You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
RandomX v1 hash: 4194304 VM instructions + 262144 AES = 4456448 total ops
RandomX v2 hash: 6291456 VM instructions + 524288 AES = 6815744 total ops (+52.9%)
RandomX v2 includes the following changes:
Program size increased from 256 to 384
CFROUND is tweaked to change the rounding 16 times less often
Group F and group E register mix is done using 16 AES operations (the extra 262144 AES ops come from this change)
Dataset prefetch is 2 iterations ahead instead of 1
RandomX v2 hash is 1.5 times "heavier" than RandomX v1 hash, so hashrate comparison is only for the reference. The real metric is work per Joule (the last two columns).
Results are obtained with randomx-benchmark tool and HWInfo64 to monitor the CPU power consumption. Large pages on, MSR mod on (XMRig was used to turn it on).
@Fountain5405@Gingeropolous please re-run your tests with the latest code. One more RandomX v2 tweak has been added, hashrate and efficiency increased a lot.
The exact power usage is not very important for the "Relative Work/Joule" column because it cancels out. The power draw of the whole box was 140 W at the wall, but that includes a GPU, SSD and RAM.
Here are the tests on my laptop:
AMD Ryzen AI 9 HX 370 (Zen 5), DDR5-5600
Algorithm
Hashrate
Relative Speed
Relative Work/Joule
RandomX v1
6597.15
100.0%
100%
RandomX v2
7121.69
108.0%
165%
The TDP is 28 W. HWInfo was showing package power around 26 W in both tests.
The exact power usage is not very important for the "Relative Work/Joule" column because it cancels out
Yes, I already observed that if power usage stays the same, relative numbers don't depend on the absolute wattage value. I'll keep it at 88 watts just to be able to fill in all the fields.
(-- deleted because i didn't understand the assignment --)