08:21:46 <sech1> Something new to tinker with: https://github.com/xmrig/xmrig/releases/tag/v5.2.0
08:22:06 <sech1> 1GB pages, HW prefetcher control on Intel
08:22:10 <sech1> but only on Linux
08:23:44 <Inge-> wait, does it not require bios changing of prefetcher behaviour?
08:26:02 <sech1> No
08:26:07 <sech1> it's command line switch now
08:26:26 <sech1> https://xmrig.com/docs/miner/randomx-optimization-guide#intel-specific-optimizations
08:26:50 <sech1> It changes the same MSR register as BIOS does
08:29:44 <Inge-> nice
08:30:06 <sech1> 1GB pages give +1-3% depending on CPU
08:30:10 <sech1> +1% on Ryzen/Skylake
08:30:15 <sech1> +3% on AMD Opterons
08:36:01 <Inge-> so running as root and adding --randomx-wrmsr --randomx-1gb-pages should enable this stuff?
08:38:12 <Inge-> or rather, don't need to run as root as long as huge pages have been configured ...
08:39:07 <cohcho> It probably changes the same since there is no proof.
08:39:35 <Inge-> hm.  * 1GB PAGES    disabled even with --randomx-1gb-pages
08:41:28 <sech1> 1GB pages are only available on Linux
08:41:37 <sech1> and you need to configure them
08:41:54 <sech1> not all CPUs support them
08:42:36 <Inge-> this is a configuration change of how Huge PAges are set up then I presume?
08:44:25 <sech1> The easiest way to enable 1GB pages on single CPU systems: GRUB_CMDLINE_LINUX_DEFAULT="default_hugepagesz=2M hugepagesz=1G hugepages=3"
08:44:45 <sech1> and then sudo update-grub and reboot
08:56:01 <Inge-> so checking /proc/meminfo after this, it is correct in still reporting 2M size?
08:56:17 <Inge-> but if I run as root, it will configure a few 1GB pages
08:56:59 <cohcho> check number of 1gb pages with `cat /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages`
08:58:07 <sech1> or run "hugeadm --pool-list"
08:58:59 <cohcho> There is no "hugeadm" here.
09:00:12 <sech1> sudo apt install hugepages
09:00:16 <sech1> or hugepage
09:00:24 <sech1> I don't remember exact package name
09:00:52 <Inge-> yeah found it
09:00:53 <Inge-> c# cat /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages
09:00:54 <Inge-> 7
09:01:03 <Inge-> (dual cpu so I sent 6 there instead of 3)
09:01:14 <sech1> And the correct path for NUMA is /sys/devices/system/node/node0/hugepages/hugepages-1048576kB/nr_hugepages
09:01:26 <sech1> node0, node1 and so on
09:01:32 <Inge-> and node1 for a dual socket ... right
09:02:03 <sech1> xmrig writes to this file to allocate 1GB pages, it should work under root
09:04:01 <Inge-> hmm. still says disabled
09:04:21 <Inge-> Waiting on an apt upgrade to complete so I can install hugepages package
09:05:45 <sech1> What CPU is it?
09:06:28 <sech1> And how exactly do you try to enable it?
09:07:25 <Inge-> Dual E5-2670 (v1)
09:07:39 <sech1> it should support 1GB pages
09:07:41 <Inge-> modified /etc/default/grub and ran update-grub and rebooted
09:07:49 <sech1> Xeon E3 series don't have support
09:07:53 <Inge-> then ran sudo bash -c "echo 3 > /sys/devices/system/node/node0/hugepages/hugepages-1048576kB/nr_hugepages"
09:07:57 <sech1> you also need to modify config.json
09:07:58 <Inge-> and sudo bash -c "echo 3 > /sys/devices/system/node/node1/hugepages/hugepages-1048576kB/nr_hugepages"
09:08:05 <sech1> find "1gb-pages" there and set it to true
09:08:17 <Inge-> appended these to commandline: --randomx-wrmsr --randomx-1gb-pages
09:08:24 <Inge-> not using config gile
09:08:26 <Inge-> file*
09:09:40 <Inge-> # hugeadm --pool-list
09:09:40 <Inge->       Size  Minimum  Current  Maximum  Default
09:09:40 <Inge->    2097152     2560     2560     2560        *
09:09:41 <Inge-> 1073741824        6        6        6
09:09:48 <sech1> can you paste full miner console output?
09:09:57 <Inge-> kk
09:12:19 <Inge-> http://paste.debian.net/1120623/
09:13:04 <Inge-> commandline: ./xmrig -o pool.supportxmr.com:5555 -u wallet -p xyz:bvla⊙nn --coin monero --randomx-wrmsr --randomx-1gb-pages
09:13:39 <sech1> well, command line doesn't work apparently :D
09:13:42 <sech1> try with config.json
09:13:46 <Inge-> hehe ok
09:13:54 <sech1> take default one from src folder and enable it there
09:14:01 <Inge-> hm. these cpu's are L2 cache limited it seems. unexpected
09:14:56 <sech1> I think you forgot to add a number after wrmsr in the command line
09:15:05 <sech1> try ./xmrig -o pool.supportxmr.com:5555 -u wallet -p xyz:bvla⊙nn --coin monero --randomx-wrmsr 6 --randomx-1gb-pages
09:15:59 <sech1> yep, confirmed
09:16:09 <sech1> you forgot to add 6 after --randomx-wrmsr
09:16:28 <Inge-> a) it works with config.json
09:17:28 <Inge-> b) adding the 6 also enables it from command line
09:17:50 <sech1> nice
09:17:56 <sech1> so what's the hashrate difference?
09:18:19 <Inge-> it was not clear for me that I needed the 6 by looking at the releases here: https://github.com/xmrig/xmrig/releases/tag/v5.2.0
09:19:07 <Inge-> currently it looks like 0.5%
09:19:16 <Inge-> 6664.8 vs 6630
09:19:28 <Inge-> note that I already did prefetcher changes in BIOS
09:20:12 <sech1> Sandy Bridge Xeons should've got the most speedup from it. Maybe they're already limited by CPU and memory access is fast enough.
09:20:44 <Inge-> now running without root, and it says [2019-12-11 10:20:03.610]  rx   cannot set MSR 0x01a4 to 0x0006
09:21:09 <Inge-> not sure if it warns if those settings were already made in BIOS when run as non-root
09:21:23 <Inge-> hr is identical
09:21:36 <cohcho> xmrig doesn't revert msr value
09:21:42 <cohcho> so it's 6 from previous run
09:21:48 <Inge-> it still warns however
09:23:21 <cohcho> It warns since read/write msr requires root and xmrig can't update or read msr.
09:25:47 <Inge-> ah it can't read it either ...
10:01:23 <sech1> Someone actually tried to mine with GPUs: 6xRX 580 - 2800 h/s, 400 W https://youtu.be/DnLE48xi2k8?t=50
11:12:51 <Inge-> Vega cards mining as we type: [2019-12-11 12:11:48.415] speed 10s/60s/15m 8811.7 8824.8 8828.7 H/s max 9318.2 H/s
11:51:05 <wow-discord1> <sech1> https://www.reddit.com/r/MoneroMining/comments/e962fu/9526_hs_on_ryzen_7_3700x_xmrig_520_1gb_pages_msr/
12:00:55 <Inge-> I wonder if the 3950x can get higher with the msr and 1gb page tweaks:
12:01:00 <Inge-> https://www.reddit.com/r/MoneroMining/comments/e7dm1p/amd_3950x_final_results_of_hashrate_after/
12:01:58 <wow-discord1> <sech1> Yes it will. Prefetchers read more stuff from memory than needed so turning them off always improves hashrate.
12:02:13 <cohcho> sech1, What was the default value of those registers?
12:02:23 <wow-discord1> <sech1> I didn't check
12:02:31 <wow-discord1> <sech1> I can reboot now and check
12:03:29 <cohcho> How long have you bruteforced those registers to find 0x510000 and 0x1808cc16?
12:04:48 <wow-discord1> <sech1> I didn't, they're from l1/l2 prefetcher options on some AMD BIOS
12:05:40 <wow-discord1> <sech1> default values: MSR 0xC001102B = 0x2008CC17
12:06:02 <wow-discord1> <sech1> MSR 0xC0011022=0xC00...0002500000
12:06:44 <wow-discord1> <sech1> so it changes 2008CC17 -> 1008CC16
12:06:55 <wow-discord1> <sech1> and 0x500000 -> 0x510000
12:09:27 <wow-discord1> <sech1> xnbya found these options in BIOS and just took register values from there
12:11:37 <cohcho> xnbya: Have you looked into firmware asm or just advanced options in UI?
12:41:55 <sech1> https://webcache.googleusercontent.com/search?q=cache:7rEwStD4ffoJ:https://github.com/coreboot/coreboot/blob/master/src/vendorcode/amd/agesa/f12/Proc/CPU/cpuRegisters.h+&cd=1&hl=en&ct=clnk&gl=se
12:42:01 <sech1> #define MSR_DC_CFG                  0xC0011022
12:42:10 <sech1> so ..22 register is data cache config
12:42:19 <sech1> #define MSR_BU_CFG3                 0xC001102B
12:42:27 <sech1> no idea how to decipher this abbreviation
12:43:59 <sech1> but this is for older CPUs (Bulldozer etc.), probably Ryzen just has the same registers
12:45:45 <sech1> Found it all here: https://www.amd.com/system/files/TechDocs/42300_15h_Mod_10h-1Fh_BKDG.pdf
12:45:49 <sech1> MSRC001_1022 Data Cache Configuration (DC_CFG)
12:45:56 <sech1> MSRC001_102B Combined Unit Configuration 3 (CU_CFG3)
12:46:18 <sech1> 1022, bit 13, set to11=Disable the DC hardware prefetcher.
12:46:59 <sech1> 1022, bit 4 is also interesting: 1=Disable speculative TLB reloads
12:47:47 <sech1> ohhh my gawwd, bits 10, 22:19 and 23 are even more interesting
12:47:52 <sech1> so much to test this evening
12:48:58 <sech1> 102B register - no idea, the changed bits (28:29) are not documented there
12:50:40 <sech1> hmm, bit numbers that are changed are different from what's written here, so I guess Ryzen changed them all
12:50:55 <sech1> But 0xC0011023 register is also interesting
13:06:51 <cohcho> 0xc001102b: 0x2008cc17 -> 0x1008cc16 or 0x2008cc17 -> 0x1808cc16 ?
13:11:10 <cohcho> 0x1808cc16 at your photo and 0x1008cc16 here in irc are different values
13:16:34 <sech1> 0x1808cc16
13:16:49 <cohcho> and default ?
13:17:00 <sech1> 2008CC17
13:17:17 <cohcho> more bits has changed then
13:17:38 <sech1> bit 28 was set to 0, bits 26:27 were set to 1
13:17:44 <sech1> and bit 0 was set to 0
13:18:02 <sech1> no, bit 29 was set to 0, I'm bad at counting
13:18:14 <cohcho> type binary of both values and check carefully
13:18:19 <sech1> and bits 27:28 were set to 1
13:22:41 <cohcho> You can bruteforce only changed bits of 0xc001102b to remove any redundant bits. It's likely only 2 of them shoud be changed.
13:25:29 <sech1> Pasting the link here: https://community.hwbot.org/topic/190114-rog-crosshair-viii-hero-formula-impact-ln2-oc-guide/
13:25:38 <sech1> Zen Perfboost : https://bit.ly/2kBs15c (After loading CineBench R15 or Geekbench3 bench then run the boost, just click defaults button after benching to prevent bsod)
13:25:45 <sech1> it also messes with MSR registers
15:23:33 <kico> anyone knows HR for a 2950x ?
16:01:23 <gingeropolous> im mining with gpus sech1 . cause im an idiot
16:01:57 <sech1> not really, they're nice space heaters in this weather
17:10:53 <sech1> https://www.reddit.com/r/MoneroMining/comments/e962fu/9526_hs_on_ryzen_7_3700x_xmrig_520_1gb_pages_msr/fahdy11/?context=3
17:11:08 <sech1> 18% hashrate increase
17:11:19 <gingeropolous> with the 1 gb pages?
17:11:25 <gingeropolous> nice
17:11:58 <sech1> no, it's the magic MSR mod
17:12:03 <sech1> turning off HW prefetchers
17:12:15 <sech1> and 1GB pages, yes
17:17:01 <tevador> yeah, memory starved systems will see a bigger jump in hashrate
17:17:07 <kico> is it equivalent to disabling in BIOS ?
17:17:10 <sech1> yes
17:17:17 <kico> thanks
17:17:19 <sech1> but my BIOS doesn't have this option
17:17:48 <kico> oh ok I see :)
17:31:32 <sech1> hmm, I'm getting 9614 h/s now (up from 9526 h/s)
17:31:42 <sech1> just changed "Performance Bias" option on BIOS
17:31:45 <sech1> interesting
17:32:02 <sech1> I guess it's some other MSR register
17:33:32 <sech1> how to dump all MSR registers?
17:36:07 <cohcho> apply rdmsr to all values within [0xc0011000, 0xc00110ff] or even wider range, it should fail for unsupported addresses
17:36:39 <cohcho> Since there are no docs about available registers
17:36:46 <sech1> yep
17:36:58 <sech1> I'll try all performance bias options and then dump the best one
17:37:19 <sech1> "aida/geekbench" performance bias got me down to 7000 h/s :D
17:51:21 <sech1> so 9614 h/s was the best, I'll dump registers and start comparing
17:54:15 <cohcho> Do you have direct link to the firmware you're using?
17:57:30 <sech1> cohcho https://dlcdnets.asus.com/pub/ASUS/mb/SocketAM4/PRIME_B450M-K/PRIME-B450M-K-ASUS-2006.zip
17:58:30 <sech1> "Performance bias = CBR15 gentle" was the best option
17:59:49 <sech1> I think I've found it
18:00:02 <sech1> another MSR register, I just need to retest it to confirm
18:01:49 <sech1> sudo wrmsr -a 0xC0011021 0x40
18:01:57 <sech1> 9601 h/s
18:02:09 <sech1> so I think some another register adds 13 h/s
18:03:12 <sech1> sudo wrmsr -a 0xC0011020 0
18:03:14 <sech1> this is it
18:03:16 <sech1> 9614 h/s
18:12:55 <sech1> MSRC001_1020 Load-Store Configuration (LS_CFG)
18:13:04 <sech1> bit 28: 1=Disable streaming store functionality.
18:13:14 <sech1> MSRC001_1021 Instruction Cache Configuration
18:14:15 <tevador> looking for that last +0.1%
18:14:19 <tevador> :D
18:15:35 <sech1> hey, it's +0.9%
18:17:08 <cohcho> It isn't clear yet whether this is the last +0.1%
18:17:27 <sech1> this can give more ideas: https://www.amd.com/system/files/TechDocs/42300_15h_Mod_10h-1Fh_BKDG.pdf
18:17:52 <sech1> It's not for Ryzens, but register names there are what to look for
18:18:01 <sech1> MSRC001_1xxx group
18:20:46 <sech1> MSRC001_102D Load-Store Configuration 2
18:21:00 <sech1> ForceSmcCheckFlwStDis 0=Force a
18:21:00 <sech1> self modifying code check when a cache probe hits a store that has not retired
18:21:03 <sech1> interesting
18:21:16 <sech1> but probably won't give anything
18:21:37 <cohcho> I've become a bit scary when saw all those configurable things inside cpu 1 month. Are you really sure that you current cpu configuration is within 10% from the best?
18:21:56 <cohcho> 1 month ago*
18:22:25 <sech1> I'm sure that 9850 h/s is the limit for my 3700X @ 4.1 GHz
18:22:38 <sech1> because this is where it's bottlenecked by instrution execution, not memory
18:23:07 <sech1> I tested it at 2.05 GHz and got 4925 h/s, so this is how I calculated it
18:23:30 <sech1> so it's within 2.5% from the best
18:25:01 <cohcho> This approach test&dump is much faster than reading firmware asm.
18:25:34 <sech1> damn, so much progress since even 1 month ago
18:25:41 <sech1> 1 month ago 8700 h/s was the limit for me
18:30:14 <sech1> I'm getting consistent 600+ h/s per thread, beautiful :)
18:30:24 <sech1> 1200+ h/s per physical core on all cores
18:38:56 <tevador> getting closer to the ideal CPU ASIC performance
18:40:02 <sech1> yeap
18:40:05 <sech1> this is the idea, right?
18:40:58 <sech1> I guess I can set it to 3.6-3.8 GHz, undervolt and it'll not be bottlenecked by memory anymore
19:00:48 <hyc> nice
19:05:00 <sech1> I wonder what 3950X could do with all this tuning...
19:06:19 <gingeropolous> 9 bajillion h/s
19:07:02 <sech1> 18-19 kh/s is possible
19:31:22 <tevador> yes, that's the goal, to squeeze the last bit of performance out of the CPU hardware
19:31:33 <tevador> to make it harder for ASICs
19:33:07 <sech1> 9614 h/s, 156 W at the wall (but it's 51W when idle)
19:33:22 <sech1> not very efficient, but I'll try to find more efficient speed/voltage tomorrow
20:01:52 <tevador> energy-wise, the bitcoin network is roughly 360x more secure than Monero
20:01:58 <tevador> based on the most efficient hardware
22:29:02 <tevador> another trash article about RandomX: https://www.publish0x.com/miner-insights/cpu-mining-dominates-monero-randomx-upgrade-xmrgdz
22:29:20 <tevador> ... botnets ... botnet mining ... more botnets ... battle against botnets ...