Win Modern Go

When I first started playing Go, I was looking for something simple to make new friends. I did not expect the burning necessity to build latest and greatest state of the art dedicated Ubuntu Linux computer around it.Nakigoe

Go Opening Moves

Use AlphaGo Fuseki to find the best opening moves.

The larger the numbers are, the greater the chance for black to win.

The smaller the numbers are, the greater the chance for white to win.

Mid‑Game

Use Kifu Search to find the best continuations.

Pay attention to the win-rates, they are shown at the right side of the screen for the current player (independent of colour).

Shift + Mouse Click puts a black stone on the board.

Control + Mouse Click puts a white stone on the board.

Choose "Around Stones" in the top menu to perform area search.

Press the mouse button and drag to select specific search area!

Choose "Search W" to find White continuations.

Choose "Search B" to find Black continuations.

There are about 80,000 professional games stored, so on rare occasions the database isn't fully searched, but all results are cached, so repeated search can gain new results. The number of games searched is displayed above the win-rates on the right side of the screen.

Artificial Intelligence

Use all modern free engines:

Engines Pack they are all start ready, click and analyze.

GoGui from FueGo, GoGui is a visual interface for generating plain moves. GoGui is the most simplistic, lightweight and the most straightforward Go GUI (Graphical User Interface) to run Go AI.

KataGo

KataGo engine is tricky to set up, takes a lot of reading, trial and error, but it's doable, especially if You are pressing forward to win.

Before You plug in KataGo AI into any GUI (Graphic User Interface), and every time You want to try out an absolutely new weights set, You have to run the following command in the command line terminal to generate GPU settings automatically:
katago.exe benchmark

KataGo AI works just fine with GoGui. If You experience any problems with KataGo, that's probably the weights file size fault, download 20 block weights manually from katago website: 20 block weights Store it inside the Katago folder and name it default_model.bin.gz. Run the benchmark command for every new weights file you install!!!
katago.exe benchmark

numSearchThreads is another key setting, which depends on Your GPU memory size. That is, number of threads to use in search. Set numSearchThreads = 16 and double it until the thing crashes. You can also check GPU memory in any app for GPU load monitoring. If KataGo crashes, decrease the value twofold.

You want to keep the original settings maxVisits=500 to receive stable results on the board.

Comment maxTime out to find the best move.

It very well may take up to 5 minutes to calculate just one move on a weaker GPU. Leave the time control to Katago to receive best stable results on the board.

nnCacheSizePowerOfTwo = 27 for approximately 1.5 Gb of RAM, increase nnCacheSizePowerOfTwo to 28, 29 in steps and monitor the RAM usage via Task Manager or any other RAM monitoring app. You cannot use 20 and 40 blocks models unless Your RAM settings are maxed out.

Close GoGui when not computing any move to free up GPU and RAM from KataGo, and let the equipment rest.

Zenith Go

For out of the box solution try paid Zen engine for CPU, it's best for simple equipment and a great opponent against Japanese players Zenith Go 7 Zen

Taiwan players have a strong tendency to beat Japanese Go AI into pulp. Taiwan Go gameplay is different from Japanese. Even 15‑year‑old Taiwan kid can beat Zenith Go home PC version sometimes.

Anyway, set the mode to 120sec.

How to do it really well, the fine points:
You can expand the upper time limit for the AI from 120 seconds to 210 seconds!!!
Edit the SGF save file ZT value in any text editor, change ZT[120] to ZT[210]

Sometimes the app plays overly aggressively, attacking when it's not appropriate with no predictable outcomes,
or way too far,
or shrivelly inefficiently defensively extending a corner to the center of the side, when it's time to expand twice as much, up to the opposite corner.

In that case you'll have to run the app 2‑9 times to find the move You consider the best.

In general You should expand the corner up to another corner (AlhpaGo style), and settle the corners themselves safely, making life as early as possible.

Expand far.

Study AlphaGo games to find safer, and / or more territory gaining moves.

Every AI plays better when the board position is settled and simple.

Eliminate any technical debt to reduce the number of future computations. Settle positions early! Avoid overly aggressive attacks with no easily predictable outcomes.

And remember - there are no stones!

Proper Hardware

For the love of God, do not use Your laptop to run any AI, for fear Your video‑chip will burn. This is extremely computational process with intense overheat for the chip for very prolonged time, from 2 to 20 minutes. The video‑chip just overheats.

Set up a simple, but separate PC under $1000 with the best cooling system at Your hand. Motherboard, RAM, and processor are paramount. For every AI, the processor inner memory cache size is everything. The frequency of the processor is vital also. You should absolutely consult AI developer what kind of CPU technology they put their money on. It's an AI processor type, processor with huge cache size and frequency.

Grab cheap monitor, and cheap ATI video‑card.

Buy SSD and SLI video‑cards pack later in life. You cannot run a good video‑card without having top‑grade CPU, RAM, and motherboard.

Trust me, You'll save on future repairs of Your laptop if You invest into a bare, but separate PC.

You'd better have a separate PC to run Zen engine, altogether below $1000:

Cheap monitor.

Cheap ATI video-card.

Probably you might want to buy a pair of nVidia a6000 48Gb video‑boards later.

Best MotherBoard which supports up to 128Gb RAM and multiple GPUs (you can buy them later).

Best & cheap (cheap is good) seventh generation dual‑core Intel processor: i3-7350K (4M Cache, 4.20 GHz) , that processor supports up to 64Gb RAM, (and has Intel® SSE4.2 instruction set, see below).

Please note minimal system requirements for Adobe Photoshop, which runs a modern computer vision AI: «Graphics card with at least 4 GB of GPU memory» and:

Intel® or AMD processor with 64‑bit support and with instructions set of SSE 4.2 or later (Intel® Core™ i7 and later)

AMD processor might be a good choice if You are going after the CPU frequency.

«The sweet spot for running Adobe Premiere Pro is a fast CPU with 8 cores. Core i7 or Core i9 Intel processors or AMD equivalents are strongly recommended. Fast clock speed at least 3.2 GHz, or higher. 8 cores are ideal for Premiere Pro. The application can use more cores, but without significant added benefit. Depending on the task, Premiere Pro runs at 93‑98% efficiency with 8 cores».

If you want to go fancy about Go and to use more than 2x GPUs, You might have to install a server CPU that has support for 4x GPUs and even 8x GPUs like in Facebook Elf (the best Go AI for home PC, requires Ubuntu Linux system).

Tower Workstation

Above all, use the best CPU & GPU cooling system available around you. Explain to computer technician that the CPU & GPU are going to run for hours (like every game of Go), on their top frequency and maximum voltage. Ask to use the best thermal paste, they start with thermal conductivity of 12,5 W / mk Kryonaut

For laptop users: new 2mm thermal pads for other micro-chips, good pads start with thermal conductivity of 8 W / mk Minus Pad

Oh my God Jah, please, give me a workstation. And a country. And a place to live. Pharaoh, do let us go from Russia! We want to play Go in the Silicon Valley for three days.