The graphrate for the 1070 quoted in that table is for cuckatoo30, while the others are for cuckaroo29. There is an expected slowdown for cuckatoo30. We should maybe split up the table on the wiki according to algorithm.
cuckaroo is slightly slower than cuckatoo. perhaps around 5%
for benching purposes, it’s better to use cuckaroo, as that’s the main GPU PoW.
In most other algos the performance increase from 1070 & 1070ti is anywhere from (5-20%). It’s usually closer to 5% than 20% and can be more dependent on other factors like OC/ brand)
Unless I’m reading this wrong; Tromp is saying below that the decrease in performance between cuckaroo and cuckatoo is only around 5%, so that still wouldn’t explain the reported 101% increase in GPS between the 1070 & 1070ti.
You missed the fact that the cuckatoo30 graph is twice as large. 1070 TI has always been 10% faster.
Arhh I’m with you now! Thanks for clarifying.
RTX 2070 support on linux is good enough to get mining using the standard methods? I’m trying to think of any reason not to start a 2070 rig.
How do you guys think, hat is the best option for now?
i would consider to purchase a used gtx1080ti to be able to use the c31 mean miner (from grin-miner)
and for the platform (cpu+mb), maybe a mobo with amd x399 chipset and the threadripper cpu looks to be the best bet. there are x399 mobo’s with 4x pcie16 slots, each connected directly to cpu.
has anyone tested the cuckatoo_mean_cpu_avx2_31 plugin on amd threadripper ?
I suspect that any 8 GB AMD card will end up being a better value than the 1080Ti but the current miners don’t do C31 well in under 8GB… yet.
Why do you suspect this? It seems AMD cards are way behind Nvidia in both C29 & C31. Even more so in C29.
Meanwhile the calculators says that C31 is more profitable anyway.
Only in raw hashrate, but Nvidia cards are also expensive. You must consider the capital cost and power cost as well. If you can get 25% of the hashrate for 10% of the cost, it’s a good deal. I haven’t done all the math, but generally Nvidia charges a premium for being the absolute fastest, although usually AMD is faster per dollar.
Also consider all the miners so far seem to have focused on Nvidia development first, so we might expect AMD speeds to see greater improvement.
Are you still developing an Nvidia/ AMD miner? The typical 1.5-1.6GPS of an 8gb RX580 is very disappointing, how much head room could there be?
i tend to agree with @timolson that amd cards are most of the time on a par with nvidia in price/performance or performance/watt comparison. especially for memory intensive algos. nvidia is ahead when the algo requires “raw” gpu performance. but this might change slightly with 7nm vega as well.
Yes, but I haven’t looked at AMD at all yet. Nvidia has better performance profiling tools so I’m starting there.
1 Like