Overall F@H Stats Graph?

[screen317]

Hello everyone!

On the Wikipedia page for F@H there is a graph showing overall statistics for (I imagine global) F@H usage, but it stops at 2007. Is there a more up to date graphic? If not, I'm curious as to whether or not it would be possible to have one that updates itself on some sort of time basis?

Regards,

Chris


Edit: There is now a chart to plot this data!! Anyone can now edit my stats page to help:
https://docs.google.com/spreadsheet/ccc ... pY1E#gid=0

Please ask if you have any questions. Historical data is of primary interest since I'm updating it (nearly) daily with current data.

[Jesse_V]

That image simply serves to illustrate the participation spike when the PS3 client was launched.

AFAIK, the closest you're going to get is http://www.stanford.edu/group/pandegrou ... Us2010.png and then of course there's kakostats and other third-party stats reporting websites.

Glad you're checking out the article though.

[screen317]

Thanks for the reply Jesse.

Would you or anyone know if the data is at least available? I could help put something together graphically if there was a way to access that sort of data over time.

[bruce]

If you're willing to start collecting the data yourself, you can try
http://fah-web.stanford.edu/cgi-bin/mai ... pe=osstats

[screen317]

Cool, thanks bruce. I'll start a spreadsheet, collecting data from there. Do you know how often that updates? It says last updated today, so I think I'll collect data around the same time every day.


Will post back after about a week!


Question: the page isn't too clear about this: what is the difference between the Native TFLOPS and the x86 TFLOPS column?

[bruce]

Question: the page isn't too clear about this: what is the difference between the Native TFLOPS and the x86 TFLOPS column?

http://folding.stanford.edu/English/FAQ-flops

[screen317]

This is the part I don't quite understand:

What are native FLOPS?

We refer to the FLOP count on a given hardware as the native FLOP count. For example, an exponential on a GPU is one native GPU FLOP but many native x86 FLOPS.

What does "...count on a given hardware" mean? That's the only part that's confusing. No idea what "an exponential on a CPU" means.


Edit: Here's how the spreadsheet looks! Will be able to start constructing the graph tomorrow. Suggestions are always welcome.
Adding to it will be easy now that the structure is laid out.

[bruce]

Computer code consists of instructions that perform specific operations like add/subtract/multiply/divide and there's a different set of instructions for integer numbers and for floating point numbers, etc. The CPU hardware has a circuit that can perform each of these operations with a single instruction.

There are more complex operations that are used less frequently like square root or sine, cosine, exponential, etc. If these functions are done very rarely, they can be done with a subprogram made up of a number of more basic operations. (The hardware is cheaper but it will run slower if you use those functions.) If the function is done frequently enough, the hardware designer can speed up the program by providing a extra hardware circuit that performs the entire calculation with a single instruction.

On one computer chip, you may need to perform 50 FLOPS to find a square root. On another computer chip, the same calculation might take one FLOP. Which way of counting FLOPS is correct?

[screen317]

Thanks bruce. I understand it better now.

Here's something else that perplexes me. Two days ago the Native TFLOPS (total) and x86 TFLOPS were 5091 and 7252 respectively; today they are 5074 and 7209 respectively. How does that make sense?

[Jesse_V]

Here's something else that perplexes me. Two days ago the Native TFLOPS (total) and x86 TFLOPS were 5091 and 7252 respectively; today they are 5074 and 7209 respectively. How does that make sense?

I don't see anything strange here. Participation waxes and wanes, and people add/remove hardware or reconfigure their systems over time. I've heard that participation picks up in the winter months, but seems to ebb a little bit during the summer. You can see from that page that GPU and PS3 folding adds far more to the x86 FLOPS than they do the Native FLOPS, so that affects those numbers as well. Like stocks, such fluctuation is normal, but they really should always be going up.

[iceman1992]

On one computer chip, you may need to perform 50 FLOPS to find a square root. On another computer chip, the same calculation might take one FLOP. Which way of counting FLOPS is correct?

This is a bit off-topic, but since the whole fah project is often compared to the fastest supercomputers, what type of FLOPS count do they use to measure the supercomputers (like in the top500.org website)?

[7im]

Seriously? Skim a few sentences on top500.org website. http://top500.org/project/introduction You'll quickly find...

In the present list (which we call the TOP500), we list computers ranked by their performance on the LINPACK Benchmark

Oh, that's an original list name.

[iceman1992]

the computer system must be able to be used to solve a range of scientific problems

That doesn't imply that they run FAH though

[7im]

And it doesn't imply that that don't run FAH.

If you want apples to apples, run the Linpack test on a computer similar to the fah benchmark computer. Then multiply by the active number of fah clients. Not accurate, but a close enough rough estimate to make it interesting...

[iceman1992]

If you want apples to apples, run the Linpack test on a computer similar to the fah benchmark computer.

I have no idea how to run linpack on a GPU. As some supercomputers use GPUs too (I recall one using HD4870s)