Folding Forum

I've been wondering if anyone has complied a database on how each brand of a certain series of GPU performs on one specific work unit. Maybe Windows vs Linux also.

Has this been done?

tks
Rick

Guess it would be GPUs vs FAH project which consists of many identical work units. All we know currently is that projects with low amount of atoms have a bad performance on GPUs with a high amount of shaders. And Linux is 20% faster than Windows with FAHCore_21 but in future there will be FahCore_22 which speeds up Windows to equal Linux.

I was thinking about something like this:
………………………….WU 1530
………………………...eVGA………….ASIS......Gigabyte...etc...…...etc
GPU 1060 PPD

GPU 1070 PPD

GPU 1080 PPD

Have a look at: https://docs.google.com/spreadsheets/d/ ... pub?gid=2#
Although it doesn't have manufacturer It may have some of what you're looking for ...

I'm not aware of any significant differences between a GPU packaged by eVGA or Gigabyte or ASUS or ... except based on the number of shaders and the clock rate. The actual GPU chip sets the number of shaders, Many offer two models, one with a standard clock rate and another with an OC somewhere in the name that is overclocked. Sometimes there are two types of GPU memory (VRAM) and that makes a small variation in throughput. There can be a variation depending on the width and PCIe version of the connection between the CPU and the GPU.

Just look up your graphics card and 'PPD' in google. YOu'll find many references on this forum, as well as on other sites about what points on average people get with their cards.
I find my cards are slightly above the average ratings on most WUs. There are some projects that the card is slower.
But I don't think anyone actually has compiled a list of projects and card performance.
Some projects go, and new ones come in. it would be a lot of work.
Instead an average is taken of multiple projects and WUs.

So from what I'm reading the only difference of one brand from another is the packaging. One fan vs two or three fans. A light here or one there and so on. Kind of figured that as it happens in so many other products as well. Remember the old 3 in discs when all you had to do to change a single sided disc into a double sided disc was cut a hole in a square that was automatically cut in the double sided discs.

When I choose a new GPU, I decide which chip (NVidia vs AMD and then how much I want to spend on the highest ranked chip or on a chip with less active shaders). Then I pick a board designer based on warranty policy or reputation.

Most products will also say 'boost clock speeds'.
In case of NVidia, all 2070 and 2080 cards now will have higher binned GPUs, meaning that in theory an RTX card of Zotac, MSI, PNY, EVGA or Asus all will perform the same, and it will be on the manufacturer's power and cooling solution to determine the difference between boost speeds.
Like Bruce mentioned, once you have found a card you like, the next step is the cooling solution. Single fan, triple slot cards usually are the slowest, and triple fan, dual or 2.5x slot solutions are better at extracting heat, which usually translates to higher performance, but at the cost of being less compact.

Comparing between brands and models, boost clocks speeds can make a 10% or more of difference in folding applications.


Making a database of each card, for each project would take a massive amount of time, as each card has different cooling solutions, boost speeds, and also depends if you're running the card in a 16, 8, 4, or 1x slot.