Is my GTS 250 no longer

[GTS250]

Is my Nvidia GTS 250 no longer able to participate? I found a previous thread in this forum but it would not let me post the link (anti-spam measure). Still, you can find it by looking for:

Poster: barebear
Date: 30-July-2014 at 11:37 p.m.
Title: NVidia GeForce GTS 250 not getting work assignments

Scroll to the bottom and you will find a reply from: bollix47

I interpret that reply to mean the GTS 250 is no longer able to participate in the "Fold at Home" project.

Is that interpretation correct? If yes, then does anyone know of other GPU donation projects I could consider?

Thank you!

[ChristianVirtual]

I fear bollix47 is right. You could look over at BOINC and GPUgrid; but most likely the same situation will be there too. You might want to consider replacing it with something newer; which could have beside a performance increase also a reduction of energy consumption.

[Joe_H]

I am unable to find the posts with more details right at the moment, but the situation can be summarized as follows:

1. The Core_11 projects that used to be assigned to the pre-Fermi nVidia video cards such as your GTS 250 have ended.

2. The newest projects using Core_17 & 18 will not run on these older cards.

3. Some pre-Fermi cards can successfully process WU's that use Core_15, but not all. It depends on the CUDA version supported and the driver installed. Assignments to this group of cards requires the use of the beta flag and we ask that you monitor to see if your card does work. You can read a bit about this here - viewtopic.php?f=66&t=26646.

4. Core_15 was announced as approaching end of life. So how long the remaining projects will run is not known. But some projects have taken a couple years or more to finish.

I have no suggestions for other projects that could use your GPU. However other readers here may send you PM's with suggestions if they know of ones that have projects that can use your GTS 250.

[jcoffland]

Clinton, it depends a lot on what your goals are. If you goal is solely to earn points with your existing hardware then you may be disappointed.

If your goal is to support disease research in the best way possible then I would ask you to consider that we must weigh the environmental impact of using electricity against the scientific value of the results we expect to get from various types of hardware. Newer hardware is both faster and more energy efficient by a large margin. If we only gain, say a 1% performance increase at the cost of a 10% increase in energy usage then that is a bad tradeoff. I just made those numbers up but they are not unrealistic when considering these older GPUs.

There is yet another factor we also regularly have to consider. That is the cost of man power. Which we have in limited supply. Focusing on supporting old GPUs is not very effective for us. However, listening to old folders is very important to us so our ears are open.

[7im]

This is the latest info that I know, with the last recommendation likely soon in your future.

viewtopic.php?p=273584#p273584

[GTS250]

I noticed "Core 11", "Core 15", etc. What does that mean?

[Joe_H]

I noticed "Core 11", "Core 15", etc. What does that mean?

The various "Cores" are the part of the client that actually do the calculations - "folding" - for the client. As new features have been needed to do the simulations on different proteins, new cores have been released over the years. So Core_11 is an older GPU folding core that was supported on a range of video cards available when it was released and for a while after, and had specific projects that used it.

Core_15, 17 and 18 are all newer releases of GPU folding cores with specific projects associated with each of them. They also have requirements for various GPU functions being present that might not be available in older cards to be able to run. There are also other cores for folding using your CPU.

Using the current version 7 client as an example, processing a WU involves the FAHClient portion which downloads and uploads work and downloads the folding core if needed. FAHClient starts and stops the core folding the WU. FAHControl or the Web Control is used to control when folding is done and what settings are used by FAHClient.

[bruce]

CPU hardware has been around a long time and changes that represent a new hardware generation are relatively infrequent and/or make relatively minor changes in capability (as well as changes to speed and power efficiency). In comparison, new generations of GPUs happen more frequently and involve bigger changes in intherent capability (as well as speed and power efficiency).

[toTOW]

The GTS 250 is the same as my 9800 GTX+ (G92b chip with Compute Capability 1.1), so it should happily fold the core 15 WUs in beta (see the thread referenced by Joe_H).

But that's definitely not a choice for the future : these cards will be unsupported sooner or later, so it can only be a temporary solution until you replace it by a modern GPU.

[bruce]

There is more than one GPU that was designated as a GTS 250. The output from lspci will tell us which one you have. It may contains a G92 or a G70 which is quite different. Please post the system section of your log. It may help us determine if it supports Compute Capability of 1.0 or 1.1 or not. In any event, most active FAH projects currently require Compute Capability 2.0 or higher, so even if we find a way to enable some projects, the GPU's days are numbered.

[GTS250]

I checked the card itself. The GPU chip states G92. Not to sound ignorant but, it is 1 GB card. How could 1 GB not be enough power? Or are GPUs more than just their ram?

[Joe_H]

Yes, your video card is more than just the RAM. RAM itself does not provide any processing power, just memory to store the results whether they are video elements or the results of folding calculations done by folding. The GPU processing chip is a G92 from nVidia, first introduced back in early 2008. Compared to GPU chips that have come out since, it is lacking many features that are used by more recent GPU folding cores. In addition its raw processing power in GFLOPS is a fraction of that possessed by more recent GPUs.

[bruce]

Increased RAM is important if you drive high resolution monitor(s); it's not particularly important to FAH. The next time you buy a GPU, if you have a choice of RAM sizes AND if the primary use is FAH, choose the one with the smaller RAM because it will consume less power. If the primary use is high-resolution video, then choose whichever one with support your video needs.

For use by FAH, if you have a choice, choose one with more shaders and higher clock-rates. FAH computations make good use of massively parallel floating point operations (commonly rated in Single Precision GFLOPS though that's only a rough estimate of the GPUs capabilities for FAH)..

[7im]

I checked the card itself. The GPU chip states G92. Not to sound ignorant but, it is 1 GB card. How could 1 GB not be enough power? Or are GPUs more than just their ram?

After following the links provided above, and changing the max-packet-size option setting to small does the GPU not download any core_15 work units?

Anything with GPU folding power below a GTX 460 will typically get better PPD using this option and folding core_15 work units due to the slower frame times on core_17 work units (where supported) and the nature the QRB points curve.

And core_15 is the only and last option on cards below the Fermi level (<4xx series).

[GTS250]

Oh, I haven't even gotten that far. It seems like even if it is supported, then it won't be supported for long. I will see if I could find some other use for it.