Folding Forum

AnandTech has run the FAHbench Single Precision test on the RTX 2060 and it has scored higher than the GTX 1070 Ti and 1080 Founder’s editions and slightly slower than the RTX 2070 which it shares a stripped-down version of the TU106 core with.

$349.99US for 850,000 to 1,000,000 PPD at 150W power cap.

It seems we may have a new value king in the quest for the lowest GPU PPD/W

Well,when they come up for sale next week i plan an upgrade to the 1050ti in my second rig, upgrades to the rest of my bigger cards can wait for prices to get better

I have the GTX 1060, and unless you're planning on overclocking with no more than a few Mhz, it's not going to work as well for folding.
When gaming there's more tolerance to errors.
For folding, it's recommended to keep the cards running stock.
Even at a safe overclock (50% from the max safe overclock) on my GTX 1060, my card still errored some Work Units.
I wouldn't buy those cards with the intention to overclock them to fold.

Whose GTX 1060 do you have, and which model? Many of the models on the market from Gigabyte, EVGA and others come already factory overclocked, and it is the luck of the draw as to how much, if any, additional overclocking a particular cards will take.

By prices getting better I mean for the RTX 2070/2080, I won't be replacing my 1070, 1070TI and 1080 with anything lower than a 2070.

As for overclocking you are right, but since I am running them under linux I don't know how to overclock anyway, I let the card do its boost thing and thats it

MSI is very well tuned and hard to further overclock.
EVGA I can tune a bit more. Even though these cards come overclocked from the factory.
I found out the benefits of overclocking are virtually nihil; and may even be counterproductive (failing WUs, or lower life expectancy of the hardware).
Overclocking works a lot better in games or Bitcoin mining.
For FAH, each calculation made depends on the one previously made.
Mess up the first calculation, and the string of calculations after, may give an error, and the batch is considered 'faulty'.
FAH is more like a chain of calculations, rather than bitcoin where each calculation is only affecting a tiniest portion of the overall
If one block chain, one hash fails, no other block chains or hashes are affected.
Graphics cards process millions of hashes per second, and thousands of blockchains at a time.
Basically, if your card errors 0.01% of the time, is enough for FAH to fault a Work Unit (or, perhaps fault the part from the last save point before the error). With bitcoin or in games, aside from a small glitch (and possibly a few points) the remainder of the batch goes through fine.

The RTX series is optimized for Compute workloads. That is why RTX 2060 matches GTX 1080

ProDigit wrote:I have the GTX 1060, and unless you're planning on overclocking with no more than a few Mhz, it's not going to work as well for folding.
When gaming there's more tolerance to errors.
For folding, it's recommended to keep the cards running stock.
Even at a safe overclock (50% from the max safe overclock) on my GTX 1060, my card still errored some Work Units.
I wouldn't buy those cards with the intention to overclock them to fold.

Not my experience with my two EVGA Gtx 1060 6GB models. Sounds like you lost in the Silicone lottery.

foldy wrote:The RTX series is optimized for Compute workloads. That is why RTX 2060 matches GTX 1080

And the same could be said for the previous generation. The GTX 1060 is a pretty close match for the GTX 980 and the GTX 760 is a pretty good match for the GTX 580 and the RTX 2060 is a pretty good match for the GTX 1070 T (with respect to OpenCL applications like FAH). If there also has been a die shrink, there will be power savings, too.

Not exactly, of course, but that's what they do when they come out with a new generation.

I was planing to get a 2060 but its still not listed in Denmark. Looking forward to some real world results.

I have Strix GTX 1070 @ 2000 MHz @ Linux Mint.
At Linux FahBench it draws 130 ns/day, but card's PPD nowhere near to other cards with more CUDA cores and with ~130 ns/day and only 800k @ 2000 MHz.
Keeping in mind that 2060 have same 1920 CUDA cores as 1070 but with power draw at stock boost 1680 MHz much closer to power draw of stock 1070 Ti - which draws 30-40 watts more than stock 1070 - so it may be not so rosy ppd/watt king as one could suppose from FAHBench data only without looking for real average PPD and power draw.
Strix GTX 1070 @ 2000 MHz at Linux Mint eats 145-165 watts depending on WU.

gordonbb wrote:

ProDigit wrote:I have the GTX 1060, and unless you're planning on overclocking with no more than a few Mhz, it's not going to work as well for folding.
When gaming there's more tolerance to errors.
For folding, it's recommended to keep the cards running stock.
Even at a safe overclock (50% from the max safe overclock) on my GTX 1060, my card still errored some Work Units.
I wouldn't buy those cards with the intention to overclock them to fold.

Not my experience with my two EVGA Gtx 1060 6GB models. Sounds like you lost in the Silicone lottery.

Seems like it's temperature related.
I live in South Florida where the temps are always above 75F (24C).
Last night the temps dropped to the 50s (70F indoors), and my OC margin doubled.

There's still too little info on power consumption of these newer RTX cards available, while folding.
It's a different number than stress testing, or gaming.

The RTX 2060 has almost a quarter less Cuda cores, at the same GPU frequency, but is equipped with GDDR6 VRAM, that's almost twice as fast as the GDDR 5 VRAM of the 1070Ti.
14Gbps vs 8Gbps. It accounts for something.

So which is the best bang for the buck, of the two?
It would be nice to have someone compare PPD with Watt usage, and see if owning an RTX benefits a GTX power wise.
I guess the lowest cost card, makes most sense for people folding occasionally, and the higher cost card, for those who fold 24/7


On a side note,
My PNY RTX 2070 arrived broken
I'm waiting for another month, until AMD releases their cards.
Hopefully we'll see a price drop on the Nvidia cards, and some PPD scores on their new 7nm cards soon!
In the mean time, I'm sitting on my $500 cash pile, it's not going anywhere until some or another good deal comes along!
Cards with 1+M PPD, are still too expensive.

One of my local Bricks & Mortar stores had them in-stock for a better price than new egg so I picked one up.

Initial impression ~ 20% better than a 1070Ti

After removing the cuda toolkit and installing 415.27 from the ppa ...

Running a 14156 currently and the Advanced control shows 1,091,520PPD at 38% of the way with a TPF of 1:46

HFM.net shows current results for this WU
2070 1300kPPD
1070Ti 881kPPD
1060 6GB 476kPPD

Running with a RTX2070 in a Gigabyte z390 SLI on Ubuntu 18.04 LTS (x8/x8)
Haven't adjusted the fan curves yet on the motherboard.
FanControl for the GPU does not work from the command line using nvidia-settings on 415.27 for the 2060 but now does for the 2070. I borked my X and couldn't be bothered to fix it yet. Running at 65C ~44% Fan
1950MHz Shader with a +75MHz O/C
210W Power-Limit set but averaging ~157W

Mstenholm wrote:I was planing to get a 2060 but its still not listed in Denmark. Looking forward to some real world results.

I finally received my EVGA RTX 2060 (XC Ultra) and can report a consistent 1.1M PPD on Linux using NVIDIA's 415.27 drivers. The first WU came in before I had set client-type = advanced and only netted 450k PPD. I didn't catch what project that WU came from, but I'll make a note and report out if it happens again.

Nathan_P wrote:By prices getting better I mean for the RTX 2070/2080, I won't be replacing my 1070, 1070TI and 1080 with anything lower than a 2070.

Well you might be missing out then.
Here are my yields under Ubuntu 18.04 LTS in a dual card (z370) rig with a small overclock
Card MSRP PPD Limit
EVGA 2070 XC Ultra $600 1.3M 170W
EVGA 2060 XC Ultra $380 1.1M 150W