..So I guess that means "no" ?
It means "complicated"
Here's an example: One of my systems that I've tested on is a new, nice Haswell-based machine from Dell.
CPU: i7-4770
GPU: GTX 650Ti
CPU hashing gets somewhere in the 330 c/s range using Yam. Maybe more, maybe less, but in that range.
GPU hashing gets almost exactly the same. I think I can sneak it a little higher in the next release, but for now, they're nearly identical.
The CPU costs $300. Not counting motherboard and memory.
The GPU costs about $140, 1GB of DRAM included.
The CPU draws about 70-80W when running full tilt. The GPU uses about 15-30W when running my code full speed.
You need one motherboard (~$120) and DRAM (~$40) for every CPU or two. You can instead fit 3-6 GPUs into a motherboard, reducing those costs further.
Also, my code doesn't really take advantage of the zoominess of something like the 650Ti. People have reported very good numbers from older cards. The memory bandwidth and number of memory channels is pretty important here, and that's not increasing as radically fast as core counts.
In total, running PTS on a GPU probably costs 1/4th as much as doing it on a CPU. That's not insignificant, but it's also not earth-shattering. I suspect that given the current price of PTS and the hash rates, mining PTS on a CPU will actually remain power-profitable, which is more than you can say for most other options (DOGEcoin on a Haswell CPU isn't too bad, though. *grin*)
As I said initially: This doesn't (yet) change the game as radically for PTS as it did for things like Bitcoin. The amount of memory accesses help level the game to a decent degree. But I'll note that I think there's more room for optimization in the GPU code than the CPU code, so the gap will widen a bit more.
Show Posts
