smatovic wrote: ↑Wed Sep 02, 2020 7:15 pm
Hmm, something is fishy with those RTX 30xx FP32 numbers, better wait for the first LC0 benchmarks imho, cos Turing added dedicated INT32 cores to the SM, these are still present in the whitepaper for the A100, but now with RTX 30xx Nvidia seems to count these INT32 cores as FP32 Cuda Cores, therefore the doubling of cores and FP32 throughput, but some tech sites claim it is yet unknown if the doubling of FP32 throughput per SM holds in real world applications....
--
Srdja
Hmm, don't want to play the kill-joy in here, but here a official Nvidiia benchmark with RTX 20xx vs 30xx with and wo Ray-Tracing...the first three benchs to left are wo RT...
smatovic wrote: ↑Wed Sep 02, 2020 7:15 pm
Hmm, something is fishy with those RTX 30xx FP32 numbers, better wait for the first LC0 benchmarks imho, cos Turing added dedicated INT32 cores to the SM, these are still present in the whitepaper for the A100, but now with RTX 30xx Nvidia seems to count these INT32 cores as FP32 Cuda Cores, therefore the doubling of cores and FP32 throughput, but some tech sites claim it is yet unknown if the doubling of FP32 throughput per SM holds in real world applications....
--
Srdja
Hmm, don't want to play the kill-joy in here, but here a official Nvidiia benchmark with RTX 20xx vs 30xx with and wo Ray-Tracing...the first three benchs to left are wo RT...
--
Srdja
Followup...
<speculation mode on>
If we assume that LC0 won't proft from a) the 2xFP32 per SM, b) the 2xTensorOps
and c) the 2xRTX, then the RTX 3080 would be on par with the RTX 2080 TI, but
for a lower price.
Dann Corbit wrote: ↑Wed Sep 02, 2020 8:50 pmGive in to the dark side, and suck power like a titanic inductor, screaming through a magnetar's magnetic field. Sure, you will eat one potato a day, but you will have beautiful chess. Which is more important?
If it's not too inconvenient, it would be interesting for you to watch two computers, one playing against itself at 3000 elo and the other playing against itself at 4000 elo. It must be randomised so that you yourself don't know which is which. Then see whether you can tell which is which.
Last year, I got annoyed with one of my friends who drinks coca cola, which is well over 5x more expensive than supermarket cola*, so I made him do a blind tasting: it failed - he called out the supermarket cola on the very first sip! But maybe "cheap chess" can do better than than the supermarket cola...
*IIRC coca cola was 80p for a 20cl can, but supermarket cola was 50p for a 1 litre bottle
Got it! When you say "beautiful chess", you don't mean "aesthetically pleasing", you mean "beat my online/correspondence opponents"!
Writing is the antidote to confusion.
It's not "how smart you are", it's "how are you smart".
Your brain doesn't work the way you want, so train it!
Winning chess is beautiful chess.
But I mean more than that.
The strongest programs will make moves that take my breath away, just like the strongest players.
Often, I won't see how amazing the move is while watching a game, it is during post game analysis that I marvel.
I'm 63, so I don't think as fast as I used to. But I can still get there. It just takes a little longer than it used to.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Werewolf wrote: ↑Sun Sep 27, 2020 8:24 pm
Seems like there is actual data appearing now
Very early results by people quoting ankan suggest if we take the 2080 Ti as a reference point:
3080 is about 1.4x faster
3090 is about 1.6x faster
A100 is about 2.7x faster
Seems like the batch size makes a big difference
That last result contradicts what I saw earlier. Anyway, if this forum had a means to insert an image I'd post the screenshot from the discord...
If the speed refer to Leela speed, it it seems being acceptable.
Only the prices are questionable.
Werewolf wrote: ↑Sun Sep 27, 2020 8:24 pm
Seems like there is actual data appearing now
Very early results by people quoting ankan suggest if we take the 2080 Ti as a reference point:
3080 is about 1.4x faster
3090 is about 1.6x faster
A100 is about 2.7x faster
Seems like the batch size makes a big difference
That last result contradicts what I saw earlier. Anyway, if this forum had a means to insert an image I'd post the screenshot from the discord...
If the speed refer to Leela speed, it it seems being acceptable.
Only the prices are questionable.
Yes, it is acceptable when running Lc0, and much more when running games. But remember most are just buying a upgrade, and with Lc0. You can use 2 GPU's of different generations. So for the people who have room and power for 2 GPU's. And a lot of system can do this. Run the 3000 series for your games, and run 2 GPU's when running Lc0! It is the best of both worlds.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Stupid to consume 300 watts of energy per hour just to play chess with LC0. With my android SD865 and stockfish 12, I have the best chess engine, I can play 6 hours, and consume just 25 watts of energy (1 charge) . Stockfish can run on 100% of the PCs of the world, LC0 with same strength, needs at least 2000 Cuba cores (2080 GPU) minimum which is installed on 2% of all PCs in the world...
RogerC wrote: ↑Mon Sep 28, 2020 6:19 am
Stupid to consume 300 watts of energy per hour just to play chess with LC0. With my android sd865 and stockfish 12, I have the best chess engine, I can play 6 hours, and consume just 25 watts of energy (1 charge) . Stockfish can run on 100% of the PCs of the world, LC0 with same strength, needs at least 2000 Cuba cores (2080 GPU) minimum which is installed on 2% of all PCs in the world...
This is a joke. 300 watts an hour is nothing for running a chess engine. I pay more for turning on my yard lights. If you can not pay this cost. Then why buy a desk top computer. Or any computer! Unless you live in NY or California and French Polynesia. If this is the case then Stockfish is the best bang for the buck. And just use it on a smart phone.
And on the other side. If you can afford a high end CPU, Who is not going to have a high end GPU, unless your an idiot. One goes hand in hand with the other.
And on the third side, my CPU uses as much energy as my 2080 TI. And the new gen GPU's use less energy, and are faster.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Werewolf wrote: ↑Sun Sep 27, 2020 8:24 pm
Seems like there is actual data appearing now
Very early results by people quoting ankan suggest if we take the 2080 Ti as a reference point:
3080 is about 1.4x faster
3090 is about 1.6x faster
A100 is about 2.7x faster
Seems like the batch size makes a big difference
That last result contradicts what I saw earlier. Anyway, if this forum had a means to insert an image I'd post the screenshot from the discord...
That's good news. It means that 3080 is about 190% of 2080 as Lc0 speed goes, so represents a very good deal. Do you know the 3070 estimate? To me 3080 is a bit too hungry Watt wise for my PC.
But maybe "cheap chess" can do better than than the supermarket cola...
If this is the case then Stockfish is the best bang for the buck. And just use it on a smart phone.