driver '396.54'.
9 computeUnits @ 1835MHz, memSize=3018MB, cacheSize=144kB, cacheLineSize=128B, localMemSize=48kB, maxWorkGroupSize=1024.
Time: 78,402s ~ 21h47m
____________
My DC mathematical side :)
| |
|
|
|
|
GFN21 with Radeon VII:
Running on platform 'AMD Accelerated Parallel Processing', device 'gfx906', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2766.5)' and driver '2766.5 (PAL,HSAIL)'.
60 computeUnits @ 1802MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256.
Run time: 21,911.91 (~6h05m)
____________
Found prime:
12982084138455*2^666666-1 Sophie Germain Prime Search
3988720718715*2^1290000-1 Sophie Germain Prime Search | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
|
|
What's going on?
I have one task each of GFN 22 and DYFL.
They have IDENTICAL runtime and CPU time, right to the 100ths of a second.
Not only that, all the times are composed of the digit 4 to the left of the decimal point.
The chance of this happening is astronomically small. Something's fishy.
https://www.primegrid.com/results.php?hostid=966448&offset=0&show_names=0&state=4&appid=
EDIT: It's a borrowed host attached with my weak key, so I can't see what's going on there. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14044
ID: 53948
Credit: 482,370,180
RAC: 568,853
|
What's going on?
I have one task each of GFN 22 and DYFL.
They have IDENTICAL runtime and CPU time, right to the 100ths of a second.
Not only that, all the times are composed of the digit 4 to the left of the decimal point.
The chance of this happening is astronomically small. Something's fishy.
https://www.primegrid.com/results.php?hostid=966448&offset=0&show_names=0&state=4&appid=
EDIT: It's a borrowed host attached with my weak key, so I can't see what's going on there.
The 4444 is obviously faked. There's also no stderr text. What happened is that the host completed the tasks and uploaded the result file, but didn't for whatever reason actually report the result as completed. Left to its own normal processing, BOINC would treat this as "no response", and at the end of the deadline the task would be dead and a new task would be sent to a new host.
We don't think that outcome is fair to you, and from the project's perspective it's a waste of computing time.
We have a process that periodically looks for tasks such as those and manually marks them as "completed". The tasks then get credit if the result passes validation, and everyone wins. We don't, however, have any of the information your BOINC client would send to the server, so we have to make it up. The elapsed time is set to 4444 to make it obvious that we filled in the data.
____________
My lucky number is 75898524288+1 | |
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 990,114,734
RAC: 55,798
|
|
You are correct, it's not normal. Those jobs uploaded but then when they tried to report the system decided that the wrong host was reporting them. They were resurrected by a cron job that runs once an hour. The 44444 time is because we don't know how long it took and is meant to stand out so that we don't go trying to figure out why the time on that workunit is so wrong. It also means we can screen those out of any averages about how long those jobs take.
$ grep 609110193 *.log
scheduler.log:2019-04-28 13:32:58.5301 [PID=31636] [CRITICAL] [HOST#966412] [RESULT#995940844] [WU#609110193] got result from wrong host; expected [HOST#966367]
scheduler.log:2019-04-28 13:32:58.5304 [PID=31636] [CRITICAL] [USER#26495] [HOST#966412] [RESULT#995940844] [WU#609110193] Not even the same user; expected [USER#55391]
unreported.log:2019-04-29 14:24:46 Workunit 609110193, result genefer22_24110274_3 was abandoned but had an upload. Now fixed.
$ grep 609935887 *.log
scheduler.log:2019-04-29 20:44:39.3904 [PID=10450] [CRITICAL] [HOST#966412] [RESULT#995117283] [WU#609935887] got result from wrong host; expected [HOST#966448]
scheduler.log:2019-04-29 20:44:39.3905 [PID=10450] [CRITICAL] [USER#26495] [HOST#966412] [RESULT#995117283] [WU#609935887] Not even the same user; expected [USER#55391]
unreported.log:2019-04-30 21:24:43 Workunit 609935887, result genefer_extreme_24115859_1 was abandoned but had an upload. Now fixed.
Not sure what was going on with that host. Was it possibly signed into PrimeGrid twice with different keys? It expected the user to be you, but instead it was Tornlogic. That's probably his hostid for that same machine. | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
|
Not sure what was going on with that host. Was it possibly signed into PrimeGrid twice with different keys? It expected the user to be you, but instead it was Tornlogic. That's probably his hostid for that same machine.
It's his machine, and he was fiddling with the BOINC configuration while my jobs were running on it because I had loaded it up incorrectly. It was running for a couple of days and he didn't want to abort the long jobs, so they completed with a weird configuration. He thought the results were lost because he didn't see them on my tasks list but apparently they were reported and the PG server cron job put them in my list. | |
|
|
|
|
Hey everyone. Yea, so this is my machine, and essentially my F up. I attached composite's weak key to that client using the command line method. Worked great, the client began crunching for him.
Once I got an understanding of which tasks composite was sending to my client, I then made a change to my app_config file to configure multithreading for the tasks he was running. Oddly enough when I went to make changes to the app_config file, it simply was gone. I restored a backup of that file, and that's when everything went crazy.
It then detached composite from my client, all tasks got messed up as indicated below, and I'm still scratching my head on what's going to happen with the 8 SoB tasks that the client is processing now for composite. I do not have "No New Tasks" enabled, (double negative, sorry) and it's not getting new tasks. (triple negative!!) In other words, I'm trying to get new tasks for the GPU, but nothing.. It was getting "Do you Feel Lucky" tasks, but quit once I restored the app_config.xml file. My GPU is sitting idle right now, kinda sad. I'm going to let it finish the 8 SoB tasks, and try to fix the problem.
Is there a way to get composite's weak key associated with my client AND get multithreading enabled for his SoB tasks? I just went into BIOS and disabled hyperthreading as a work around, but still not crunching at optimum speeds.
Thanks,
Drew | |
|
compositeVolunteer tester
Send message
Joined: 16 Feb 10
Posts: 1172
ID: 55391
Credit: 1,221,586,795
RAC: 1,513,610
|
...I then made a change to my app_config file to configure multithreading for the tasks he was running. Oddly enough when I went to make changes to the app_config file, it simply was gone.
I think that's what happens to running tasks when you change app_config.xml on the fly. You need to let tasks complete or abort them so that they don't get lost. Correct?
As for not getting work, it might be because I merged the old machine id and the new one under my account. So if your machine has gotten the old id again, then it's not associated with any account and it's effectively in the twilight zone. The remedy I suspect is to detach the machine from primegrid and reattach it with my weak key (after the SoB tasks are done). | |
|
Message boards :
Generalized Fermat Prime Search :
GFN21 GFN22 Scoreboard, post your times/hardware |
