GPU work units keep expiring. It should be assigned smaller work units. How?

[vahid.rakhshan]

My GPU work units all expire at percentages below 10%. And I don't want to keep the laptop on and charging (still it is on for at least around 12 hours a day). But even if I keep it on all the time, still it won't finish a work unit until the ETA deadline (at its full speed). The same GPU could be used for successfully finishing a much smaller work unit. Is it possible for the server to "see" these repeated failures of my GPU and accordingly assign smaller work units to it? Or at least, it would be good if the user could throttle down the demand of the system (via some slider, for example) and request smaller work units from the server.

From FAQ, I understand that the server cannot elongate the ETA deadline. So at least, it should assign my GPU smaller chunks: That is it should (1) either break down the current work units into many (for example 10 or 100) smaller pieces so that each of them can be completed by computers like mine. (2) Or if it is not technically possible to break down large work units into smaller ones, at least the server can assign my GPU smaller work units from other projects. I think there may be many projects with varying sizes of work units available.

But I wonder why it shouldn't be technically possible. If a whole simulation can be broken down into millions of smaller pieces (i.e., work units), why each work unit cannot be also broken down into smaller pieces? I think with these large work units constantly expiring and with the all-or-nothing requirement to complete a work unit using a single device, many people might be actually contributing to nothing (because their work units keep expiring before 100% and the system truncates their work).

--------------

Another workaround may be to change the current all-or-nothing way of contributions to some more flexible ways. For example, if my GPU processed 35% of a work unit until its deadline, at least the next computer that is being assigned the same work unit should start from that 35% and not from zero.

This way, a lot of processing power that currently goes to waste will be actually used and the whole simulation will speed up.

[Joe_H]

Is it possible for the server to "see" these repeated failures of my GPU and accordingly assign smaller work units to it?

No. That would require a major rewrite of the server code on both the Assignment Servers and the Work Servers. The best that can be done at this time is placing GPUs in several performance categories and assign WUs accordingly.

Another workaround may be to change the current all-or-nothing way of contributions to some more flexible ways. For example, if my GPU processed 35% of a work unit until its deadline, at least the next computer that is being assigned the same work unit should start from that 35% and not from zero.

Again this is not possible, and would also require major changes to the software to support it. There is just one full time paid software developer working on the server code, a second was added about a year and a half ago to work on client code. They did try this out a number of years ago, you may occasionally find mentions of tests of a "streaming core". It did not work out well.

Finally over a total of 3 posts you mentioned enough information about what you are using to fold on. That is a laptop with a GT 750M GPU that you tend to only run about 12 hours a day. The processing power of that GPU is about the same as the desktop GT 730 which was about the minimum usable to process current WUs when running close to 24 hours a day. That is no longer the case for many projects, support by folding for use of Nvidia's CUDA which speeds up processing compared to OpenCL requires a Maxwell based GPU or later. The GT 730 and your GT 750M are both Kepler based.

There may be some GPU projects that your GT 750M can finish, but they tend to be a minority and in short supply. The 750M is already in the lowest performance category that still gets WUs assigned. At some point in the near future F@h may drop support entirely for Kepler based GPUs, Nvidia has already placed them on legacy support for drivers.

[vahid.rakhshan]

Thanks Joe for your detailed response. Much appreciated. So I am better off without my old and useless GPU. I might run lighter projects like Rosetta or others on it.

Finally over a total of 3 posts you mentioned enough information about what you are using to fold on.

Sorry if that made any inconvenience. I intentionally left the type of my GPU undisclosed in the first two questions because they were generic and for all devices and not just my own. But my third post was directly relevant to my own GPU model, so I said its model in the third post. Yes, I am running an Acer Aspire laptop with an Intel Core i7 4500U CPU with 12 GB of RAM, and a GT 750M with 4 GB of RAM for the GPU.

Thanks for the detailed info on it.

There may be some GPU projects that your GT 750M can finish, but they tend to be a minority and in short supply. The 750M is already in the lowest performance category that still gets WUs assigned. At some point in the near future F@h may drop support entirely for Kepler based GPUs, Nvidia has already placed them on legacy support for drivers.

Thanks. I was previewing/sending my new post when I saw your answer. Here it goes:
I found the answer to one of my questions (How to get smaller work units?). However, apparently, it doesn't work because the server doesn't have such small-enough work units for GPUs (all it has for GPUs is relatively large WUs). This is how to get smaller WUs: https://www.overclock.net/threads/info- ... ns.979691/

If you do not know how to use these advanced options correctly, please ask before you do so.

Folding slot options

This flag will grant you access to WUs in late-stage beta. (-advmethods)
client-type
advanced

This flag will allow you to get bigadv WUs on capable systems. (-bigadv)
client-type
bigadv

New flag in v7.1.24 it is related to (-gpu x)
gpu-index
x

FAHControl will start the slots in the paused state; Works on individual folding slots
pause-on-start
true

FAHControl will start the slots normally; Works on individual folding slots
pause-on-start
false

It tells when the next WU should be downloaded after the current WU reaches X percentage; X should be between 90 and 100.
next-unit-percentage
X

Will configure your Slot to get small WUs (~5MB)
max-packet-size
small

Will configure your Slot to get normal WUs (~10MB)
max-packet-size
normal

Will configure your Slot to get big WUs (~500MB)
max-packet-size
big

Will set the priority of the FahCore being used.
core-priority
idle

Will set the priority of the FahCore being used.
core-priority
low

It tells when to pause the Slot after X WUs are completed; X can be any number; 0 is indefinite.
max-units
X

It will create a checkpoint after X minutes; X is a value between 3 and 30.
checkpoint
X

X is the VendorID of the GPU.
gpu-vendor-id
X

X is the DeviceID of the GPU.
gpu-device-id
X

X is a 32 digit requested from Stanford here. This is option is used for multiple passkeys (Mainly used for the Team Competition)
Passkey
x

[vahid.rakhshan]

Update (just some info):

I set the max-packet-size for my GPU to "small" and started a new Work Unit. The server assigned me this:

18448 (4, 251, 167) 2.96 days
Work Unit (PRCG) Work Unit (ETA)

Interestingly, it also told me "My computer has 2.00 days to complete this work unit."

This means that even at full speed running for 24 hours, my GPU can't finish a SMALL-sized work unit.

[Joe_H]

That information from overclock.net is a bit dated, for example the options bigadv and advanced are no longer used. Normal for WU size is now considered up to 25 MB. Even when it was used regularly, the max-packet-size was a holdover from the end of the dial-up era and when I was still on DSL was barely useful by the end of 2010s. That was more related to the file upload and download size than how long a WU would take. There are other minor changes since that post from 12 years ago.

As for the estimated time, until your client has processed a WU for several percent it will not be accurate if a WU from that project has not been processed before.

[vahid.rakhshan]

Thank you very much for taking the time to clarify all these. I think it would be very good if someone could put all such basic questions (and answers to them) up in the FAQ. The current FAQ doesn't have any of these good details. And I think FAH can reach a broader audience (of non-tech-savvy) users if it has a more user-friendly, clearer, and more easily available GUI / HELP / FAQ.

[Joe_H]

The forum FAQ mostly covers use of the forum itself and is provided as part of the software. Many of the folding related questions are covered by the Welcome topic - viewtopic.php?t=26036. It may need some minor updates as some of its details are based on a slightly earlier minor revision of the 7.6 client.

This forum is volunteer run. Some of the posters who most helped out have moved to the discord, others are posting less or have left for various reasons. A new FAQ for here or major revisions to the material already covered will need someone to put it together.

The FAQ material on the foldingathome.org currently mostly was copied over from the original Stanford site when that was closing down. The person who had been maintaining it passed away about the same time, and the F@h Consortium has not recruited new volunteers to update it. Volunteers did work on updating the main pages there, but further work remains.

[toTOW]

What GPU do you have ? We might have to remove it from the list of supported GPUs ...

[vahid.rakhshan]

What GPU do you have ? We might have to remove it from the list of supported GPUs ...

nVidia GT 750M with 4 GB of RAM

Though as Joe said, it is still possible that it might finish a WU (at least some lighter ones) if running full-time at full speed.

[vahid.rakhshan]

The forum FAQ mostly covers use of the forum itself and is provided as part of the software. Many of the folding related questions are covered by the Welcome topic - viewtopic.php?t=26036. It may need some minor updates as some of its details are based on a slightly earlier minor revision of the 7.6 client.

This forum is volunteer run. Some of the posters who most helped out have moved to the discord, others are posting less or have left for various reasons. A new FAQ for here or major revisions to the material already covered will need someone to put it together.

The FAQ material on the foldingathome.org currently mostly was copied over from the original Stanford site when that was closing down. The person who had been maintaining it passed away about the same time, and the F@h Consortium has not recruited new volunteers to update it. Volunteers did work on updating the main pages there, but further work remains.

I was mainly talking about the FAQ material on the main (foldingathome.org) page which is the first place most potential volunteers visit first, and not about the forum's FAQ. Perhaps a more comprehensive set of guides/FAQ/help on the official website as well as on the Web Control page and also on the Advanced Control GUI would give a better sense of control and understanding to them. Otherwise, it is possible that many of them think of F@H as too complicated, user-unfriendly, and difficult (and even vague) and leave it.

But now that I see this project is not well-funded (at least the way I thought it would be) and is somehow a one-man job, I understand this is not much possible for now. I hope in the future F@H can become a part of major operating systems of famous devices (e.g., as some pre-installed program), so that it can reach a much larger audience and at the same time receive a lot of good funding from big tech companies and a great number of paid developers who can redesign the whole procedure into much more flexible algorithms.