THE COMPUTER CHESS HAS REACHED ITS CRITICAL POINT

[corres]

CPU is one thing and TPU is another thing.
With the help of TensorFlow library one can use CPU as a TPU (but smaller effectiveness) but one can not use TPU instead of CPU in every respect.
Have you seen the pictures about TPUs of Google in their server room? There is no any chance to miniaturize that system to the scale what one can place them in a living room.
Maybe you do not know but the development of Windows is going to the direction of making it to be a part of a big Windows network.
We are standing at the beginning of the era of big networks and the key machine (the brain of them) of these networks will be in the server parks of Google, Windows, etc.
The own PC, mobile, TV will be a local machine with limited capability only.
This is the not far future.

[Ras]

This means that very soon a computer with 1-2 CPUs and 8-64 cores will become obsolete and regular PC will come from the shop fitted with 1, 2, 4, 8 TPUs

Won't happen anytime soon because Google will not sell them. Renting them for cloud is more profitable.

or whatever else parallel unit comes into fashion.

So we first need a competitor to get up and going, develop and produce that en masse. That's at least two years if Intel decided to go ahead right now.

And then it won't be mass produced anyway because next to nobody needs them in terms of numbers. CPUs? Of course, all these office PCs and web servers. Graphics cards? Sure, all the gamers were what made them big (and put SGI etc. out of business).

But who is the mass audience to make TPUs a mass product? I don't see them, and that's why the regular shop PC will not have 4-8 TPUs in the foreseeable future. Besides, the whole development cost would spread over not-too-many units, making them expensive and limiting their use even more.

I can see the technical feasibility, but the economics are quite a problem.

[brianr]

I don't know what price points they are targeting, but here is one company:

https://www.graphcore.ai/posts/prelimin ... plications

https://supercomputersfordl2017.github. ... phCore.pdf

[Leo]

I think AK is the real A Karpov. If so, welcome.

[syzygy]

I think AK is the real A Karpov. If so, welcome.

He was kicked when guest accounts were banned, sometime between now and 10 years ago...

[Ras]

I don't know what price points they are targeting, but here is one company:

Looks promising, though they havn't actually delivered yet. The current main question is what the actual goal of that company is: delivering a product or getting bought by Google, Intel or the like?

[FWCC]

With the introduction of AlphaZero brute force is now dead and obsolete.Self-learning is now the way to go with new algorithms. Here it is AlphaZero comes along and makes old Computing obsolete destroys the best chess-playing entity in the world ( though they should have used the latest version) and has all the chess programmers not getting any sleep. A new era has arrived. Time to think outside of the box.



FWCC

[lantonov]

But who is the mass audience to make TPUs a mass product? I don't see them, and that's why the regular shop PC will not have 4-8 TPUs in the foreseeable future. Besides, the whole development cost would spread over not-too-many units, making them expensive and limiting their use even more.

I can see the technical feasibility, but the economics are quite a problem.

When your TV and cooking stove start to use neural networks to learn about your viewing and cooking preferences, or your vacuum cleaner uses it to learn its way about the house, not to speak about self-driving cars which are fact at present, this is already a mass production. All I am saying is that it is not for the far future but for a quite forseeable future because we have beginnings of it even now. No need to list these, most people interested in this topic have read about it.
As for companies, it's not only Google. I even suppose that Google will soon lag behind in this competition. Companies like Nvidia, Intel, AMD, Foxconn, Apple who now produce CPU / GPU and other computer components will orient themselves according to demand and complement CPUs with parallel units.

[lantonov]

Out of the tons of articles on this topic, I will give only one:
https://www.netformation.com/industry-p ... e-new-cpus

[pilgrimdan]

Out of the tons of articles on this topic, I will give only one:
https://www.netformation.com/industry-p ... e-new-cpus

thanks for the link...

interesting that the article doesn't mention Apple as one of the companies...