Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Deep Blue has the same evaluation as Deep Fritz 10
- - By rocket (***) [se] Date 2017-12-10 14:39 Edited 2017-12-10 14:59
Wanna know how it felt to play Deep Blue on a personal computer? Well, nostalgia buffs, you can.

From a study in 2006 with the program version that the same year beat Kramnik:

In positions such as the 17th move from game 1 of the match, Deep Fritz (with the black pieces) plays the same move as Deep Blue and anticipates the same continuation. The evaluation is to a depth of 16 plies (8 moves) topping out at 4.737 million positions per second with all four CPU cores at 100% utilization after roughly 3 minutes of analysis.

Editors note: it was discovered that the multiprocessor implementation was not working, resulting in Deep Fritz only utilizing 1 core when set to 4 cores.

I can confirm however with the bugg fix 10.1 that it indeed plays the same continuation after several minutes, and this game was no exception. Deep Fritz 10.1 has the highest overlap/ponderhit rate with Deep Blue out of all its generation of PC-programs. It naturally discards some of its moves but the average move overlap is substantial.
Parent - - By Labyrinth (*****) [us] Date 2017-12-11 00:39 Upvotes 2
I think it's weird that you talk about Fritz 10 so much.
Parent - - By rocket (***) [se] Date 2017-12-12 01:55
It's the only engine that was actually programmed to play anti-human chess at a super gm level:
Parent - - By turbojuice1122 (Gold) [us] Date 2017-12-12 02:23
I think you would find that AlphaZero would be the best ever by far in this department, even though this wasn't its specific purpose.
Parent - - By rocket (***) [se] Date 2017-12-12 02:44
No evidence of that from the 10 games posted. It used 2000 cores (equivalent) against 64. It's a marketing scam.
Parent - - By turbojuice1122 (Gold) [us] Date 2017-12-14 03:54
I think it's safe to assume that the setup has gained at least 100 elo since these results were published.  The moves made in the games are eerily super-GM-human-like, and in many cases beyond that.
Parent - - By mist (**) [gb] Date 2017-12-14 13:03

>programmed to play anti-human chess at a super gm level:

So basically what is the point of that?

As for this AlphaZero machine/brain the question really is, like Deep Blue will you or I, or anyone be going out and buying a copy in a year or so?!

I do not think so.
Parent - By zaarcis (***) [lv] Date 2017-12-14 14:04
Chess community could make its own alphazero in less than year, if really tried.
If it happens, the resulting engine should runnable on everything with GPU (and with some software libraries even on bare CPU, only much slower).

Of course, the more and better GPUs one has, the stronger will it become. :) But the same thing happens with standard engines. :D
And it's reasonable prediction that good GPUs (or maybe even TPUs) for machine learning will become more and more mainstream, easier to buy and more efficient.
Parent - - By zaarcis (***) [lv] Date 2017-12-14 14:16
Bullshit. It used 4 TPUs for the match, that can be replaced with 2 Titan V for 6000$.
I don't know exactly how much does 64 core computer cost, but maybe 4000$ or smth (I randomly surfed in the net, and can't be completely sure).

Seems almost equal, if judging by costs. And it's not like you can run stockfish on video cards. :D

Marketing scam, lol. :D All your chess games are marketing scam, all your board games are marketing scam. :D
Deepmind 100% proved that such approach to chess is working, that's all that matters for them. Btw, it's more like Stockfish gained tons of publicity from this "scam" while Deepmind already had more than enough.
Parent - - By mist (**) [gb] Date 2017-12-14 16:42 Upvotes 1

>can be replaced with 2 Titan V for 6000$.

Is that all......bloody hell I will put my order in for a couple now! :eek:
Parent - By zaarcis (***) [lv] Date 2017-12-14 17:10
We still have to train our own AlphaZero (similarly to ongoing LeelaZero project, for example). Until it happens, even I can slowly save all necessary money.
Parent - By Carl Bicknell (*****) [gb] Date 2017-12-15 10:29

> It used 4 TPUs for the match, that can be replaced with 2 Titan V

Are you sure?
Parent - - By user923005 (****) [us] Date 2017-12-26 20:19
A GPU is not a TPU.
Parent - By Banned for Life (Gold) Date 2017-12-26 22:16 Upvotes 1
NVIDIA is spending a lot of money competing in the inference engine market, and are currently the leader in this market measured by either volume or the value of this business area. See for a discussion of the performance of the Tesla V100 on a variety of recently developed network benchmarks.

Note that it is the inference engine rather than the machine learning component that 99.99% of the chess engine using population are interested in...

Here is NVIDIA's discussion of their competition with the TPU2:

The Rise of Accelerated Computing
Google* has announced its Cloud Tensor Processing Unit (TPU), and its
applicability to deep learning training and inference. And while Google
and NVIDIA chose different development paths, there are several themes
common to both our approaches. Specifically, AI requires accelerated
computing. Accelerators provide the significant data processing
necessary to keep up with the growing demands of deep learning in an
era when Moore’s law is slowing. Tensor processing is at the core of
delivering performance for deep learning training and inference. Tensor
processing is a major new workload that enterprises must consider
when building modern data centers. Accelerating tensor processing can
dramatically reduce the cost of building modern data centers.
According to Google, the Cloud TPU (also referred to as "TPU2") will
be available later this year, and that a single Cloud TPU can deliver 45
teraflops of computing horsepower. NVIDIA’s Tesla V100 can deliver 125
teraflops of deep learning performance for both training and inference.
An 8-GPU configuration such as DGX-1 can now deliver a petaflop of
deep learning computing power.
NVIDIA’s approach democratizes AI computing for every company, every
industry, every computing platform and accelerates every development
framework – from the cloud, to the enterprise, to cars, and to the edge.
Google and NVIDIA are the clear leaders – we collaborate closely while
taking different approaches to enable the world with AI.
Up Topic The Rybka Lounge / Computer Chess / Deep Blue has the same evaluation as Deep Fritz 10

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill