I wish they would include a Distributed Computing test in these kinds of things, pick a project, both with and without DP, and run them head to head on a test unit.
I wish they would include a Distributed Computing test in these kinds of things, pick a project, both with and without DP, and run them head to head on a test unit.
Hmm, i can see why they don't. Benchmarks needs to be reproducible, we might want a project app to be a "benchmark", but some difficulties - as we all know,
each work unit is different, good benchmarks must be exactly the same, you should always get the same results if you run it next week.
work units can take a very long time to run, good benchmarks should be reasonably quick, so can be repeatedly run to get a significant number of results.
several options for tuning exist (such as x2), and this would need automating for good benchmarking.
I think i might ask the application developers, "What benchmark application(s) closest match each science application?" and then see if they are tested and maybe ask if they are not.
Source? Sounds like a rumor mill number to me. Nvidia has not been making official release date disclosures anywhere near that far in advance for the Pascal cards.
Apparently it was a rumor, and a bad one - the card has been released by now. Without reviews yet, but its performance is going to be somewhere between 11 and 0% slower than a GTX 1060 6 GB, so nothing to worry or get excited about.
@Mikey & AgentB: creating BOINC / DC benchmarks would be a hassle. But easier than you think, if the app can run in stand alone mode. E.g. for Einstein you can grab the contents of a WU running in a slot and run that WU over and over again as benchmark. Someone posted a very short benchmark task, which I have used to check whether the Intel Skylake driver still yields wrong results for Einstein (yes, it does).
I wish they would include a Distributed Computing test in these kinds of things, pick a project, both with and without DP, and run them head to head on a test unit.
Hmm, i can see why they don't. Benchmarks needs to be reproducible, we might want a project app to be a "benchmark", but some difficulties - as we all know,
each work unit is different, good benchmarks must be exactly the same, you should always get the same results if you run it next week.
work units can take a very long time to run, good benchmarks should be reasonably quick, so can be repeatedly run to get a significant number of results.
several options for tuning exist (such as x2), and this would need automating for good benchmarking.
I think i might ask the application developers, "What benchmark application(s) closest match each science application?" and then see if they are tested and maybe ask if they are not.
I started running my 1060 6GB a couple of days ago and it seems to be working well so far. I'm just running 1 Einstein task at a time right now and it's doing about 1credit/sec on Parkes XT work units. Core load has been in the high 70% - low 80% range. I haven't bothered with a custom fan curve yet and temps have been around 60C.
Any of the recent video card reviews from the last couple of years on Anandtech web site always does a Compute Benchmark which includes for instance both Single and Double Precision test comparisons of Folding@Home work on the video card on review. They have benchmark comparisons for cards going back to the Nvidia 600 series and AMD 7000 series.
Any of the recent video card reviews from the last couple of years on Anandtech web site always does a Compute Benchmark which includes for instance both Single and Double Precision test comparisons of Folding@Home work on the video card on review. They have benchmark comparisons for cards going back to the Nvidia 600 series and AMD 7000 series.
@Keith: correct, but this doens't tell you anything but the performance at Folding@Home. If you want to know how CPUs perform in Photoshop, you wouldn't benchmark them in Corel PhotoPaint, would you?
Hi MrS, Not a problem, I run
)
Hi MrS,
Not a problem, I run those rigs only between 0100 hrs and 0800 hrs [cheap rate]:-)
I have an economy 7 tariff, so any power used after 0100 for 7 hrs is at a lower cost.
Regards,
Cliff
Cliff,
Been there, Done that, Still no damm T Shirt.
Recently published at
)
Recently published at Phoronix on Ubuntu 16.04 LTS
16-Way OpenCL Compute Comparison Of The Latest Polaris & Pascal GPUs
AgentB wrote:Recently
)
I wish they would include a Distributed Computing test in these kinds of things, pick a project, both with and without DP, and run them head to head on a test unit.
mikey wrote:I wish they would
)
Hmm, i can see why they don't. Benchmarks needs to be reproducible, we might want a project app to be a "benchmark", but some difficulties - as we all know,
I think i might ask the application developers, "What benchmark application(s) closest match each science application?" and then see if they are tested and maybe ask if they are not.
archae86 wrote:Source? Sounds
)
Apparently it was a rumor, and a bad one - the card has been released by now. Without reviews yet, but its performance is going to be somewhere between 11 and 0% slower than a GTX 1060 6 GB, so nothing to worry or get excited about.
@Mikey & AgentB: creating BOINC / DC benchmarks would be a hassle. But easier than you think, if the app can run in stand alone mode. E.g. for Einstein you can grab the contents of a WU running in a slot and run that WU over and over again as benchmark. Someone posted a very short benchmark task, which I have used to check whether the Intel Skylake driver still yields wrong results for Einstein (yes, it does).
MrS
Scanning for our furry friends since Jan 2002
AgentB wrote:mikey wrote:I
)
That works too!!
I started running my 1060 6GB
)
I started running my 1060 6GB a couple of days ago and it seems to be working well so far. I'm just running 1 Einstein task at a time right now and it's doing about 1credit/sec on Parkes XT work units. Core load has been in the high 70% - low 80% range. I haven't bothered with a custom fan curve yet and temps have been around 60C.
Any of the recent video card
)
Any of the recent video card reviews from the last couple of years on Anandtech web site always does a Compute Benchmark which includes for instance both Single and Double Precision test comparisons of Folding@Home work on the video card on review. They have benchmark comparisons for cards going back to the Nvidia 600 series and AMD 7000 series.
Keith Myers wrote:Any of the
)
Thanks I will look!!!
@Keith: correct, but this
)
@Keith: correct, but this doens't tell you anything but the performance at Folding@Home. If you want to know how CPUs perform in Photoshop, you wouldn't benchmark them in Corel PhotoPaint, would you?
MrS
Scanning for our furry friends since Jan 2002