Folding Forum

Beside all the technical KPIs like number of client or GigaTerraPeta FLOPs couldn't/shouldn't we add a "business language" defined KPI like nano/milli seconds of folding done per day or per WU.
And based on finished WU also feedback that to donors. It might be indirect part of the calibrated base points but getting that number into day light would be nice.
It might be a bit more complicated given the number of atoms in each WU is different; leave that up to the experts.

ChristianVirtual wrote:It might be a bit more complicated given the number of atoms in each WU is different; leave that up to the experts.

That's what I like about the base point idea. It equalizes everything for everyone across all platforms, drivers & whatnot, beginging at day 1 and across all times.

Assuming that the tests WU work & setup has been done correctly (which I assume is the case), this is a very safe & accurate way of going about things.

ChristianVirtual wrote:Beside all the technical KPIs like number of client or GigaTerraPeta FLOPs couldn't/shouldn't we add a "business language" defined KPI like nano/milli seconds of folding done per day or per WU.
And based on finished WU also feedback that to donors. It might be indirect part of the calibrated base points but getting that number into day light would be nice.
It might be a bit more complicated given the number of atoms in each WU is different; leave that up to the experts.

To many of us, flops etc. makes no difference, I do not understand it nor do I care to, to me it is just a irrelevant #, the number of people folding is the only measure of the true health of FAH.
When using flops etc. the only thing that tells you is tha X amount of work is getting done. The question is, is 1 person doing that or is that 100,000. The number of active folders in a 24hr period or 7 day period is easy enough to find just by going to EOC or Kakao 3rd party stat sites and sorting users by 24hr or last 7 days, recently the number of active folders in a 24 hr. period has been below 11,000 according to those 2 sites.

If you look at how many actual people are donating and if that number is going up or down you then have a somewhat more accurate / relevant measurement of what your public support is.

+1 on the people vs flops.

Maybe we should try to list all relevant ways to track, list, show day to day progress. Then, and only then trying to figure out how many stats we show and maybe a way to combine them all in a single figure.

Again total base point adresses part of that issue. But this excludes the # of people which is also very important.

# of physical items (1 cpu, 1 gpu, 1 core). Those are all meaningful too, but they lose meaning across time because of power throughput.

Using cpu&gpu Ghz is rather pointless because this figure is not the same across vendors & technologies. But maybe they are "stabely" uneven and might be usable in some sort of total?!?!?

Grandpa_01 wrote:To many of us, flops etc. makes no difference, I do not understand it nor do I care to, to me it is just a irrelevant #, the number of people folding is the only measure of the true health of FAH.
When using flops etc. the only thing that tells you is tha X amount of work is getting done. The question is, is 1 person doing that or is that 100,000. The number of active folders in a 24hr period or 7 day period is easy enough to find just by going to EOC or Kakao 3rd party stat sites and sorting users by 24hr or last 7 days, recently the number of active folders in a 24 hr. period has been below 11,000 according to those 2 sites.

If you look at how many actual people are donating and if that number is going up or down you then have a somewhat more accurate / relevant measurement of what your public support is.

Fair enough and I would define an "active distinct donor" count as business language KPI and also publish. It also need to be seen with what each active distinct donor contribute for the overall project. 10 distinct donors with each a pentium CPU have a different impact on the project goals compared to one donor with 10 GPU.

But yeah, I understand (too well) where you are coming from and for transparency and early-warning reason reporting on active donors contributes to a better overall picture.

Well, the ATI GPU issue is now resolved (https://folding.stanford.edu/home/updat ... -ati-gpus/) so the number of ATI GPUs are 1048.

From Client Statistics by OS: I must not understand the difference between a core and a CPU in this context. Why are there more active cores than active CPU's in several of those categories? Specifically, address Fermi GPU's considering there isn't any SMP-GPU capability.

Windows/Mac/Linux make sense; a quad-core processor running a single CPU:4 slot will show up as one CPU but 4 cores. I'm not sure what happened to Nvidia Fermi, however, as they've never counted GPU cores before, and if they were actually counting GPU shaders, that "core" count would be a heck of a lot higher.

Perhaps multiple GPU cards in a single PC?

David

That would imply that the over the entire folding community that average number of Fermi video cards per computer is nearly 2 and that is hard to believe. I would argue that the vast majority of people have only one video card and to get that average close to two means there are an awful lot of 3+ video card machines.

P5-133XL wrote:That would imply that the over the entire folding community that average number of Fermi video cards per computer is nearly 2 and that is hard to believe. I would argue that the vast majority of people have only one video card and to get that average close to two means there are an awful lot of 3+ video card machines.

All the dual-GPU-cards being counted as two GPU's and any multi-GPU-rigs being counted as having multiple GPU's is of course a possibility. Still if this is the explanation, it's interesting how where's 1.90 Fermi/computer even many users has only a single GPU while for Windows where's only 1.71 core/cpu even it's AFAIK many years now since single-core-cpu's was discontinued.

Also, I find it highly unlikely not a single Amd-user has either a dual-GPU-card or has two Amd-cards in the same computer.

BTW, while Amd has a large increase due to the mentioned stats-problems, both Nvida-types has also large increases since yesterday possible due to similar reasons. For CPU on the other hand Windows lost 6706 cpu's (5.5%), Linux 933 cpu's (21.6%) and Mac 2169 cpu's (27.7%) since yesterday. It will be interesting to see if these cpu-numers is just an artefact of the stats-changes and the numbers will be back up again tomorrow, or if it's a lasting drop in CPU-contribution.

P5-133XL wrote:That would imply that the over the entire folding community that average number of Fermi video cards per computer is nearly 2 and that is hard to believe. I would argue that the vast majority of people have only one video card and to get that average close to two means there are an awful lot of 3+ video card machines.

Historically I have had up to 5 GPU clients running in a single box. After that, heat was a major issue. AtlasFolder had many dual GPU cards running in individual boxes. Recently I run 2 or 3 in a single box. Difficult, but not rocket science, so they are certainly out there. Shrug.

I'm not saying that multi-gpu setups don't exist. I just don't believe there are enough out there to boost the average to 1.9x. There are just too many people with only one video card to move that average enough.

I really wonder: the assignment server should have a good insights on the client hardware and slot setup (e.g single, double, triple GPU slot). Making some (anonymous) statistics out of it shouldn't be that hard. It would have exact data at any given time in the whole landscape.

The amount of data shouldn't be that high a that a little R-script can't deal with it.

I can fully understand the Linux count going down, Its where BA units reside and a lot of heavy hitters have already shut down