13:38:57 <\x> https://www.phoronix.com/review/amd-ryzen-7900x-7950x-linux/21
13:39:01 <\x> hyc sech1
13:42:38 <nioc> that shows the 5900x at 18% faster than the 3900x   
13:42:43 <nioc> I thought that those 2 were virtually equivalent for mining 
13:44:15 <nioc> also shows the 3900x 31% faster than the 3950x  lolol
13:48:44 <sech1> they have no idea what they're doing
13:49:02 <sech1> XMRig will need an update for Zen4 MSR mod
13:49:03 <hyc> yeah, phoronix benchmarks are notoriously bad
13:49:56 <hyc> hm, they show 5800x3d as faster than 5800x
13:51:14 <sech1> it's probably just "this system had huge pages, and that system did not"
13:51:48 <hyc> lol
13:57:59 <\x> https://xmrig.com/benchmark/2cgHoB
13:58:00 <\x> hyc
13:58:08 <\x> some one did a tuned run prolly
13:58:20 <\x> its the fastest 6c/12t run there
13:58:23 <\x> https://xmrig.com/benchmark/12
13:58:31 <\x> everything higher is non-6core chips
14:00:06 <sech1> hmm, maybe they didn't change the relevant MSR registers, for once :D
14:21:11 <hyc> always possible...
14:21:22 <hyc> funny looking at the Multi Threads page
14:21:28 <hyc> all AMD Epyc
14:21:46 <hyc> oh, top 25 anyway ;)
14:22:23 <sech1> actually, reading up on Zen4, it's just an evolution of Zen3 architecture - more caches and buffers everywhere
14:22:32 <sech1> so maybe MSR registers for caching didn't change
14:22:49 <sech1> Zen5 will be the big overhaul
14:22:58 <hyc> yeah, can't see them changing for no reason
14:23:49 <\x> https://xmrig.com/benchmark/5mCeSC
14:23:51 <\x> kekts
14:23:54 <\x> this exists
14:23:56 <\x> 7900X ran
14:23:59 <\x> 17.5khs
14:24:20 <\x> 584 seconds for 10 million hashes
14:24:24 <\x> ASIC cheat ser
14:24:43 <sech1> where are they getting them, release is only tomorrow?
14:25:05 <\x> anyway sech1 i can ask some xOC results for sometime, but thus is a new platform so the xOC guys i know will be doing shit that gets points first
14:25:13 <\x> tomorrow sech1
14:25:51 <sech1> good to see some progress over 3 year old Zen2 in RandomX, finally :D
14:28:20 <\x> and they ran a 10M test
14:28:23 <\x> not a 1M
14:28:31 <sech1> so kind of a stable oveclock
14:28:32 <\x> so definitely not a record chaser guy
14:28:38 <\x> its XMP
14:28:43 <\x> the dimms are rated for 6000
14:28:44 <sech1> maybe they used it as a stability test
14:28:49 <\x> maybe yeah
14:30:04 <sech1> 7900X seems to be the sweet spot in the line up. High base (4.7) and top (5.6) clocks, only 100 MHz lower than 7950X, but 7950X has lower base clock.
14:30:22 <sech1> So probably 7950X will be only a tad faster than 7900X in most workloads
14:31:36 <\x> sech1: always since zen happened, the 6 and 12 cores are the best balance
14:31:50 <\x> you get the full L3 size on them anyway
14:32:12 <\x> unlike intel, less cores, less L3. on amd, less cores, same L3
14:32:36 <sech1> I wonder if they'll do 7950X3D and 7800X3D next year...
14:33:34 <\x> i wonder if someone has benchmarks of the 6 core, I kinda want to downsize a bit
14:34:00 <\x> if tuned 7600X can match or be in 10% of 3900X in compiles, ill grab it
14:34:06 <\x> if not, then ill wait for a while
16:24:21 <hyc> so how does Intel Raptor Lake support 24 cores and 32 threads?
16:24:34 <hyc> seems like quite a lopsided design
16:25:54 <hyc> ohhh. they're using heterogeneous cores
16:38:59 <gonbatfire[m]> So, no sick RandomX gainz with Ryzen 7000 then?
16:39:13 <gonbatfire[m]> Man I really wanted to burn some money on a new rig!
16:41:19 <hyc> decent gains, for sure. 
16:41:58 <sech1> 5-10% gains
16:42:21 <selsta> 5-10% while using the same amount of power?
16:42:32 * gonbatfire[m] uploaded an image: (68KiB) < https://libera.ems.host/_matrix/media/r0/download/monero.social/GIZxvtYCBJkNVcjgQIWinpOG/Screenshot_20220926-134145.png >
16:42:44 <gonbatfire[m]> selsta: Oh okay
16:42:55 <selsta> I don't know, I was asking :P
16:43:04 <gonbatfire[m]> That's nice, guess we have to wait for the more tuned results
16:44:04 <\x> selsta: nah, it pulls more.
16:44:09 <\x> but you do get more.
16:44:39 <\x> amd dialed in the silicon for real this time
16:44:52 <\x> shit works with a temp limit, it unleashes everything sles
16:44:55 <\x> else*
16:45:19 <\x> so you give it more cooling it gives you more power, more frequency, more voltage, more everything
16:45:51 <nioc> more efficiency? 
16:45:55 <\x> not really
16:46:20 <\x> that curve tapers off as with all silicon
16:46:26 <\x> they work like capacitors.
16:46:41 <\x> sech1: overclocking
16:46:42 <\x> https://www.youtube.com/watch?v=20r4NhtvOj8
16:46:46 <nioc> more everything, YMMV
16:47:01 <\x> https://i.imgur.com/ER8qqfM.png
16:47:18 <\x> so for bclk oc, you can skew the cpu core clock now safely
16:47:23 <\x> clock1 = cpu core
16:47:29 <\x> clock2 = pcie, memory, fclk
16:47:46 <\x> so its safe to skew cpu core speeds now with bclk
16:48:04 <\x> but if you want to skew either memory or fclk, you have to kind of give up pcie
16:48:17 <\x> so you either lower it or you dont daily an nvme drive
16:49:02 <\x> ofcourse an external clockgen will only be available on higher end boards
16:49:14 <\x> if you want the feature, you know which boards to pick.
16:51:01 <\x> one of the good things too with this design is that usb clock is a separate domain so no need to really use ps/2 for some weird overclocking
16:51:24 <\x> but yeah be wary of memory and fclk overclocking with bclk as that will skew pcie.