Folding Forum

I'm starting to wonder how long the FLOPS will go. Any predictions?

long or low?

total folders continues to drop off by a few thousand a day.

ps3= obviously a very mature platform..amazing that the numbers are as high as they are.

move to advbig16 is certainly not making any friends with the competitive folding (OC quad core) set.

Consequently, FLOPS will continue to decline until there is support for Kepler and AMD GCN.

I don't see any other near term positive drivers for FLOPS increasing. JMHO

I've added a comparison to the fastest supercomputer to this graph and to the Wikipedia article. I found the result quite interesting: see the graph and also http://en.wikipedia.org/wiki/F@H#Performance. The F@h stats spreadsheet was necessary for this to happen, so thanks!

However, I don't have any data on how F@h was doing around November 1, 2009, and I don't know where the numbers came from for June and November of 2010. I think it would be fantastic if we figured it out. Sorry for the "say where you go it!" mentality, it's a side-effect of immersing myself in Wikipedia's culture.

It's also interesting if you log the Y axis - clearly supercomputer performance is increasing more or less with an order of magnitude, while FAH has plateaued in relation

I think accurate referencing is always preferred, but if not referencing is the difference between people adding data to the spreadsheet and people not adding the data then I'll always go with more data is better. If someone wants to use the data and its sources for the wiki entry then its largely up to them to verify the data.

I think a lot of the boost has to do with the relative proliferation of GPGPU over the last few years. Tons of these new supercomputers are just lots and lots of GPGPUs, rather than arrays of CPUs as it was before.

It's down more to the economies of size. All but the most extreme folks won't have more than one videocard in their PC; the ones who put 2-4 in are a relatively small minority compared to the general population. And CPUs... well, it's still uncommon even for quadcores, and dualcore penetration is maybe 50-60%? There's still lots of people out there on 10, even 15 year old PCs, let's not forget.

The third factor is of course, FAH is opt-out. If you don't want to fold anymore, you just stop running the program. People come and go every day.

Obviously as things get more powerful, our folding capabilities will increase. Just look at how things were 20 years ago. I remember as a kid handling 5.25" Floppy disks on an Apple II that were 140 kilobytes to a side.

And now I am holding a USB drive in my hand that holds 128 GB and transfers things in the range of hundreds of MB a second; an increase of several thousandfold in both capacity and in transfer speeds. Heck, this thing is bigger than the HDs in my first three HDs (which were 20, 60, and 120 GB)!

Time marches on, really... it's just that supercomputers tend to get the "latest and greatest" first, and then stuff trickles down to the consumer sector. In that sense, FAH is bound to go through ups and downs; right now, we're simply in a down, that's all.

Can someone fix the TFLOPS (per client) graph? It's including the physical numbers and I can't figure out how to change it.. Also, could the NVIDIA numbers be added?

It looks like someone tired to add NVidia numbers. and borked it.

I've come across the same outcome but only previewed it (not actually applied the changes). Removing the NVidia column tends to fix it... It looks like Docs either has a problem with extra column and/or with the Nvidia data (Perhaps there are too many null values to start with or perhaps there is a sneaky string value in there or something). Unless someone knows a magic trick or two I've just decided to leave Nvidia off it.

On a positive there are initial hints of PPD bottoming out - lets hope the worst of the summer shutdown has passed!

It's been cooler in most of the country lately, and well, September is a week and a half away. I'd say it's likely.

Someone made a typo for the Linux TFlops entry for 9/14/2008- it's listed as 63016, which is clearly way too high, and it's throwing off the scale of the graphs. I assume it's supposed to be 63, but I don't have the relevant data.

Zagen30 wrote:Someone made a typo for the Linux TFlops entry for 9/14/2008- it's listed as 63016, which is clearly way too high, and it's throwing off the scale of the graphs. I assume it's supposed to be 63, but I don't have the relevant data.

My mistake. I was copy-pasting some of the entries from that Chinese forum, and somehow the previous entry for OSX Total CPUs also ended up in the Linux TFlops. The source says 60. Fixed.

On a positive there are initial hints of PPD bottoming out - lets hope the worst of the summer shutdown has passed!

CPU have gone down 20K (from 277k to 255k) in a month, but total Flops seem to have reached a bottom .

I guess that nets an increase in efficiency.

mdk777 wrote:I guess that nets an increase in efficiency.

I'm not sure that's properly called "efficiency" but everybody who upgrades their CPU/SMP hardware from N cpu-cores to M cpu-cores (M>N) or from a supported GPU to a more powerful supported GPU will increase their PPD (and FLOPS) without changing the number of FahCore(s) that they're running.

I guess it's efficiency if it's measured in terms of Flops/client - i.e. getting better performance out of a given client.

bruce's example would be the case where the type of folders remains proportionate [i.e. everybody]. However, it is quite possible the composition of folders is changing. One scenario is a lot of small-time folders leaving the community and a small number of the more hard-core, dedicated, folders turning a small amount of more powerful hardware back on as temperature's fall (An i7 could replace many P4's). Alternatively, it could be that the situation is just as a result of the hard-core continuing to upgrade their hardware 'as usual', with the loss of small time folders nullifying that increase and dropping the total # of clients in turn..

Of course, a reduction in low power clients is not necessarily a bad thing per se, particularly if the computational power is offset by a small number powerful machines. It basically comes down to whether PG is aiming for greater FLOPS or faster returns for a given amount of FLOPS. The QRB and their public comments indicate that the latter is the case and actually point to FLOPS/client as an important metric...

It looks like FLOPS and participation has leveled out. Following the last two years as an example, I think we'll start to see a good rise. Perhaps it's time to buy some shares of FAH.

I hope you're right.

Maybe and maybe not right now there is a pretty big push going on between evga and the [H] in the battle for #1 allot of rigs have started up on both sides starting on Sept 1.