cryptozeny / cpuminer-opt-sugarchain

Optimized Cpuminer for Sugarchain using YespowerSugar (based on Yespower 1.0.0)
https://bitcointalk.org/index.php?topic=5177722.0
Other
72 stars 46 forks source link

RAM effect #22

Open cryptozeny opened 4 years ago

cryptozeny commented 4 years ago

telegram Doctor Who

系统都是win10 64位 i5 7500

这个是32G内存的 857 h/s image

这个是16G内存的 724 h/s image

这个是8G内存的 463 h/s image

cryptozeny commented 4 years ago

image

-t4

cryptozeny commented 4 years ago

@volbil @itwysgsl

zkyykj commented 4 years ago

内存提高了运算速度

cryptozeny commented 4 years ago

Alex on telegram:

All configurations with the exact same timings? Dual or single channel, how many banks in play?

Second, can this behaviour be reproduced in other machines (at least on chipset level)? That would be interesting to close out that it's BIOS/UEFI related and possibly to a sole manufacturer

@zkyykj

cryptozeny commented 4 years ago

SandelinosToday at 12:31 AM Okay I found some weird shit Apparently the amd B350 chipset is a lot slower at mining than B450 I tested a Ryzen 7 2700 and it got 1100 H/s with 3200MHz RAM While a Ryzen 5 1600 I tested on a B450 motherboard with slower ram got 1110H/s With 2 less cores and older architecture

crypt?zenyToday at 12:42 AM whats the point?:thinking: sounds strange mother board effect?

SandelinosToday at 12:45 AM Maybe I'm super low on battery but I got some testing done on ram speeds The hashrate scales pretty linearly with ram speed

@Sandelinos

cryptozeny commented 4 years ago

I get 1110H/s with 8gb dual channel on R5 1600 and only 570H/s with the same cpu with 8gb in single channel

by @sandelinos

image TL:DR you really really really want to run dual-channel ram Also faster ram and tighter timings help a bit but the difference isn't too big