Folding Forum

whocrazy wrote: ↑Tue Dec 12, 2023 4:20 am Hi.
When you change gpu options, where are the settings saved to, the registry? or a file?
Thanks.

I have no idea, but it might write within the driver settings. That is assuming you are asking in regards to nvidia -smi. But I'm not even sure if that would take place at all unless you used the persistent setting for it to apply on a reboot.

I'm fairly sure the commands to both set and check power limits in nvidia-smi would work well with your screen reader.

As far as PPD per KWh it goes as such:
4080
4090
4070ti
4070

The best bang for the buck when it comes strictly to folding in my opinion is a 4070ti.

Next month several new cards are being released, 4080 Super, 4070ti Super, 4070 Super. Supposedly at very similar pricing as current, so it may be best to hold off on buying a new card for a bit.

The 4070 looks like a bit of a dud with the 7-8M PPD showing on the LAR Systems database, but with the GPU typically costing $550-600+.

My 4060 Ti routinely shows 6-7.5M PPD estimates, shows 6M avg on LAR and was on sale recently for under $350 but still isn't bad at $400.

Now there are rumors of new NVIDIA models coming next month:

4070 Super, available Jan 17, $599-$649, 220W TDP, 3-6% slower than 4070 Ti
4070 Ti Super, available Jan 24, $799-849, 285W TDP, 14-22% faster than 4070 Ti
4080 Super, available Jan 31, $999, 320W TDP, 6-9% faster than 4080

https://videocardz.com/newz/nvidia-gefo ... ng-the-way
https://www.youtube.com/watch?v=W5K8GM2fNDM

whocrazy wrote: ↑Mon Dec 11, 2023 8:18 am Hello.
I'm going to exchange my Intel video card for an NVIDIA, and I am wondering which one is the best to use for both gaming and folding?
I had a Geforce 9600 gt on a PC many many years ago, but I'm sure they have come forward in leaps and bounds since then.
What do you suggest?
Thanks.

Best choice for most applications is to get a x60 GPU of current or previous generation. You might want to push to x70 with extra budget.

I would suggest you to fold on Linux to achieve an additional 15% vs Windows.

4070S hitting 12m ppd on stock settings and Windows10. Massive improvement over 4070 for the same money.

Calculated daily PPD for 4070 Super on Linux:
4060 Ti on Linux:

Running RTX 4070 Super @ 185W and RTX 4060 ti @ 140W on Manjaro Linux nets 18 - 20,000,000 PPD on Ryzen 5950X

Throw in a graphics co-processor CPU for another .5 - .75 million

To answer the question as of now, my math tells me that an L40S gets an average of 28,809,869 PPD. I have 6 in a lab, my 24hr avg is 172,859,218.

That's some monster science being done. Up in the top 5 of all users.

But I think it's safe to say that most won't be shelling out the cash for even one L40S, much less 6 of them!

I consider this definitive; dig down to individual GPU pages will show Points Per Watt.

https://folding.lar.systems/gpu_ppd/overall_ranks

While up to date for most for mid-range to high end GPUs, it's not a definitive list of all GPUs.

Which leads me to wonder how many Teams / Users are using Data Center versus Consumer grade GPUs.

Y'all ought to catch me by the first of June.

Nah, I'm fixing to re-allocate these GPUs to some other project. Just doing some testing with the FAH core running on them.

pcwolf wrote: ↑Fri May 10, 2024 1:51 am I consider this definitive; dig down to individual GPU pages will show Points Per Watt.

https://folding.lar.systems/gpu_ppd/overall_ranks

Its not always, I found my PPD tanked if I left Chrome open running that extension so I do not contribute to the page any more. Though right now, that does look fairly accurate, but the creator has to periodically clean up results as its shown some very odd results before.

It also doesn't take into account points per watt, you can dramatically underclock the 40x0 series cards and lose just a few PPD while reducing power consumption dramatically, especially on Windows where you can run the OC Scanner but you then have to manually change the clock rate with nvidia-smi.

The 4090 FE at different clock rates:


Notice interestingly the ASUS 4090 TUF Gaming OC is scoring higher (first entry in table) and using more power than the FE card, despite them both locked to the same clock rate. I suspect this is differences in silicon quality and/or VRM efficiency (assuming the wattage is measured at the 12V input).

Of course, it may also be differences in the WU being processed too, but it does seem to consistently score higher in general.