Over 2 exaflops!

[gunnarre]

The "NVidia SA team" user and is rising in the ranks at least:
https://folding.extremeoverclocking.com ... &u=1162743
and "NVIDIA_Saturn_V" have also increased folding this last month:
https://folding.extremeoverclocking.com ... &u=1096093

[MeeLee]

There is a significant increase in the number of Nvidia GPUs, from 220000 two weeks ago to 423000 today, all the increase is on linux machines. What is that? A test farm installed by NVidia to test their upcoming 30xx consumer grade GPUs? All other machine numbers (Windows Nvidia, AMD on Linux and windows, CPUs... are falling).

Some of the stats haven't been updated for years...

[Jonazz]

There is a significant increase in the number of Nvidia GPUs, from 220000 two weeks ago to 423000 today, all the increase is on linux machines. What is that? A test farm installed by NVidia to test their upcoming 30xx consumer grade GPUs? All other machine numbers (Windows Nvidia, AMD on Linux and windows, CPUs... are falling).

I wonder how long these will continue to fold. They've come at a good time to help with the moonshot project.

Currently at 2.11 exaflops.

[Endgame124]

Now listed at over 2.6 exaflops, but I’m doubting the numbers. The assign rate, the points / day, and WU / day don’t seem to be increasing at all.

[Severian]

3 x86 exaflops !?
Maybe a supercomputer from a company has joined the force ?

[Jonazz]

That can't be right, I'm not seeing more points being awarded.

[psaam0001]

I am slowly resurrecting my Windows 7.0 SP1 machine. New GT 1030 card and upgraded memory have been ordered.

Paul (trying to add FLOP''s on my end--even if it takes awhile for the effort to show up)

Edit: New parts will be here on Thursday. I'll be getting that system upgraded, before taking on the issue of getting the GT 710 folding on my Dell T105 (one of my Fedora v32.x systems)....

[PantherX]

When a donor re-installs the client or re-initiates a VM, the client count can increase without any real increase in the number of clients. That issue would fix itself in 50 days but I don't know if any method to speed up this process.

[gunnarre]

I've seen some cloud/infrastructure operators say that they're doing folding by launching a fresh instance every time they're folding a work unit and then destroying the instance. That would presumably create a bunch of new CPU IDs which only deliver one WU each.

[ThWuensche]

How is that rate calculated?

For me it does not seem to fit to the observations, the figure seems way to high. From different calculations (see here viewtopic.php?f=16&t=36081) I'm assuming that my systems contribute about 1/3000 to 1/5000 of the general compute power. The calculation mentioned there seems to be supported by observations on moonshot 13426 WUs, as far as I know a total of 1.8M WUs, with 5% of that finished per day resulting in 90000 13426 WUs/d. My contribution of ~60 WUs/d is about 1/1500 of the delivery of these WUs. Taking into account that most of my contribution comes from eight Radeon VII (with a theoretical single precision preformance of 13.8 TFlops per unit) theoretically providing about 110TFLOPS, I would come up with a total performance of the FAH system about 3000*110TFLOPS -> ~ 330PFLOPS. That's by far lower than 2EFLOPS.

Or would that 330PFLOPS GPU by the x86 FLOPS magic lead to the 2 EFLOPS?

[PantherX]

The measurement of FLOPS is a tricky process and while I don't know the exact details, some of the common issues are addressed here: https://foldingathome.org/support/faq/flops/ CPU and GPU FLOPS aren't the same which makes it even more tricky.

[cocolacre]

So this graph is fake news after all ?

[bruce]

I've seen some cloud/infrastructure operators say that they're doing folding by launching a fresh instance every time they're folding a work unit and then destroying the instance. That would presumably create a bunch of new CPU IDs which only deliver one WU each.

true.

FAHClient does not tell the server when it's being overwritten. The server sees only a new client come on-line. With no report from the old client the server cannot tell the differerence between a new client plus an old client that won't check in until it needs more work and a new client minus an old client. How long should the server wait before deciding that the old machine is never coming back. 50 days is too long, but there's no better timeout.

A container-based machine with persistent storage can retain the old ID but that storage is not required.

[Severian]

Well, if there isn't any usable stats, I suppose that there isnt' any world record, so maybe the wikipedia article should be corrected :

Folding@home is one of the world's fastest computing systems. With heightened interest in the project as a result of the COVID-19 pandemic,[7] the system achieved a speed of approximately 1.22 exaflops by late March 2020 and reached 2.43 exaflops by April 12, 2020,[8] making it the world's first exaflop computing system.

[8] Pande lab. "Client Statistics by OS". Archive.is. Archived from the original on April 12, 2020. Retrieved April 12, 2020.

I used these high scores to try to convince friends to join F@H.
Are you sure that F@H is the world's first exaflop computing system ??
I feel a little stupid now.

[Joe_H]

Back in April the means used to generate that statistic were relatively accurate. They have since been distorted by the results coming from people running containers w/o maintaining configuration data between runs. So each time a container starts up, it registers as a new system. and its previous usages still remain until they exit the window used. It has been brought up, how they will handle such statistics in the future has not yet been announced or determined to my knowledge.