ccminer icon indicating copy to clipboard operation
ccminer copied to clipboard

4x1070 rig, 2 cards have half hashrate on x17

Open streaml1ne555 opened this issue 6 years ago • 10 comments

Using the latest version 8.21-cuda91 on a 4x1070 rig on Windows 10, 2 cards are identical. For some reason I'm getting half or less the hashrate on 2 of the 1070's on the x17 algorithm. I've tried playing with --cuda-schedule options and --no-cpu-verify but not having any success.

[2018-03-24 16:28:42] accepted: 2/2 (100.00%), 21.62 MH/s yay!!! [2018-03-24 16:28:43] Received new x17 block header [2018-03-24 16:28:43] block height 1979712, 0 transactions [2018-03-24 16:28:43] GPU #0: Zotac GTX 1070, 7610.05 kH/s [2018-03-24 16:28:44] GPU #3: ASUS GTX 1070, 7090.20 kH/s [2018-03-24 16:28:44] GPU #2: ASUS GTX 1070, 3067.58 kH/s [2018-03-24 16:28:44] GPU #1: MSI GTX 1070, 3066.29 kH/s

streaml1ne555 avatar Mar 24 '18 20:03 streaml1ne555

Actually, all 4 cards are pretty slow. With my single 1070 I get 8250 kH/s Looks like there's something wrong with your system. Too much heat? Weak power supply? I don't know

KlausT avatar Mar 24 '18 20:03 KlausT

Open air case and a fan profile that never lets them get over 60 degrees C. 850 watt corsair supply which is more than enough and the cards perform on par with other 1070's on other algorithms. The CPU is a Ryzen 7 1700 with 16GB of ram which is more than fast enough. Also verified that wasn't the bottleneck with --no-cpu-verify.

streaml1ne555 avatar Mar 24 '18 21:03 streaml1ne555

If you only mine with one card, is it faster then?

KlausT avatar Mar 24 '18 21:03 KlausT

Negative. Actually, it seems that X13, X14, X15 and X17 are showing slower results. I'm using Awesome Miner and have my prior good benchmarks to compare against. When I run a single "slow card" it's still consistently slow. Nothing has changed on this rig in the past few weeks except I just upgraded the nvidia driver to latest to try and help diagnose (it did not help).

streaml1ne555 avatar Mar 24 '18 21:03 streaml1ne555

For the sake of completeness I rebenchmarked in Awesome Miner, here are the results

Old Value:New Value

Groestl 130.45 MH/s : 131.22 MH/s Myriad Groestl 241.26 MH/s : 217.68 MH/s Quark 110.42 MH/s : 93.45 MH/s Qubit 107.28 MH/s : 76.58 MH/s X11 50.21 MH/s : 49.50 MH/s X13 39.16 MH/s : 28.93 MH/s X14 38.63 MH/s : 24.24 MH/s X15 35.73 MH/s : 24.97 MH/s X17 35.07 MH/s : 20.31 MH/s

streaml1ne555 avatar Mar 24 '18 21:03 streaml1ne555

That's wierd. I still think it's a hardware issue.

KlausT avatar Mar 24 '18 21:03 KlausT

I'll do the DDU dance and reinstall drivers again. Maybe something is screwball there.

streaml1ne555 avatar Mar 24 '18 21:03 streaml1ne555

Maybe try to remove 2 or 3 cards, and see what happens.

KlausT avatar Mar 24 '18 21:03 KlausT

This appears to be a PCIe bandwidth issue. The two cards that were performing better were on the x16 slots of the motherboard via 1x risers so running PCIe 3.0x1 . The others were on PCIe 2.0x1 slots. Ryzen's X370/B350 chipset provides PCIe 2.0 only. When I took the "bad" card and put it alone in the x16 slot I was able to get ~8200 KH/s. I guess the X algorithms send more data back and forth over the PCIe link?

streaml1ne555 avatar Mar 25 '18 02:03 streaml1ne555

Close this issue nobody uses KlausT for x17 anymore.

platinum4 avatar Dec 02 '18 09:12 platinum4