scott@bjorn3d
Not sure how to make it real big, but I saw I can click on it and make it bigger, not bad for 2 2080TI's and 2 1080TI's

I only see two cards, both 2080ti's but holy smokes!!!!!!

STR1D3R_2

Ghost13
i see that most run with two of the same cards,  So with that can you say a R9 280 x and a GTX 490 at the same time or will that cause any issue  ?
cause i have those two cards, but the amd folding machine i am working on still needs a CPU.  

Not sure about the r9 280 (dont have them)  but I can say I wouldnt bother with anything below a 1060 these days. Moving that statement  to I Wouldnt bother with anything under a 1070 as soon as I can 

I would say nothing below a 1650 super, or 1660 is worth it.

STR1D3R_2

yodap

Cool GTX

yodap

Ghost13
what about using two 1070 FTW cards ? 

Easily 1.6M ppd with core 22 on those.

Core 22 - getting 1.0X to 1.1 Million PPD on  single EVGA 1070 Ti FTW2

The large work units are not as point friendly, here's my 1070SC for the last 3 days @ team 41608. My SC does not OC well so those are stock results. Older MOBO and ram and Win7, and all wu's are core 22.
 

How old of a Mobo? I have 3 1060's on an X58 that are pulling 450-500+k each. Running Mint, stock, fans also stock (cool room)

It's an 1155, z77 model with a 2500k and 8 Gb ram. I've run 5xx through 10xx series cards on it for a long time.

    well got my first  .. 

STR1D3R_2

scott@bjorn3d
Not sure how to make it real big, but I saw I can click on it and make it bigger, not bad for 2 2080TI's and 2 1080TI's

I only see two cards, both 2080ti's but holy smokes!!!!!!

Client0 is living room machine with 2 1080TI's in it.  That's running an old X79 mobo and  i7 3960X CPU.  Both machines in that screen capture are doing over 10million PPD at that moment.

STR1D3R_2
Tried to add a few pics in post #56
Got 1 pic uploaded and that whole  post went bye bye. So here, to add the beta flag,


 


 
 Edit> TY Cool GTX , Rock on!!
edit by Cool GTX rescued from the Spam Filter

I wonder if with multi GPU setups, you can just set the client-type to beta in the 'configure'>>'expert' window instead of doing this for each individual GPU?

Ghost13
bill1024
New




crunching milkyway tasks ? 

bill got a nibble.
 
Congrats Ghost13 for getting it done.

LOL ...  now that right there is funny i dont care who ya is  ..

Not bad for Beta.
Please understand I still haven't fully configured them.
Running them around 3,5M PPD per GPU right now.
 

Tesla GPUs have been around for quite some time.
Some older models fold a lot worse than even a GTX 1650, both in performance and power draw.
The newer models do double precision a lot better than 2080Tis, but this isn't supported by FAH.
For single precision, they're only marginally better.
 
If a $1k 2080Ti folding nets 100%, I would guess a $3k Titan does ~110-115%, and a top of the line, $5-8k Tesla would do ~125%.
It would make much more sense to double or triple up on 2080Tis.
Then again, fah's qrb might reward you more generously with a modern Tesla GPU, even if less actual work is done than with 2 RTX 2080 Tis.
 
Last time I folded, there really was only 1 project (in beta) that could make full use of a 2080Ti and net over 4M PPD. All other Core 22 projects would net much lower PPD, not to mention that they would mix core 22 with core 21, which was a total pain to maintain.
Power limiting a GPU to match core 21 WUs, would cause a serious reduction in PPD on core 22 projects, and allowing core 21 to run at power levels of core 22 WUs, would unnecessary waste power, and sometimes cause overclocking errors...
 
In short, there's really no benefit on running Teslas. They cost too much, consume too much under load, and where the speed difference might matter (core 22), the PPD benefits get negated by the mix of core 21 WUs they'll send you. That and there might only be 1 project that could fully utilize a modern day Tesla.

This is outdated information.
DP is required by newer projects on FAH.
GPUs without sufficient DP performance are no longer getting work.
The reality remains that Tesla's and quadros are not cost effective for folding. The price premium is just too dang high.

ipkha
This is outdated information.
DP is required by newer projects on FAH.
GPUs without sufficient DP performance are no longer getting work.
The reality remains that Tesla's and quadros are not cost effective for folding. The price premium is just too dang high.

DP is only a fraction of a WU. 
Folding still uses SPP for all Wu's, with a tiny bit of DP.
If they only coded it better, the CPU could do the DP part, leaving the SP for the GPU.
Without any real numbers, I would say anywhere between 80-99.8% of a WU is done in single precision. Some are 100% in single point precision.

For Linux the procedure is more complicated, but for Windows, each card manufacturer has their version of an overclocking utility, which allows you to both overclock, adjust fan speed curve, and limit power consumption, as well as monitor voltages, frequencies (of GPU and ram), fan speed, and temperatures and more.
Some programs also allow you to change the led colors and their patterns.
 
The most used ones are MSI Afterburner, or EVGA XOC (this one being my favorite). Zotac also has a program which I've never used. EVGA also has a paid version of their overclocking utility called 'precision xoc', or 'precision x1'. It's more elaborate, but not really needed for fah. and perhaps there are more overclocking programs out there.
 
These programs give a lot of control to fold better. Primarily reducing power consumption (to lower heat and reduce your power bill), increase fan speed (lower heat), and overclock to compensate the performance for the lower power draw.
The latter being an art to get it all just right.
 
But you can start with the first two, power limiting and fanspeed. Lowering power consumption, or increasing fan speed from the auto settings, won't harm your GPU.
Overclocking might... You'll need to read up on it 

Using the power slider alone, will allow the voltages to automatically adjust.
There are those that play with the voltages, and are successful, but in my experience, what works for one Wu, doesn't for another.
For that reason, I only use the power slider, and leave the voltages alone.
And I set my fans to 100%, save for the living room unit, which I crank up to just below where it becomes noisy.
 
You can safely drag the power slider down from 100% (~170W) to the minimum 125W without failing WUs.
2060, 2060 Super, 2070 runs fine at 127W, with an overclock.
2060KO, 2070 Super, and 2080 need 135-144W (or more).
 
From there, you can apply the overclock.
 
Memory overclock on  my RTX GPUs is +1400Mhz, regardless of what version. (Some of software will display half of that, 700Mhz, and 7500Mbits for memory frequency).
2060, 70, 80 and 80Ti will run at 15Gbps. The 2070Super and 2080 Super run at 15,5Gbits.
It's between 1000 and 1900 MHz (Wu's on many GPUs error at over + 1800Mhz overclock, and there's no real performance improvement over 1000-1200Mhz anyway.
I just keep em running at what they're made for, hynix and Samsung modules run at 15 and 15,5Gbps from factory, and are downclocked when installed on GPUs to 14 or 14,5 Gbps.
 
For GPU, run a WU at max power settings when your system is cold (120%).
Record the target GPU frequency, without any overclock.
Set the power limit to the above setting (127/133/145 W depending on the GPU)
If your GPU frequency cold with 120% power setting gave you a boost frequency of 2Ghz, and after capping the power to above it drops to 1600Mhz, I would start with a +100Mhz GPU overclock.
 
Then add in increments of 20Mhz every 5 or so minutes, until your WU errors.
When it does, take 25Mhz off. From there adjust 5Mhz every 10-20 minutes. Increase in case successful, decrease when the WU errors.
Sometimes a setting working for 1 hour might still get you errors over the course of the day.
If you run 1 error per day, you can lower the overclock to the MHz.
While the Nvidia readout reads in increments of 15Mhz, you can adjust overclocking by a single MHz, and wattage by 0.1Watts in Linux. I suppose the same in Windows.
 
So it's important to keep a log of OC settings (GPU name, GPU frequency, GPU temperatures, and/or fan settings if you adjust them).
For every WU that errors, you'll have to decrease your OC settings.
Once you have found the GPU's stable settings, where it runs days without errors, you can increase the power slider to get higher boost frequencies, or leave them like above, and add another GPU.
Saving power allows you to add more GPUs.
My experience is your stable minimum wattage OC might not work when you raise the wattage by 20-40 Watts, and might need to be lowered a bit.
 
GPU-Z shows a lot of info on your GPUs as well!

Seems like your memory speed is displayed at half the speed it is on mine.
Take that in consideration when overclocking. The memory should be running at around 7000Mhz, and should work fine at 7500Mhz.