Is the jury still out, or was there a conclusion about 3950x that it tops out around 15KH/s?

Inge-: I saw one report at 16 KH/s

this was using 28 threads, 7 threads per CCX gives better results because of the L3 cache

so although you get 33% more cores vs 3900X you can only use 17% more threads

*shrugs* i.imgur.com/tELTwb4.png

nioc: limits of memory sustem, most likely

Look, a motherboard for the Up-to 64-core arm v8 tomshardware.com/news/huawei-introd…rd-for-kunpeng-920-armv8-processors

cool. too bad the US has bannedt huawei

hm, their desktop offering is weak though, only up to quadcore, only DDR4-2400, only PCIe3.0

strange to limit memory controller and PCIe when their server chips already support DDR4-3200 and PCIe4.0

Does it looks like a fake for you: tevador/randomx-sniffer #1#issuecomment-562083663?

look like*

it's always "shield and weapon" competition

one can detect the sniffer and stop mining while sniffer is working

that means you must poll for sniffer existence quite frequently though

or you can just hook all processes, replace GetThreadContext() code and "fix" registers for your thread

but that counts as OS modification though

but very simple to implement

and doesn't require kernel driver

I'm calling BS

bull...?

Joined GitHub on December 2, 2019

quite a perf/watt boost

no RandomX tests, disappointing

have you suggested them?

I wonder how the 32-core threadripper compares to the 32-core epyc

total miners 27247

4GB computers can't mine RandomX: reddit.com/r/MoneroMining/comments/…eed_na_na_na_hs_what_did_i_do_wrong

one more proof

sech1: I'm not sure I understand why 3950x hashes faster if you free up some MB of L2 cache from the CCD (ref reddit), isn't it 64MB cache for 32 threads as per spec?

mining with 4GB is possible on linux

Doesn't this indicate that the threadripper would be more hash/s/watt than a 3900x by a fair bit? phoronix.com/scan.php?page=article&item=amd-linux-3960x-3970x&num=12