Help with overclocking 1080Ti?

I’ve just put together a small mining rig consisting of 2x ASUS 1080Ti Strix O11G, waiting for an upcoming offer before I buy and install another four cards. Now, before I built this rig, I tried mining on my desktop single GPU GTX 980.

On my regular desktop I experimented with overclocking, and I got to the understanding that just regular overclocking through MSI afterburner etc. wouldn’t get me what I wanted, namely increased speeds in P2 state. I’ve tried every possible tweak to make the GPU mine in P0, but I haven’t figured it out. In the Nvidia Profile Inspector, there was supposed to be a “disable P2 state” or something, but the closest I have is the “Power Management Mode”, which does nothing to the P-state.

So in order to overclock my old 980, I overclocked the GPU/memory in the P0 state first (using NVIDIA Inspector), and then I set the P2-state to the new boost speed to match the P0 speeds. I did all this manually and created some shortcuts on the desktop to quick-clock after restart.

Now, I had hoped to create something more automatic on my new 1080Ti-rig, and I’m not sure where to start. Firstly, I can’t seem to make sense of the clock speeds of my new cards out of the box. The clock speeds shown in the Nvidia Inspector doesn’t match with what ASUS claims as specs or what the boost speed is supposed to be (screenshot attached).

According to the ASUS website, my card has a base clock of 1569MHz and can boost to 1683MHz at default settings. During mining, the NVIDIA Inspector and GPU-Z claims that the core is running at 1923-1924MHz. This is while mining and is way above what it is supposed to be. Also, NV Inspector shows that my memory runs at 5505MHz (which should be correct as the mem frequency is double that at 10010MHz). GPU-Z, however shows my mem speed at 1251.5MHz (which could be just how GPU-Z presents the number, as it is 1/8 of 10010MHz).

Now, I’m a bit confused (to say the least) about how to overclock this correctly. Do I need to use NV Inspector? Is there an easier way of just increasing core/mem-speeds and auto-clock at startup (together with starting the miner).

I’m not really complaining about the hash speed out-of-the-box (720-730 sol/s on each card using EWBF miner v0.3.4b), but I would like to see what speeds I can get, even trying to make the cards draw a little less power (currently at 260-270W each).

Are there anybody out there that can shed some light on my problem and nudge me in the right direction? Or better yet, just tell me what I should do?

GPU-Z statsGPU-Z sensors

Nvidia GPU BOOST 3.0 is what is driving your core clock above the advertised stock clock speed. The frequencies you report are pretty common on 1080 Ti’s (My EVGA ran ~1910 MHz, for example, with appropriate temperatures).

Regarding memory speed…GPU-Z is showing you the actual memory speed. the 10 GHz you’re quoting is just an effective memory speed. If you take the GPU-Z frequency and multiply by 4 for quad-data rate chips which your GDDR5X are, and then you add in another factor of 2 (for 2 operations per clock cycle (s^-1)), you get your large effective frequency quoted.

I haven’t played around too much with the GPU states, to be honest, but I would imagine the option for CONSISTENT PERFORMANCE is what you’re looking for. I have been running mine with modest +100/+100 Core/Mem clock offsets and getting around 740-770 sol/s, but I left them at 100% TDP for the time being to see how stable these settings are (going for 4 days so far).

One thing to keep in mind…if you’re just now getting a 6x Ti rig up and running, you’re looking at a very, very long ROI in all likelihood. I’d overclock with caution since you’re going to need them to run 24/7 for a year or more to get your money back.

Thanks for the reply. I figured that there had to be something with how GPU-Z and NV Inspector reports the memory speed, thanks for explaining. Now, about the ROI, I am aware that there should be at least 3/4 of a year to break even (last I checked). But then again, the next cards will be bought with a 20% discount (a special Mastercard discount in one of the local electronic stores, they do 5-6 weekends each year). And in a year or so, the cards are still good for resale. And then there is the everlasting hope of an increase in Zcash value. There are uncertainties, yes, but unless Zcash crashes completely, I guess the risk will be acceptable.

I just did some experimenting with overclocking. I only changed the P0 state in NV inspector, and I got them running at +120 core, +700 mem and at 80% TDP. Temps are about 63 centigrades, and each card draws around 210-220W calculating appr. 730 sol/s. That gives them around 3.3-3.4 sols/W. Think this is something I could run 24/7?

EDIT: Looking at the core speed after adjusting with +120MHz, the current speed has dropped from 1924 before overclocking to 1860MHz after (have I actually downclocked the core by increasing the offset?). Memory runs at 1425,6MHz x 4 x 2 = 11405MHz (which would be an increase of 1400MHz (effective frequency).

EDIT#2: I just realised that the lower core speed probably is due to a decreased TDP. If I increase the power target to 100%, the core speed increases to about 2000MHz. And that gives me about 775sol/s, but unfortunately a dip in efficiency to 2.7-2.8 sol/W. I’ll probably leave it at the lower setting for longevity.

Great! WIsh I had that kind of deal at my local electronics store! :slight_smile:

I’ve seen those overclock numbers thrown about before, but typically close behind is a discussion about the relative robustness gap between Samsung and Micron memory. Anecdotally, Samsung memory tends to lend itself to more aggressive and stable overclocks, but individual results always vary . The other thing is, you really need to write down what your hash rates are and overclock in a methodical fashion. If you’re getting approximately the same results with a lower overclock, I’d encourage you to think about just letting it go at the lower settings. I personally didn’t notice much improvement going from +100/+100 to +100/+200…but again, I can’t speak for all cards.

Commenting on your efficiency (3.3-3.4 sol/W) and your HR (730), I think those are pretty respectable numbers…but no one can definitively predict how stable those settings are. You’ll just have to see how your individual cards perform.