For those that haven't read all my posts on the subject, I attempted to water cool my original EVGA GTX 1660 Ti XC Gaming card.  Unfortunately, the hole spacing on the 1660 Ti PCB changed to a non-standard spacing, so I was left 1 option...., carefully drill out the mounting holes to make the card work with existing GPU brackets.  Okay, so I had 3 options, but 2nd option would be too costly, and I didn't like the third option at all which would have been to stay with air cooling.  Bottom line, I killed the card, RMA/warranty not an option due to making physical changes to the PCB.
 
So after some questions, and contemplation, I ended up going back with a 1660 Ti instead of getting a 2060 Super (or 2070).   This time I went with a Gigabyte Windforce card - GV-N166TWF2OC-6GD - for the improved cooling, same factory clock speed as the EVGA card.
 
It appears to have worked, at least so far as synthetic testing goes.  With the EVGA card, I was able to run Firestrike at just 2055MHz boost clock, the Gigabyte card ran it at 2130MHz boost clock!  Both cards finished Firestrike at about 58C.  
 
I'll start up folding again later tonight, this time around, I'm going to try out different driver versions to see if any driver is better than the others.  For the driver test, I will be running factory stock speed, and 90% target power for all driver versions.  I'll also discard any folding data from work units that are paused while using computer - something I did not do when I posted the results from the EVGA card.

Cool GTX
looks like a silicon lottery winner  2130 MHz
 
Looking forward to you updates

Indeed, ran Firestrike again, boost clock peaked at 2145Mhz for about 5 seconds into the benchmark, with an average clock rate of 2094.4MHz and peak temp of 58C.  BUT, that's at 147Watts according to GPUz.  Highest wattage I saw on the EVGA card was just 132 Watts.
 
Will be a few weeks till I get folding data across the various driver versions, but I'll post 'em once I have 'em.
 
In the meantime, here's my Firestrike run "at 2145Mhz" https://www.3dmark.com/fs/20383914 using driver 419.67

Some quick results:  only doing about 2 - 3 days per driver, still getting some work unit overlap.  Probably should do a week per driver, but I tend to get impatient....
 
Nothing notable so far....

 
All the results are less than 2% difference across same work units, with the lone exception of project 14184, so statistically, looking at identical performance, with 1 anomaly.  
 
419.35 does appear to have best performance on several work units so far, but not a significant difference.

DrSchmidt
Hey SM-71.
Can I ask where you get your data formatted for your table above?  I usually have to dig through my logs on the C drive in Windows 10; am wondering if I'm missing something easier.

Check out the sticky topic "Client Monitoring Application" https://forums.evga.com/Client-Monitoring-Application-for-the-FoldingHome-with-HFMNET-m2642139.aspx  I use HFM.NET to collect the folding data, then export the history view to CSV, and import that into a spreadsheet, lastly, I delete the columns that aren't really necessary, and sort it however I need it.
 
When I setup my HFM configuration file, I usually just use the driver version and Windows version as the "Name" for easy future reference.  

What's the lowest power cap you can do on those?
I recently found out on the RTX 2060 and 2070, it is 125Watts; and both cards (though running overclocked at P2) remain in P0 state boost speeds.
That's 1.03M PPD at 125W for the 2060, and 1.136MPPD for the 2070 (about 10% faster than running at my prior 136W, due to the GPU running below 55C instead of 60-65C).
 
That's a PPD over Power ratio of:
2060: 8.24
2070: 9.088
 
I would be interested in the 1660Ti numbers, for perhaps future projects.

Wow, thanks for all the Testing & Posting your results

Thank you for posting these results. I am curious if anyone has results in Linux os

Once you'll hit a power level where you'll hit 50c, and running at max overclock, the performance won't be able to increase more.
As you may have noticed, performance increases with a higher power cap, but there's more at work than just that.

On RTX cards, there are a couple of limitations that play a role in optimizing a GPU for folding:
1- the min power (watt) limit the card can handle
2- the max GPU P0 frequency the card can handle (boost frequency)
3- the max overclock frequency the card can handle (overclock frequencies can slightly increase as power consumption goes down).
4- The max L2 boost frequency depending on temperature.

Number 1 is fixed for most cards and determined by Nvidia.
Eg, RTX 2060 to 2080 all have a minimum of 125Watt. (Which doesn't always mean they're running .let efficiently at these settings).

2 usually when starting a wu, with the stock settings, max power consumption (reverse power cap), and at a temperature of below 45c, the card will run in P0 power setting. In this setting, it'll show the max boost frequency that the card can run safely.
Most cards can handle 2010Mhz fine, but heavier cards are usually capped at 1995, 1935 or even 1810Mhz.
It will only be shown the first 10 or so seconds. This frequency is your reference frequency (frequency you'll want the card to run at continuously).

3 this is very card dependent, and no set rule that cards with higher overclock, or even cards with higher boost frequency, will therefore fold faster. Some cards I own, fold just as fast as others of the same kind, using much lower reported boost frequencies, and/or overcloks.
Higher overclock cards could indicate lower from the factory overclock.
But it's safe to say, that whatever overclock works at stock setting, a higher one will work at a power capped card. The higher the power curve, the lower the overclocking capability of the card, as the closer it'll run to the GPUs max boost frequency.

4- Once you got a certain power cap, the GPUs temperature will fall below 50c.
When it does that, and you haven't yet reached the card's max GPU boost frequency, but you already are running the highest safe overclock levels (eg: +100Mhz), and you'll lower power consumption further from there, the cards performance will drop, without a way to compensate with overclocking, and you'll be below the GPU's efficiency range.
Nvidia made it so, that cards running below 45-50c will run at P2 speeds similar to P0 boost speeds.

I think the best course for action, in reference to tuning a card, would be to:
1- record P0 boost frequency
2- power cap to the lowest power setting
3- find highest overclock until wu records an error
4- if P0 frequency hasn't been reached yet, and at an overclock setting just below WUs fail, lower overclock frequency by a few tens of MHz, and increase power consumption profile, until values close to P0 boost frequency are attained.
5- sometimes, running at 35-100Mhz lower than boost, can increase efficiency, meaning losing +10% watts at the cost of only <5% of PPD can be beneficial. But sometimes lowering power consumption too far, can affect performance negatively.

I found on the RTX cards, that overclocking the vram, would pull down the GPU frequency.
All my cards fold with vram at 15Ghz, which is a +1Ghz overclock.
Any faster, and the PPD goes down.
I measured this by looking at avg PPD running 14 to 16Ghz vram speeds, which gave exactly the same score. Higher than 16, and lower than 14 GHz also gave lower score.
So I figured the sweet spot is between 15-15.5Ghz.
But since the speed difference was minimal or immeasurable, I kept running at 15Ghz, hoping the cards would last longer.

If I compare your results to mine, I get below 55c at 125Watt with the RTX 2060 card, with smallest cooler. I would say you should be able to optimize case cooling a bit better.
Like, switch to an open case design for folding.
That way, the card will run cooler, will run even higher boost, and as a result you can lower watts even more.

BTW... currently folding at 150MHz overclock and 60% power target.  Boost clock has dropped significantly, only folding at about 1830 - 1845MHz, while running around 77 Watts @ 46C according to GPU-Z.
 
Nice to know I can drop 50 Watts and still run if needed.

That is a really nice list of test results. While I am using a 2070 it's good to know it probably doesn't matter what driver I use.

It would be nice to see what other GTX 1660 Ti owners are seeing with this card.
@SM-71 I will pick up a EVGA GeForce GTX 1660 Ti XC ULTRA GAMING, 06G-P4-1267-KR to see how we compare with Folding.
@SM-71 What Driver Version should I install to best compare our cards under?
Current Driver I use is the Studio Driver 431.86, but now it is beings said that the Driver makes No Difference when Folding.
Bummer getting this Error on Order:
PayPal transaction failure. Please try again later on. (Error) : The remote server returned an error: (400) Bad Request.
Got it to go through, so next weekend I can do some testing to compare our cards.
I will PM you my Findings.
I will use your  code when I register this card.
 
I should add that this Graphic Card is a GTX Graphics Card and has No Real-Time Ray Tracing Hardware so it is not a RTX Graphics Card and what applies for an RTX Graphics Card would not really apply to this GTX Graphics Card
https://www.nvidia.com/en-us/geforce/graphics-cards/gtx-1660-ti/

I cleared the Cache, History, and SSL State and logged back in and it worked, thanks.

Now Testing the EVGA Version of the NVIDIA GTX 1660 Ti Graphics Card.
 
Z370FTW  i7-8700K @ 4700 MHz

12 hours later.


Going to give this card as gift to who could use it that is on both Teams.