LCZero update

[Milos]

Is a GeForce 10, 1060 any good for training?

I have one of those. I also have an older Quadro 5000 but it's not as powerful.

A 1060 is more than enough. Actually, any videocard is good. Every single bit adds to the total.

Can it run on AMD cards or does it require Nvidia?

For training not, it needs TenorFlow which doesn't run in general on AMD (there are few exceptions).
Games you can ofc run, for inference only OpenCL is enough.

[CMCanavessi]

Is a GeForce 10, 1060 any good for training?

I have one of those. I also have an older Quadro 5000 but it's not as powerful.

A 1060 is more than enough. Actually, any videocard is good. Every single bit adds to the total.

Can it run on AMD cards or does it require Nvidia?

For training not, it needs TenorFlow which doesn't run in general on AMD (there are few exceptions).
Games you can ofc run, for inference only OpenCL is enough.

Actually that's incorrect. It does NOT require TensorFlow, and I run the training games in my PC with an AMD Radeon 380 Nitro.

[Milos]

Is a GeForce 10, 1060 any good for training?

I have one of those. I also have an older Quadro 5000 but it's not as powerful.

A 1060 is more than enough. Actually, any videocard is good. Every single bit adds to the total.

Can it run on AMD cards or does it require Nvidia?

For training not, it needs TenorFlow which doesn't run in general on AMD (there are few exceptions).
Games you can ofc run, for inference only OpenCL is enough.

Actually that's incorrect. It does NOT require TensorFlow, and I run the training games in my PC with an AMD Radeon 380 Nitro.

Seems you didn't quite understand what I wrote.
I said, you can run games but not train the network without TensorFlow. I thought you'd understand such basic stuff as that running training games is not the same as training the network.

[noobpwnftw]

Seems you didn't quite understand what I wrote.
I said, you can run games but not train the network without TensorFlow. I thought you'd understand such basic stuff as that running training games is not the same as training the network.

I can see why we shouldn't say a complete sentence to explain certain things.


So, let me put it this way: I have quite a bunch of computers, they all have some mid-range AMD video cards and they run Linux, so is there anyone to spoon feed me with a package that I can run with a simple script to contribute to the project?

[CMCanavessi]

Seems you didn't quite understand what I wrote.
I said, you can run games but not train the network without TensorFlow. I thought you'd understand such basic stuff as that running training games is not the same as training the network.

I can see why we shouldn't say a complete sentence to explain certain things.


So, let me put it this way: I have quite a bunch of computers, they all have some mid-range AMD video cards and they run Linux, so is there anyone to spoon feed me with a package that I can run with a simple script to contribute to the project?

Yes, the package that I provided earlier will run perfectly fine for that purpose. I will make a new one later today, with updated binaries.

[Werewolf]

can I ask: what's the approximate strength at the moment? It'll be fun to see how fast it improves

[noobpwnftw]

Seems you didn't quite understand what I wrote.
I said, you can run games but not train the network without TensorFlow. I thought you'd understand such basic stuff as that running training games is not the same as training the network.

I can see why we shouldn't say a complete sentence to explain certain things.


So, let me put it this way: I have quite a bunch of computers, they all have some mid-range AMD video cards and they run Linux, so is there anyone to spoon feed me with a package that I can run with a simple script to contribute to the project?

Yes, the package that I provided earlier will run perfectly fine for that purpose. I will make a new one later today, with updated binaries.

Thank you, and by a bunch I mean about 100 GPUs to be fair to the amount I'm donating to fishtest, so at least I don't give a false impression to whoever thinks that way that I'm favoring one thing over the other) and I can handle some basic stuff but just in case please try to save my time with an auto updater or something if the training source or the binaries used is going to change quite often.

[CMCanavessi]

can I ask: what's the approximate strength at the moment? It'll be fun to see how fast it improves

Hard to say, as it varies A LOT with time control and GPU used, but you could estimate, in general, around 800-1000 elo.

[CMCanavessi]

Here's a new package I've made: http://www.mediafire.com/file/s99iadlt6 ... 20Gen10.7z

It has all the latest files at the time of posting this.

lczero.exe is build 101
client.exe is v0.1 from github
network is Network ID 14, Gen 10


It's NOT an official package, I just put all the files together for ease of use. Use it at your own risk or go and get the files yourself from their original place

[Spliffjiffer]

can i contribute to the "training-part" with an APU as CPU/GPU ?