One of my rigs has 8 1070's that have been running nicely for almost 4 years. But something I have never understood is that overclocking the memory makes no difference to the hashrate. But increasing the core clock does. It's precisely the opposite of what is supposed to happen. My other rigs which have 1070ti's don't exhibit this behavior.
My 1070's are tuned to use around 110 watts hashing at around 28mh/s which seems pretty optimum but I only achieve this by leaving the mem clock alone and bumping up the core clock. All my rigs hardware are identical excepting the 1070ti's on the other rigs, which I have running at around 110 watts and 32.3mh/s by increasing the mem clock.
I haven't bothered asking this before because I doubt very much there is an answer. But has anyone experienced this and know what causes it?
My 1070's are tuned to use around 110 watts hashing at around 28mh/s which seems pretty optimum but I only achieve this by leaving the mem clock alone and bumping up the core clock. All my rigs hardware are identical excepting the 1070ti's on the other rigs, which I have running at around 110 watts and 32.3mh/s by increasing the mem clock.
I haven't bothered asking this before because I doubt very much there is an answer. But has anyone experienced this and know what causes it?
This has to do with the TLB bug on 1070s/1080s/1080ti's. Increase the core and power helps increase hashrate because of this bug. 1070ti is not affected because it was released later in the year with the "fix"
https://forums.developer.nvidia.com/t/blockchain-drivers/110847
"Here is the reply. Please note that I likely will not be able to respond to any follow-up questions about this. Therefore, if you ask a question, and I do not respond, that is the reason. I’m not permitted to share any more details than what is presented here:
We’ve confirmed that the performance drop is due to the size of the DAG exceeding the total on-chip TLB capacity on the Pascal GPU. As a result, there is an increased number of TLB misses, which affects performance. Because the TLB is a fixed capacity hardware resource, and the ETH algorithm accesses the DAG randomly, we don’t believe there are any software optimizations that could reduce the TLB miss rate.
In the Volta generation, TLB coverage was increased by 4x, and large DAG sizes (up to ~8GB, which won’t be reached for many years) will still fit in the on-chip TLB. So these newer GPUs (Volta and beyond) will show much less performance sensitivity due to DAG size."
Thank you! Finally, an answer that explains whats causing the issue. Even though there isn't anything to be done to fix it, I feel a lot better knowing the why.
