This is the same application as before. We are just preparing to transition to a new one which is incompatible with the current one. It has nothing to do with the dataset.
My initial run of 1.03 Gamma under Win7 64-bit on 7 cores of an i7-4771 (and one core supporting two GTX 750 Ti's on BRP4G/BRP6) is going much more slowly. After six hours, they are projected to run for 32 hours.
It may be that the work units are larger than the cache on the Haswell chip. On my next run, I will reduce it to 4 cores, and see what happens. That is how I usually run this machine anyway, and they might do as well or better than Linux at that point. Or else maybe there are different sizes of work unit?
The 1.03 Gamma under Win7 64-bit actually finished in 9 hours 10 minutes, so the percentage complete is just not very accurate for these work units. Considering the slight speed difference between the i7-4771 Windows machine and the i7-4790 Linux machine (3.7/3.8 GHz), these work units would run about 8 hours 55 minutes on an i7-4790 Windows machine. That is not bad at all compared to Linux, and since I will be running them in practice on only 4 cores, they should be much faster.
> but under Linux Mint host 12241921 (4xGamma, no GPU) > 1.01: ~4:25:00 > 1.03: ~ 6:36:00 - 2 hours longer.
My host 12247194 under Windows 7 Pro (Dual Boot, host 12241921 under Linux) finished 4 x FGRPSSE 1.03 in ~ 5:20:00, one hour faster. No times for 1.01.
My host 12247194 under Windows 7 Pro (Dual Boot, host 12241921 under Linux) finished 4 x FGRPSSE 1.03 in ~ 5:20:00, one hour faster. No times for 1.01.
Running on only 4 cores of an i7-4771 (3.7 GHz), plus supporting two GTX 750 Ti's on BRP4G, I am now getting 6 hours 40 minutes for 1.03 Gamma (FGRPSSE). I think that is almost as fast as what Linux would produce on that machine, but the question is why isn't it faster as compared to your Xeon E3-1225 v3 @ 3.20GHz?
I will shut down the GPU work and turn off hyper-threading in the BIOS to see if it speeds things up, but it could take a couple of days.
This is the same application
)
This is the same application as before. We are just preparing to transition to a new one which is incompatible with the current one. It has nothing to do with the dataset.
OK, thanks.
)
OK, thanks.
Hi, I have calculated just
)
Hi,
I have calculated just the first FGRPSSE WU's (#1.03).
Runtime on Windows7 host 6742381 (2xGamma+2xParkes) is:
1.01: ~4:05:00
1,03: ~ 4:11:00
but under Linux Mint host 12241921 (4xGamma, no GPU)
1.01: ~4:25:00
1.03: ~ 6:36:00 - 2 hours longer.
is this intended?
Interesting. I see the same
)
Interesting. I see the same thing on my Ubuntu 16.04 machine running on seven cores of an i7-4790, with one core supporting a GPU.
1.01: 6 hours 40 minutes
1.03: 8 hours 30 minutes
I will be looking at Windows later.
My initial run of 1.03 Gamma
)
My initial run of 1.03 Gamma under Win7 64-bit on 7 cores of an i7-4771 (and one core supporting two GTX 750 Ti's on BRP4G/BRP6) is going much more slowly. After six hours, they are projected to run for 32 hours.
It may be that the work units are larger than the cache on the Haswell chip. On my next run, I will reduce it to 4 cores, and see what happens. That is how I usually run this machine anyway, and they might do as well or better than Linux at that point. Or else maybe there are different sizes of work unit?
The 1.03 Gamma under Win7
)
The 1.03 Gamma under Win7 64-bit actually finished in 9 hours 10 minutes, so the percentage complete is just not very accurate for these work units. Considering the slight speed difference between the i7-4771 Windows machine and the i7-4790 Linux machine (3.7/3.8 GHz), these work units would run about 8 hours 55 minutes on an i7-4790 Windows machine. That is not bad at all compared to Linux, and since I will be running them in practice on only 4 cores, they should be much faster.
More observations: > but
)
More observations:
> but under Linux Mint host 12241921 (4xGamma, no GPU)
> 1.01: ~4:25:00
> 1.03: ~ 6:36:00 - 2 hours longer.
My host 12247194 under Windows 7 Pro (Dual Boot, host 12241921 under Linux) finished 4 x FGRPSSE 1.03 in ~ 5:20:00, one hour faster. No times for 1.01.
Strange.
DF1DX wrote:My host 12247194
)
Running on only 4 cores of an i7-4771 (3.7 GHz), plus supporting two GTX 750 Ti's on BRP4G, I am now getting 6 hours 40 minutes for 1.03 Gamma (FGRPSSE). I think that is almost as fast as what Linux would produce on that machine, but the question is why isn't it faster as compared to your Xeon E3-1225 v3 @ 3.20GHz?
I will shut down the GPU work and turn off hyper-threading in the BIOS to see if it speeds things up, but it could take a couple of days.
On this host
)
On this Xeon E5-2660 host 12242223
Application N Average Median Mininum Maximum StdDeviationFGRPB1v1.00 6427 37969.5 38040.67 16175.79 41602.40 558.528
FGRPOLD 68 37920.9 38093 36846.11 38321.14 394.326
FGRPSSE 14 42919.7 42947.6 42673.05 43098.47 127.486
Shows a significant difference. All times in seconds.
Edit: On this i7-860 host 4918234
Application N Average Median Mininum Maximum StdDeviation
FGRPBv1.00 1342 28098.8 28053.5 22377.86 43664.31 2145.55
FGRPOLD 28 28108.9 27526.3 23603.95 32379.21 2375.66
FGRPSSE 2 33859.1 33859.1 33054.91 34663.38 804.235
Shows a significant difference. All times in seconds.
On this host FGRPB1 tasks
)
On this host FGRPB1 tasks used to take about 10 hours. The first 1.03 tasks are now at 15 hours and still running.