19:30:01
sech1:
RandomX v2 testing has started: https://gist.github.com/SChernykh/6058ecf01c929883b9d19c7eeadc8809
19:30:10
sech1:
More CPUs will be added soon (tm)
19:33:43
DataHoarder:
how does this feel on your orangepi sech1 :D
19:34:01
sech1:
I need to test it, actually
19:34:11
sech1:
It's heavily memory limited, it doesn't have L3
19:35:35
sech1:
I don't have a way to measure power though
19:36:02
DataHoarder:
usb power meter? :D
19:38:36
sech1:
I have a regular power meter, but it will show 4-5 watts, I guess. It's too rough measurement
19:40:49
DataHoarder:
ah. I have one that does mA and mV precision on their own
19:41:19
DataHoarder:
it is quite precise when I measured it against known samples HOWEVER you literally need to record the screen to get the data out
19:41:56
sech1:
I have go-pro to record the screen at 120 FPS :D
19:42:35
sech1:
Orange Pi RV2 (Ky X1 CPU, software AES): v1 99.3 h/s, v2 67.1 h/s
19:42:45
sech1:
AES hits hard when it's not supported
19:43:05
sech1:
XMRig could do 110 h/s on v1 though
19:43:24
sech1:
This thing has unstable hashrate between launches
19:43:50
DataHoarder:
start measuring 10s after starting :D
19:43:54
sech1:
Even XMRig alternates between 110 and 85 h/s on v1, after each launch
19:44:00
DataHoarder:
damn
19:44:14
sech1:
No, I mean hashrate can be either 85 or 110 in XMRig, and it's stable for hours
19:44:23
sech1:
But you need to restart it several times to get "lucky"
19:44:28
DataHoarder:
that's very odd
19:44:47
sech1:
Same behavior in randomx-benchmark, just with a bit lower hashrates
19:44:56
DataHoarder:
no identifiable cause?
19:45:02
sech1:
Probably memory allocation
19:45:09
sech1:
Maybe some parts of RAM are slower there
19:45:14
DataHoarder:
I wonder if it's due to how - yeah
19:45:25
DataHoarder:
like if some core section crosses a specific physical boundary
19:45:53
DataHoarder:
like scratchpad, allocated on big pages or similar but it's laid out across regions that cause that hit
19:45:59
sech1:
It's a 8 GB model. Maybe the upper 4 GB are slower
19:46:14
DataHoarder:
time to allocate it all
19:46:16
DataHoarder:
and test all regions :P
19:46:18
DataHoarder:
when efi randomx
19:47:06
DataHoarder:
maybe we can make a small kernel module to expose contiguous physical memory allocations via specific rules
19:47:45
DataHoarder:
afaik pagemap tells you where in physical memory your virtual memory map landed at
19:47:59
DataHoarder:
so you can start gathering data for when it's slow/fast via pagemap (can inspect this externally while it's running)