In this particular case GV100 (we don't know which) is roughly around 8x compared to my OC'ed 1060 and 6 threads on i7 firstname.lastname@example.org GHz is around 8x in nps compared to my single core. My TC (10'+6'') is around 4x longer than 2'+2'' so scaling should be roughly the same. So this 40 Elo discrepancy between my result and this one is totally within error margins once SMP loss when going from 1 to 6 cores is included. I would say it is even spot on.Uri Blass wrote: ↑Tue Aug 14, 2018 5:02 amWhat about the following result with the same net that play on TCEC that I read in another thread:Milos wrote: ↑Mon Aug 13, 2018 10:19 pmFanboys hype.CMCanavessi wrote: ↑Mon Aug 13, 2018 6:46 pmI really don't care if it wins TCEC or if it finishes last in div4, but the time management issue is real and the leela team really screwed up. Check my stream, I'm currently doing lc0 (on a gtx 1080 non ti) vs SF8 (on 8 cores) and currently they are 6-6. Why do you think leela went from dominating in div4 to playing awful in div3? It's no magic, but also, as you say, excuses are useless. Leela team will learn the lesson for next season.
Come back when you run 200 or 500 games. SF 8 used to loose of engines 300Elo weaker in TCEC. So what?
On my setup Lc0 recent test net on OCed 1060 vs single very slow core SFdev 44% at 10'+6'' TC after 200 games.
After appropriate scaling one could see that SF8 on fast 8 cores should be around 80Elo stronger than the strongest Lc0 test net on 1080.
Lets see, you have equal score and error margins of 120Elo, what a surprise.
Now you can talk about trend...
Leela Lc0 with 10520 net running on 1 GV100, versus Stockfish dev 18080112 running on 6 threads i7 5820k @4.2 GHz
TC was 2+2 and Arena perfect2010 opening book was used and the result was:
Leela 10520 - SF dev +8 -8 =69
0 ELo difference with confidence interval [-32 , +32] (95% c.l).
I know that lc0 has some hardware advantage relative to TCEC but can other engines in division 3 achieve this type of result with the same hardware advantage?
Problem with Carlos' results is that they are on way too small sample and totally off the expected value and he is just hyping on them.
And problem with comparison of GV100 and some random CPU is that current SFdev is still +100Elo on for example Ryzen 7 1700 compared to strongest Lc0 on GV100 and GV100 is more than 10x in price and at least 5x in TDP compared to Ryzen 7 1700 so it is totally unfair competition.