09:40:14 RPi4 does 100 h/s @ 5 W, even without AES: https://www.youtube.com/watch?v=Jiywpgk2wLk 09:40:18 so it's 20 h/s per watt 10:54:30 surprisingly good, but hardly competitive 10:55:07 unless electricity is free and you can get 140 of them for less than a cheap 3900x system 12:31:37 can rpis load with pxi boot? 12:31:45 or pxe. whatever it is 12:32:12 RockPro64 would be better, it has AES 14:32:02 Has anyone tested how hashrate scales with CPU speed (and the same memory)? 14:32:37 I've tested my 3700X at 2.05 GHz and got 4925 h/s, so it could do 9850 h/s at 4.1 GHz with very fast RAM in theory. 14:46:35 sech1: the only thing I have tried is 3900X at 3.8 GHz, 3.9, 4.0 and 4.1 GHz with 3600 micron E-die with settings slightly looser than recommended by ryzen dram calculator, was not stable with recommended settings 14:46:39 each 0.1 GHz increase results in an ~ 250 H/s increase 12,850 ; 13,110 ; 13,370 and 13610 H/s respectively 14:52:11 currently running at 4.1 GHz which results in an EDC of 100% of 140A that is what ryzen calls "peak current limit" 14:52:22 is that ok? 14:52:27 my vrm temps are only 63C 14:55:43 63 is like nothing for VRM. Quality VRMs can work normally at 100-110C 14:56:07 120 and higher is not good for VRM 14:59:58 but what about the EDC reading? 15:00:20 at some setting I tried it was 103% 15:00:57 that reading was from their ryzen master program 15:02:06 .btc 15:29:28 rpi is definitely not competitive 15:30:38 I updated gcc to 8.3 on my debian jessie arm64 box, to compile randomx 15:30:54 and it compiles and links, but at runtime it complains about missing glibcxx symbols 15:31:16 I wish people would stop writing fundamental libraries in C++ 15:32:27 and I don't have time right now to figure out how to build a new firmware image for this box based on a newer OS release 15:32:30 oh well 16:47:17 aha I could still update my system pointing it to archive.debian.org. upgraded to debian stretch now 16:47:54 too bad, only 2GB RAM, can only test verify mode 16:48:17 133ms/hash. 109ms/hash using largePages 16:48:44 so just under 10H/s for 1 core 16:50:02 31.3ms/hash with 4 cores 16:50:41 19.15ms/hash with 8 cores 16:51:04 RK3368, 1.2GHz 16:51:51 https://www.cnx-software.com/2015/11/19/geekbox-is-a-hackable-android-and-linux-tv-box-powered-by-rockchip-rk3368-soc/ 16:52:11 10W power supply but at least 5W of that is to power its USB-A port 16:52:45 thats pretty cool for non RX stuffs as well 16:53:46 it was my main monero node at home for a while, but since I let the OS go out of date I stopped using it 16:53:56 now it's up to date again I can restart that I guess 16:56:43 saw this guy in my feed last night: https://www.cnx-software.com/2019/12/08/rock-pi-sata-hat-targets-rock-pi-4-raspberry-pi-4-nas/ 16:58:42 going to have to bookmark that site and maybe add it to my rss feed 17:19:55 the problem of most ARM chips is insufficient cache for RandomX 17:23:22 yep 17:38:50 are you guys a fan of the orange pi stuffs over raspberry pi stuffs? not necessarily for mining. 17:40:28 haven't looked. odroid and pine seem to be good 17:44:38 i guess its more about what chip they are using. theres also the risc SoCs. 17:46:11 risc-v? didn't think there were any real products yet 17:46:50 all of these cheap SBCs are using vanilla ARM Cortex designs. at a given clock rate they're all basically the same 17:47:11 so the only distinguishing difference is whether they implement the full instruction set (e.g., including AES) or not 17:47:44 ya SiFive have a couple boards: https://www.sifive.com/boards/hifive1-rev-b 17:48:16 that domain doesn't resolve here. funny 17:48:50 anyway, risc-v is too fragmented. like 50% of the instruction set is in extensions 17:52:12 looks like my router's resolver config is confused. sigh 17:58:47 ah there it goes. 17:58:48 https://www.sifive.com/chip-designer#fe310 17:58:55 looks like a pretty weak chip. 320MHz 18:01:03 this might be usable https://www.sifive.com/boards/hifive-unleashed 18:02:07 ah i didnt see the specs on that one 18:03:22 still only 2MB L2$ on-chip, no L3. will be as crippled as most ARM SOCs 18:04:08 so there is a reason they dont have L3? too big? 18:04:17 yeah, too expensive 18:04:42 28nm process, plus they have an ethernet port on-chip. too big. 18:07:17 single 64bit memory channel 18:07:48 I don't see any reason to use one of these in production. ok for development if you're dead set on eventually deploying on risc-v 18:08:07 but I'd want to see better hardware to deploy onto 18:18:08 ya my buddy picked up a bunch for a group project that never came to fruition. not sure what he did with them. hopefully he flipped em as there was a waiting list. 18:19:38 have you heard about the gpu perf driver gains that AMD is supposed to drop early next year? i heard it will include architectures back to polaris. 18:42:05 it' only for rendering 18:42:12 not for compute tasks 18:42:27 ah ok 18:44:15 it's like automatic reduction of shader quality in "not important" areas of the image 23:52:22 Does RagerX follow randomX's license? 23:52:31 If you have a "referral code" to download, and anyone can get it any time with the code, does that bypass "open source" restrictions