few results completed (45-49k seconds), thats still 3-4h longer than the Haswell based Xeon E3-1220 v3 and E5-1650 v3 wingmen from ATLAS AEI Hannover
Their computer info page shows only 4 cores for the E3 and 6 for the E5 though. The first is a non-SMT CPU, while on the second SMT was most likely disabled (or ncpus=6 set in BOINC).
Keeping that in mind, the run times seem reasonable.
That is slower than I expected. We don't set ncpus so the number of cores you see is the actual number of threads used. It would be nice to see what effect SMT has on the runtime on the Ryzen.
My i7-4770 running under Ubuntu 16.10 was doing the Multi-Directed Continuous Gravitational Wave search CV v1.00 (AVX) in about 45 to 60 k seconds. All 8 cores were used for some BOINC/Folding project, but the variation in the type may have produced changes in the load that account for at least some of the variation.
So I think the Ryzen is still a little faster per core, and with more cores you should still get at least a factor of 2 improvement. I don't know if there are any optimizations that can help the Ryzen AVX do still better, but even now it is a nice improvement.
Using strace, I've seen it fails to open ./EAH_NO_SYNC, ./EAH_DEBUG_LEVEL, ./EAH_DEBUG_DDD and .EAH_DEBUG_GDB. Same for earth.lnk, sun.lnk, segments.lnk, but I had copied eg. the original earth over the link file and I see it being opened elsewhere successfully (O_RDONLY).
Only remaining fail is the below one, for which no file exists in /var/lib/boinc-client/projects:
/einstein.phys.uwm.eduprojects/einstein.phys.uwm.edu
stat("../../projects/einstein.phys.uwm.edu/h1_1479.90_O1C02Cl1In0C__O1MD1CV_CasA1_1480.65Hz_193_2_0", 0x7ffdcc686100) = -1 ENOENT (No such file or directory)
Might that be connected to the early termination? Any hint how to start this in standalone mode?
Your computer Ryzen1700 works with NVidia 1060 - those 140W with AVX application is for whole computer, but GPU is not crunching? And this is for 3,7 GHz for all cores with auto voltage?
With the GTX1060 3GB doing FGRP and the CPU a mix of Einstein, WCG and Sixtrack I am seeing ~210-220W usage right now. Without GPU ~150W. Isn't Einstein loading the CPU as much as others or did I capture the 140W in a lucky moment? Not sure, will keep an eye on it...
THX - what power supply do You use? I mean power efficiency - bronze/silver etc.
My present I5 3450@ 2800@0,9V 3 cores in use + GPU idle consumes ~60W. With GPU (1060 3GB) with power limited to 50% consumes ~ 130-140W with power supply rated bronze. I'm looking for upgrade of CPU - but must be low power consumption. AMD 1700 looks very promising.
Results for the standalone 8 vs. 16 thread comparison on a random WU are in. Running 8 threads they finished in 31933 seconds (average), running 16 threads completion times went up to 60740 seconds.
A small 10% throughput gain, less than anything I have seen on other projects to date.
Most of my WUs complete in shorter time though (<50k seconds). Comparing their times against the wing mates (most are ATLAS Haswell Xeon), they are of course usually 25-40% slower than the non-SMT Intels, but run 16 and not 4-6 jobs at once. Without SMT the R7 should outperform these Xeon and still run more WUs in parallel.
koschi wrote:few results
)
That is slower than I expected. We don't set ncpus so the number of cores you see is the actual number of threads used. It would be nice to see what effect SMT has on the runtime on the Ryzen.
https://einsteinathome.org/ho
)
https://einsteinathome.org/host/12437926/tasks?order=granted_credit&sort=desc
One out of many ATLAS Xeon E3-1231 v3 (Haswell, 4c/8t, 3.4-3.8GHz), showing about the same or tad better run times then my R7 @ 3.6GHz.
https://einsteinathome.org/host/12481178
An E5-1650 v3 (6c/12t) with just 6 BOINC'ing cores. Maybe SMT is disabled in BIOS or the custom Linux 3.18.25-atlas kernel?
edit:
I will do new SMT/non-SMT tests during next days and also include Einstein. Will keep you updated.
My i7-4770 running under
)
My i7-4770 running under Ubuntu 16.10 was doing the Multi-Directed Continuous Gravitational Wave search CV v1.00 (AVX) in about 45 to 60 k seconds. All 8 cores were used for some BOINC/Folding project, but the variation in the type may have produced changes in the load that account for at least some of the variation.
https://einsteinathome.org/host/12360156/tasks
So I think the Ryzen is still a little faster per core, and with more cores you should still get at least a factor of 2 improvement. I don't know if there are any optimizations that can help the Ryzen AVX do still better, but even now it is a nice improvement.
Power usage is ~140W at the
)
Power usage is ~140W at the outlet running these AVX tasks...
edit:
when trying to run the app manually, it terminates few seconds after showing "[normal]: Reading input data ...".
stderr.out shows "BOINC client no longer exists - exiting"
I called the app like this:
./einstein_O1MD1CV_1.00_x86_64-pc-linux-gnu__AVX --Freq=1480.65 --FreqBand=0.05 --dFreq=6.847512e-07 --f1dot=-1.018133e-07 --f1dotBand=1.412647e-09 --df1dot=3.881809e-12 --gammaRefine=4 --f2dot=0 --f2dotBand=6.826924e-17 --df2dot=4.033716e-18 --gamma2Refine=20 --computeBSGL --BSGLlogcorr=0 --Fstar0=65.826 --oLGX=0.5,0.5 --nCand1=10000 --SortToplist=6 --recalcToplistStats=1 -o GCT.out --printCand1 --semiCohToplist --ephemE=earth --ephemS=sun --segmentList=segments --FstatMethod=ResampBest --FstatMethodRecalc=DemodBest --gridType=3 --skyGridFile={6.123770423960928 1.026457803695149} --loudestSegOutput --getMaxFperSeg --peakThrF=2.6 --DataFiles1="h1_1479.90_O1C02Cl1In0C.y4ta;l1_1479.90_O1C02Cl1In0C.y4ta;h1_1479.95_O1C02Cl1In0C.C3rE;l1_1479.95_O1C02Cl1In0C.C3rE;h1_1480.00_O1C02Cl1In0C.NMj8;l1_1480.00_O1C02Cl1In0C.NMj8;h1_1480.05_O1C02Cl1In0C.zDNG;l1_1480.05_O1C02Cl1In0C.zDNG;h1_1480.10_O1C02Cl1In0C.FcN3;l1_1480.10_O1C02Cl1In0C.FcN3;h1_1480.15_O1C02Cl1In0C.OQNY;l1_1480.15_O1C02Cl1In0C.OQNY;h1_1480.20_O1C02Cl1In0C.Uhqc;l1_1480.20_O1C02Cl1In0C.Uhqc;h1_1480.25_O1C02Cl1In0C.pa5z;l1_1480.25_O1C02Cl1In0C.pa5z;h1_1480.30_O1C02Cl1In0C.RHkm;l1_1480.30_O1C02Cl1In0C.RHkm;h1_1480.35_O1C02Cl1In0C.Fq38;l1_1480.35_O1C02Cl1In0C.Fq38;h1_1480.40_O1C02Cl1In0C.oIjY;l1_1480.40_O1C02Cl1In0C.oIjY;h1_1480.45_O1C02Cl1In0C.GQdZ;l1_1480.45_O1C02Cl1In0C.GQdZ;h1_1480.50_O1C02Cl1In0C.u17X;l1_1480.50_O1C02Cl1In0C.u17X;h1_1480.55_O1C02Cl1In0C.SHW3;l1_1480.55_O1C02Cl1In0C.SHW3;h1_1480.60_O1C02Cl1In0C.nkiR;l1_1480.60_O1C02Cl1In0C.nkiR;h1_1480.65_O1C02Cl1In0C.rDyZ;l1_1480.65_O1C02Cl1In0C.rDyZ;h1_1480.70_O1C02Cl1In0C.BcMr;l1_1480.70_O1C02Cl1In0C.BcMr;h1_1480.75_O1C02Cl1In0C.XzhS;l1_1480.75_O1C02Cl1In0C.XzhS;h1_1480.80_O1C02Cl1In0C.AQeV;l1_1480.80_O1C02Cl1In0C.AQeV;h1_1480.85_O1C02Cl1In0C.r7-M;l1_1480.85_O1C02Cl1In0C.r7-M;h1_1480.90_O1C02Cl1In0C.iDCl;l1_1480.90_O1C02Cl1In0C.iDCl;h1_1480.95_O1C02Cl1In0C.fUWY;l1_1480.95_O1C02Cl1In0C.fUWY;h1_1481.00_O1C02Cl1In0C.8-wr;l1_1481.00_O1C02Cl1In0C.8-wr;h1_1481.05_O1C02Cl1In0C.1WEc;l1_1481.05_O1C02Cl1In0C.1WEc;h1_1481.10_O1C02Cl1In0C.LGGp;l1_1481.10_O1C02Cl1In0C.LGGp;h1_1481.15_O1C02Cl1In0C.evKM;l1_1481.15_O1C02Cl1In0C.evKM;h1_1481.20_O1C02Cl1In0C.68k2;l1_1481.20_O1C02Cl1In0C.68k2;h1_1481.25_O1C02Cl1In0C.QVUP;l1_1481.25_O1C02Cl1In0C.QVUP;h1_1481.30_O1C02Cl1In0C.rU6P;l1_1481.30_O1C02Cl1In0C.rU6P;h1_1481.35_O1C02Cl1In0C.daQ1;l1_1481.35_O1C02Cl1In0C.daQ1;h1_1481.40_O1C02Cl1In0C.eyEm;l1_1481.40_O1C02Cl1In0C.eyEm;h1_1481.45_O1C02Cl1In0C.1hDb;l1_1481.45_O1C02Cl1In0C.1hDb" --WUfpops=1.44e+14Using strace, I've seen it fails to open ./EAH_NO_SYNC, ./EAH_DEBUG_LEVEL, ./EAH_DEBUG_DDD and .EAH_DEBUG_GDB. Same for earth.lnk, sun.lnk, segments.lnk, but I had copied eg. the original earth over the link file and I see it being opened elsewhere successfully (O_RDONLY).
Only remaining fail is the below one, for which no file exists in /var/lib/boinc-client/projects:
/einstein.phys.uwm.eduprojects/einstein.phys.uwm.edu
stat("../../projects/einstein.phys.uwm.edu/h1_1479.90_O1C02Cl1In0C__O1MD1CV_CasA1_1480.65Hz_193_2_0", 0x7ffdcc686100) = -1 ENOENT (No such file or directory)
Might that be connected to the early termination? Any hint how to start this in standalone mode?
@Koschi Your computer
)
@Koschi
Your computer Ryzen1700 works with NVidia 1060 - those 140W with AVX application is for whole computer, but GPU is not crunching? And this is for 3,7 GHz for all cores with auto voltage?
Yes, whole system, but GPU
)
Yes, whole system, but GPU idle...
I am running the R7 at 3.6GHz/1.1875V currently.
With the GTX1060 3GB doing FGRP and the CPU a mix of Einstein, WCG and Sixtrack I am seeing ~210-220W usage right now. Without GPU ~150W. Isn't Einstein loading the CPU as much as others or did I capture the 140W in a lucky moment? Not sure, will keep an eye on it...
Koschi: I send you a PM to
)
Koschi: I send you a PM to discuss the standalone mode issue.
THX - what power supply do
)
THX - what power supply do You use? I mean power efficiency - bronze/silver etc.
My present I5 3450@ 2800@0,9V 3 cores in use + GPU idle consumes ~60W. With GPU (1060 3GB) with power limited to 50% consumes ~ 130-140W with power supply rated bronze. I'm looking for upgrade of CPU - but must be low power consumption. AMD 1700 looks very promising.
It's an Enermax Pro 82+
)
It's an Enermax Pro 82+ EPR425AWT (bronze), most likely 8 years old by now.
I think I've seen few people posting they run their R7 at 3GHz under 1V, which should be pretty efficient.
edit:
now 3.6GHz @ 1,1625V (BIOS) the consumption monitor shows max. 125W. These tasks consume a lot less power than other projects (WCG mix = +20W)...
Results for the standalone 8
)
Results for the standalone 8 vs. 16 thread comparison on a random WU are in. Running 8 threads they finished in 31933 seconds (average), running 16 threads completion times went up to 60740 seconds.
A small 10% throughput gain, less than anything I have seen on other projects to date.
Most of my WUs complete in shorter time though (<50k seconds). Comparing their times against the wing mates (most are ATLAS Haswell Xeon), they are of course usually 25-40% slower than the non-SMT Intels, but run 16 and not 4-6 jobs at once. Without SMT the R7 should outperform these Xeon and still run more WUs in parallel.