Dear all,

I have been crunching seti@home work units since like 20 years ago give or take.
Now i am a married guy, have kids running around and having peaceful evenings.

I have this mania to restore old computers, stick any kinds of supported GPUs and let them crunch seti@home, asteroids@home and Einstein@home 24/7.

So i have this question, how do i know which GPU is more powerful then another as far as seti@home is concerned? Which performance metric is important? Like is it shader units, ROPS, Pixel rate, FP32 floating point, FP64 double ..... ?

Which metric is important? I have more graphic cards than rigs to plug them in, so how do i tell which cards are better and should be utilized other than others?

Many thanks,
Mile

Generally the more expensive cards have more shaders or compute units and are the best for distributed computing. Depends on the project whether the science apps need or use double-precision floating point. Milkyway@home is the only common project that needs exclusively cards capable of double-precision FP. Anything over 2GB of memory is sufficient for running singles.
You can look at this chart to see where your cards fall with regard to Seti computation efficiency and production.
https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1997116

The stock Seti apps use OpenCL applications that need the OpenCL component of the respective drivers installed. If you are running a card with CUDA Capability of 5.0 and also running the Linux OS, the special applications are the fastest for Seti. Between 3X and 10X the production speed of the stock apps. Look at the Top Computers list and you can see it is dominated by Linux hosts.
https://setiathome.berkeley.edu/top_hosts.php

Thank you sir, this is exactly what i needed.

One more question though, what do you mean by special and stock applications?
Isn't it that the BOINC client is the only application since they shut down the old client and moved to BOINC?

The fastest card isn't always the fastest for every project.
The majority of the GPU projects I use, don't fully utilize the GPU's potential (which is why so many people are trying to fit 2, 3, 4, or 5 projects into one modern GPU.

I've just learned, that most projects aim to fully occupy a GTX 1050/1060.
for older hardware, the Nividia P106 GPUs are discarded mining cards, are excellent buys for this on the second hand market. They go for as low as $40, and perform just slightly slower than most RTX GPUs on one project. Plus they're PCIE 2.0 compatible (not 3.0).

For older hardware, they're probably the sweet spot on a budget for crunching!


As far as the projects go, boinc is just a client. An interface, between your hardware, and each project. There are tens (if not hundreds) of projects that work with Boinc, but not everyone of them work in the same way.

Stock - science applications delivered by the project.
Special or anonymous - applications installed by the user, often using custom configurations and frequently highly optimised for maximum performance (run times as short as possible without producing errors)

Thanks for the info.

Are there any special applications i can try and use with S@H that you can recommend me?

What operating system?
What GPU?

For example Windows 10 and Linux Mint 19.

GPUs: GF750TI, GF550ti, Quadro FX 1800 анд мултипле GF520

Thanks for the info.

Are there any special applications i can try and use with S@H that you can recommend me?

Applications that do what?
I just recommended Netlimiter to someone who wanted to reduce the network load caused by tasks being either uploaded or downloaded.
There are a lot who use VNC, to control their servers remotely.
Boinc is pretty much an application you'll want to run, with as little distraction as possible (no bloated system and preferably on Linux too).

You may be interested to try out a new version of Linux.
Installation procedures of the OS, and drivers; since Boinc benefits from the lower PCIE bandwidth usage in Linux.