AlphaZero is not like other chess programs

[Dann Corbit]

Comparing AlphaZero to Stockfish is like comparing an airplane to a motorboat.

Both can be used to cross a sea, but they are using completely different techniques to get there.

SF won't run on tensor chips.
AlphaZero won't run on ordinary CPUs.

The approach is very different.

The horsepower of AlphaZero is enormous. But that does not diminish the outcome. If we gave SF a trillion cores, the last 99.999999999 % don't help anyway,

[syzygy]

Well said.

[Milos]

Comparing AlphaZero to Stockfish is like comparing an airplane to a motorboat.

Both can be used to cross a sea, but they are using completely different techniques to get there.

SF won't run on tensor chips.
AlphaZero won't run on ordinary CPUs.

The approach is very different.

The horsepower of AlphaZero is enormous. But that does not diminish the outcome. If we gave SF a trillion cores, the last 99.999999999 % don't help anyway,

AlphaZero could actually run on ordinary CPUs. The only thing it needs is TensorFlow lib which more than successfully runs on general CPUs. The other question is how slow it would be.

If SF had access to trillion cores, you could simply just play rollouts from the root position and reach almost perfect play in every position.

[corres]

[quote="Dann Corbit"]

Comparing AlphaZero to Stockfish is like comparing an airplane to a motorboat.
Both can be used to cross a sea, but they are using completely different techniques to get there.
SF won't run on tensor chips.
AlphaZero won't run on ordinary CPUs.
The approach is very different.
The horsepower of AlphaZero is enormous. But that does not diminish the outcome. If we gave SF a trillion cores, the last 99.999999999 % don't help anyway,

[/quote]

I agree.
An AlphaZero-type system is appropriate to optimize the parameters of normal chess engines, too. I think with the help of such a system the team of Stockfish could enhance the Elo of Stockfish so high than AlphaZero have. Maybe it needs also some modifications of the code of Stockfish.

[Albert Silver]

Comparing AlphaZero to Stockfish is like comparing an airplane to a motorboat.

Both can be used to cross a sea, but they are using completely different techniques to get there.

SF won't run on tensor chips.
AlphaZero won't run on ordinary CPUs.

The approach is very different.

The horsepower of AlphaZero is enormous. But that does not diminish the outcome. If we gave SF a trillion cores, the last 99.999999999 % don't help anyway,

AlphaZero could actually run on ordinary CPUs. The only thing it needs is TensorFlow lib which more than successfully runs on general CPUs. The other question is how slow it would be.

If SF had access to trillion cores, you could simply just play rollouts from the root position and reach almost perfect play in every position.

Actually, I am curious to know if one could run it well using GPUs, such as a couple of nVidia's high-end Titans. Gian-Carlo's Leela benefits greatly from a good GPU.

https://www.sjeng.org/leela.html

[pilgrimdan]

Comparing AlphaZero to Stockfish is like comparing an airplane to a motorboat.

Both can be used to cross a sea, but they are using completely different techniques to get there.

SF won't run on tensor chips.
AlphaZero won't run on ordinary CPUs.

The approach is very different.

The horsepower of AlphaZero is enormous. But that does not diminish the outcome. If we gave SF a trillion cores, the last 99.999999999 % don't help anyway,

AlphaZero could actually run on ordinary CPUs. The only thing it needs is TensorFlow lib which more than successfully runs on general CPUs. The other question is how slow it would be.

If SF had access to trillion cores, you could simply just play rollouts from the root position and reach almost perfect play in every position.

Actually, I am curious to know if one could run it well using GPUs, such as a couple of nVidia's high-end Titans. Gian-Carlo's Leela benefits greatly from a good GPU.

https://www.sjeng.org/leela.html

"Alpha Go runs on about $25 million in hardware--it's not exactly a lightweight system... "

https://www.inc.com/lisa-calhoun/google ... learn.html

I realize the quote says alphago ... but I would imagine the same would be true for alphazero also ...

[jhellis3]

Nah, the hardware Alpha Zero runs on is much less expensive, but by no means cheap. It ran/runs on 4 2nd gen google TPUs.

For those interested, you can now buy Titan V cards (the first consumer cards with Tensor units) now for $3,000 a pop here:
https://www.nvidia.com/en-us/titan/titan-v/

Currently limited to 2 per customer but assuming you could buy 4, that would be 12k + rest of system, and get you about 2/3rds+ of the performance of the TPUs depending on the workload bottlenecks.

Or you could spend $70,000 and get a pre-built machine from Nvidia which is about 75%+ the performance.

Of course, as node shrinks happen, die sizes shrink, and the NN improves, perf/dollar and perf/watt will continue to increase.

[shrapnel]

Actually, I am curious to know if one could run it well using GPUs, such as a couple of nVidia's high-end Titans. Gian-Carlo's Leela benefits greatly from a good GPU.

https://www.sjeng.org/leela.html

That would indeed be great.
I have always found it ridiculous that Chess programs depend so much on CPUs when modern GPUs are so much more powerful.
As you know, similarly, Video Gaming really came into its own when the Developers stopped relying so much on CPU and started using the power of the GPU !

[Cardoso]

This "zero" approach really surprised me, starting from almost nothing except the game's rules. I wasn't expecting this advancement so soon, even though they used pretty powerful hardware.

One of the things that amazes me is that Alpha Zero isn't blind to deep tactics (taking into consideration it only runs at 80K positions per second).
Can someone try to explaine this?
I mean I can understand it plays well positionally, but also being tactical so strong is surprising to me.

[shrapnel]

This "zero" approach really surprised me, starting from almost nothing except the game's rules. I wasn't expecting this advancement so soon, even though they used pretty powerful hardware.

One of the things that amazes me is that Alpha Zero isn't blind to deep tactics (taking into consideration it only runs at 80K positions per second).
Can someone try to explaine this?
I mean I can understand it plays well positionally, but also being tactical so strong is surprising to me.

Quite simple.
Scientific explanation others can give you. I can just tell you what it means.
Positional Play, Tactical Play.... these are just Terms used by humans and obsolete chess engines like Stockfish, Houdini and Komodo.
The AlphaZero Program, after learning Chess in just a few Hours, has one and only one Goal, to Mate the Enemy King, NOTHING else !
What you see as Positional or Tactical Play, means NOTHING to this Monster Program. It wants to win and only win !
Positional or Tactical Play, is only a means to an end, for this Program.
It doesn't care what you call the type of Moves it makes.
It just makes the Moves it needs to make, to achieve its goal of winning.
Simple but deadly.