Since SF8 4CPU is 3386 CCRL 40/40 thus 64 core setup with ~ 2x faster cores than CCRL reference should be at least 3525+duncan wrote:do you have an estimate for elo of alphazero ?
So A0 + 4xTPU should be 100 elo above, thus 3625+ on CCRL 40/40?
Moderators: hgm, Rebel, chrisw
Since SF8 4CPU is 3386 CCRL 40/40 thus 64 core setup with ~ 2x faster cores than CCRL reference should be at least 3525+duncan wrote:do you have an estimate for elo of alphazero ?
Nope. I'm just pointing out that trending out like that is inherently risky, as I'm sure Kai would agree. It does give an interesting data point, but it isn't necessarily meaningful. I prefer to rely on actual match data as it becomes available, than try to extrapolate the impact of a 1000x increase in speed.duncan wrote:do you have an estimate for elo of alphazero ?Robert Pope wrote:Well, first off, that is a pretty big "if". Whenever you extrapolate upwards like this, there is a real risk that your assumptions won't continue to hold.duncan wrote:so if alphazero elo is 3200, it is only 100 elo stronger than lc0 which means lc0 will soon stall ?Laskos wrote:
Therefore, just a factor of 4 in time control (or hardware) (2 doublings) gives a boost of 128 Elo points compared to standard A/B engine. Or 64 Elo points per doubling. One can extrapolate:
On one CPU core at 4 s/move, from this match LC0 ID160 is about 2100 CCRL Elo. A top GPU, say Nvidia 1080 Ti, is faster by a factor of 25 compared to 1 CPU core. Tournament TC is about 40 longer than 4s/move. So, all in all, a total of a factor of 1000 time-hardware wise, or 10 doublings. So, ID160 on a top GPU and LTC would be about 2750 CCRL Elo. And on DeepMind hardware used in exhibition match, would be 3100+ CCRL Elo.
https://imgur.com/a/c04yc
Gee, you got it almost all wrong. Mainly because figure 2 is totally bogus.mirek wrote:Exactly, and what's even more remarkable is that according to the A0 paper (figure 2) 4xTPUs will do about 80k payouts in 1s and at 80k playots A0 is only 100 - 150 elo weaker than at 1 min / move (5000k playots)
Also 1x 1080Ti (11 TFLOPS) vs 4xTPU (180 TFLOPS) means nps gets reduced to 4.8k nps Even if we assumed that the TPU is somehow more effective flops to flops by factor of 4x the resulting 1080Ti playouts would be still close to 80k per minute. Thus to me it seems quite convincing that A0 on 1080Ti would be with good confidence max 150 elo weaker at 1min / move compared to 4xTPU configuration. (and most likely not more than 100 elo weaker)
If it can be easily demonstrated then please demonstrated the measured numbers.Milos wrote: Gee, you got it almost all wrong. Mainly because figure 2 is totally bogus.
First scaling of SF is bogus.
It can be easily demonstrated that SF8 on 64 cores when going from TC = 1s/move to 1min/move gains at least 40*6 = 240Elo.
They show in figure 60Elo?!???!
I like your confidence how first you disregard A0 scaling measured by DeepMind and then provide your own (supposedly better) numbers which as far as I can see you have just pulled out of thin air.Milos wrote:Second, A0 might scale better or worse than SF, but will get at least 200Elo, more probably over 300Elo when going from 1s/move to 1min/move.
OK, here I really got confused, but anyways, if I were to take your number (1080Ti = 40x slower) it would still fit comfortably above the 80k playouts per minute thus according to the graph only within 100 - 150 elo weaker.Milos wrote: However, for actual matches Google used first gen TPUs (because 4TPUs for matches give exactly 4x nps that 1TPU used for self-play gives - 800sims in 40ns) that are actually around 92TOPS. So one 1080Ti is approximately 40x slower than 4TPUs used for playing matches with SF.
Kai, I guess you are being extremely polite and didn't need any testing to know that.Albert Silver wrote:I suspect they could not care less about releasing them as general consumer products, and nor was that ever even on the table.Laskos wrote:I am even beginning to suspect that they didn't release to the general consumer their products because on a normal i7 CPU and average GPU, the performance of their AlphaZero Go and Chess programs would be not that impressive, or even pretty lame (compared to the hype), especially in Chess.