AlphaZero beats AlphaGo Zero, Stockfish, and Elmo

[jhellis3]

As has been mentioned previously, one can not really make direct core count comparisons in this case.

The most "fair" metric I can think of using is system power consumption, and I would guess that SF was at a bit of a disadvantage in this regard. Regardless, the writing is clearly on the wall.....

[Milos]

4 hours my ass (pardon my french). Try training it on state-of-the-art 1080.
Fully trained network requres 12h on 5000 gen1 TPUs for self-games and 64 gen2 TPUs for training itself.
Gen1 TPU is like 30x K80 which is like 5x 1080 in performance.
So you'd need like 375k training days with 1080, which is like 1000 years!!!

Your math is wrong. I think it is doable with a distributed effort smaller than what was used for Stockfish.

Care to elaborate, add any substance beyond your one-liner childish reply?

5000*5*12/(24*365) = 34.25 years

So you claim single gen1 TPU is just only like 5x stronger than Nvidia 1080 GPU????
Gee, you really have no clue about hardware, do you?
TPU is like 92 TOPS, 1080 is 0.3TFLOPS. Please educate yourself before trying to run discussions with one-liner replies.

[clumma]

Days of Alpha0 on normal hardware are years away.

Sounds like you're conflating model training with running the program.

For chess it would take even more.

For chess it takes less.

But keep on dreaming, no one can take that from you.

Your case isn't advanced by being an asshole.

-Carl

[jhellis3]

Why are you using old GPU architectures for your comparison. The 1080 has no tensor units, so obviously it will be much slower.

The V100 on the other hand should be ~75%+ the performance of a TPU2.

[Milos]

Days of Alpha0 on normal hardware are years away.

Sounds like you're conflating model training with running the program.

Sounds you can't really read well, coz I was comparing training time from the start and arguing about ludicrous 4 hours figure.

For chess it would take even more.

For chess it takes less.

Again no clue, no substance, no elaboration, just one-liner.

But keep on dreaming, no one can take that from you.

Your case isn't advanced by being an asshole.

-Carl

And you clearly show that your lack of manners largely surpasses any competence you might have. You are really taking it too personally.
It's not that bad being wrong from time to time .

[Dan Cooper]

Alpha hardware equivalent was somewhere 1024 standard cores.
How 1024 cores compare with 64 cores?
How scientific is that.

Has Jonny on 2000 cores ever beaten Komodo 28-0?

[Milos]

Why are you using old GPU architectures for your comparison. The 1080 has no tensor units, so obviously it will be much slower.

The V100 on the other hand should be ~75%+ the performance of a TPU2.

Show me open community that has access to V100?
I am comparing to what common ppl have access too.
You seems to be sticking to apples and oranges most of the time.

[jhellis3]

Ok, you are useless, bye.

[Milos]

Ok, you are useless, bye.

Hehe too many Google fanboys around here it seems .

[Rémi Coulom]

4 hours my ass (pardon my french). Try training it on state-of-the-art 1080.
Fully trained network requres 12h on 5000 gen1 TPUs for self-games and 64 gen2 TPUs for training itself.
Gen1 TPU is like 30x K80 which is like 5x 1080 in performance.
So you'd need like 375k training days with 1080, which is like 1000 years!!!

Your math is wrong. I think it is doable with a distributed effort smaller than what was used for Stockfish.

Care to elaborate, add any substance beyond your one-liner childish reply?

5000*5*12/(24*365) = 34.25 years

So you claim single gen1 TPU is just only like 5x stronger than Nvidia 1080 GPU????
Gee, you really have no clue about hardware, do you?
TPU is like 92 TOPS, 1080 is 0.3TFLOPS. Please educate yourself before trying to run discussions with one-liner replies.

1080 ti is 11.3 TFLOPS:
https://www.anandtech.com/show/11172/nv ... t-week-699

A TPU is 45 TFLOPS:
https://arstechnica.com/information-tec ... ute-cloud/