NVIDIA Scrypt GPU Mining Performance with CUDAminer

Everyone "knows" that AMD GPUs are best for mining the various cryptocurrencies, and the conventional wisdom is that NVIDIA GPUs aren't worth the cost or the trouble. While this may be true from a pure performance perspective, if you already own such a GPU, the December update to CUDAminer actually delivered a pretty substantial boost in performance. To be clear, we're still only talking about roughly half the performance of AMD's similarly priced hardware, but there are other reasons to go with NVIDIA GPUs -- gaming and general computational programming are both strong areas for NVIDIA.

And it's not just about games; if you're looking for a PC that can still be used for other non-GPU-intensive tasks while mining, CUDAminer tends to be far less taxing in my experience (similar to high-end AMD GPUs at intensity 13). Noise levels also tend to be much better (quieter), and stability is also good. So if you have an NVIDIA GPU, what settings should you use and what sort of performance can you expect as a result?

Let's start with the settings. I'm going to focus mostly on Kepler-based cards, but Fermi cards may also work reasonably well. The key flag to use in order to generally get optimal scrypt hashing performance is the -l flag (--launch-config), where you can specify the architecture to use for the GPU. CUDAminer will try to find the "best" solution on its own, but you can usually get better results and faster startup times by using -l. There are six options: L (Legacy), F (Fermi), K (Kepler), T (Titan Kepler), S (a variant of the Kepler core), and X (experimental Titan). I didn't even look into using L/F/S flags, as it's generally agreed now that the GPUs that require those configurations won't run all that well.

As for the remaining options, T/X require Compute 3.5 cards based on GK110, which means GeForce GTX 780/780 Ti or GeForce GTX Titan. We can again safely skip Titan -- performance is roughly between the 780 and the 780 Ti, but the cost is currently well over $1000 -- which leaves the $500 (give or take) GTX 780 and the $700 GTX 780 Ti. If you have a non-GK110 chip, you can still get reasonable performance as well, and since I have a GTX 770 I ran some tests on that as well. The basic rule is this: every SMX unit in Kepler (and Titan/GK110) has 192 cores, so take your CUDA cores and divide it by 192 to find out what your configuration should be. Here's what you can expect (without playing with overclocking):

GTX 770: ~330 KHash @ 200W (-l K8x32)
GTX 780: ~510 KHash @ 250W (-l T12x32)
GTX Titan: ~570 KHash @ 250W (-l T14x32)
GTX 780 Ti: ~580 KHash @ 260W (-l T15x32)
GT 750M: ~75 KHash @ 35W (-l K2x32) (Note: this is a notebook)
GTX 760M: ~110 KHash @ 45W (-l K4x32) (Note: this is a notebook)
GTX 780M: ~240 KHash @ 100W (-l K8x32) (Note: this is a notebook)

Again, there's obviously no reason right now to go out and buy a ton of NVIDIA GPUs to run CUDAminer, but if you already have those GPUs around -- or if you're more interested in gaming and you just want to run CUDAminer when the PC isn't otherwise being used -- the GTX 780 should still generate a return of roughly $90 per month (after power costs). The GTX 770 actually has some rebates going on right now that can bring the price down to just $320, so basically we're looking at half the performance of AMD's equivalently price GPUs. That may sound bad, but prior to the December update to CUDAminer, it was more like one fourth the performance.

GPU Mining State of the Union Address

We're coming up on two months of highly inflated AMD GPU prices, thanks to the explosion of Bitcoin/Litecoin/Cryptocurrency prices. With MSRPs of $299, $399, and $549 for the R9 280X, R9 290, and R9 290X, you could pay for the cost of the GPU(s) in about three months, give or take, at which point any future mining is pure profit (after paying the power bill of course). There are very few investments that can completely pay off in three months, and even with the prices spiking up by 30% we're still looking at four months to recover the initial investment.

But that was the state of affairs last month; has the supply/demand situation improved at all? I did a quick check of online pricing for the pertinent GPUs and came up with the following:

R9 280X at Amazon: $400-600. The XFX card is currently in stock at $400. This is probably the best bang-for-the-buck right now, as you can break 700KHash pretty easily in my experience, so you pay $0.57 per KHash.

R9 290 at Amazon: $470-$620. While the MSI 290 card doesn't show as being out of stock at $470, notice the "typically ships in 1-2 months" part -- no way! There are only four (currently) Sapphire cards left at $490, while the Gigabyte R9 290 is more widely available but at higher prices. At $490 and if you can get as high as 850KHash (though 750-850 is "normal"), the $0.58 per KHash is potentially just as good as the R9 280X. If you only end up at 750 KHash, however, that bumps you to $0.65 per KHash.

R9 290X at Amazon: $630-$790. Obviously, paying over $700 for one of these is simply crazy -- you're better off with either of the above GPUs. Even at $630, where the Sapphire R9 290X is reasonably available, you'd still be looking at 800-950 KHash, which means you're paying $0.66-$0.74 per KHash (and you're probably going to be closer to $0.74 in my experience, as most 290X cards can't seem to hit more than 850KHash).

That's Amazon, but what about other options? Availability appears better at Newegg.com, but pricing is similar: R9 280X starts at $400, R9 290 starts at $500, and R9 290X will set you back $600 or more. Searching Google didn't turn up any other significantly better deals, though I did find quite a few "too good to be true" retailers that way so I'd suggest sticking with places you know you can trust. I'm sure we'd all love to get R9 280X at $230, but right now that's just not happening so be warned.

With all the AMD discussion out of the way, there's an interesting side note to be made: NVIDIA GPUs can mine as well, just not as fast as their AMD counterparts. The older version of CUDAminer wasn't all that great, but an updated releases optimized the CUDA code in December and you can now get at least moderate hash rates. How moderate? Well, it depends on several factors, with memory bandwidth being critical, just as with AMD.

I've run CUDAminer on a few GPUs, and what I'm finding is that with Kepler GPUs and GDDR5, you can get somewhere around 20-40KHash per SMX, with the higher end of that range more for laptops with GDDR5 and only a few SMX units. GT 750M for instance, using K2x32, can give me 80KHash/sec with GDDR5; GT 750M with DDR3 on the other hand only gets about 30KHash/sec. Obviously the profitability will depend on how high you can get your hash rate versus the power draw, but here's the quick solution to getting decent hash rates with NVIDIA Kepler: use "-l K[SMX]x32". You can find the number of SMX units on Wikipedia, or if you know your CUDA core count, divide that by 192. I'll have to do some additional benchmarking, but as an example I can run the following for 220KHash on a GTX 770M notebook that draws 140W:

cudaminer.exe -a scrypt -i 0 -l K5x32 -o stratum+tcp://stratum01.hashco.ws:8888 -u trogdorjw73.tester -p tester

It's not a ton of hashing power to be sure, but it's still around $1 profit per day (after electricity costs), so that's not too bad. I'm going to have to do some additional testing on NVIDIA hardware for the higher end GPUs, but if you could hit 400KHash/sec on a GTX 770/780 ($2 per day profit), and if you already have such a GPU, there's no reason to not run it when you're not playing games. I personally still prefer NVIDIA GPUs for gaming, and $370 for a GTX 770 is a pretty good price. It will be interesting to see if NVIDIA ever tweaks their future architectures to be more hashing friendly -- I wouldn't be surprised if Maxwell did just that after the crazy prices on AMD GPUs right now!