Time settings will be 30 minutes + 3* 30 seconds byo-yomi. Japanese rules.
AQ is an open source artificial intelligence Go program. AQ’s author Yu Yamaguchi published his strong Go program AQ in late 2017. This program has played in a few computer go competitions. In AI Ryusei-sen where it was defeated in the semi-finals by FineArt, placing 4th spot. This program also played in the UEC cup where it lost to DeepZenGo in the semi-finals. Anyone can download this program at https://github.com/ymgaq/AQ/releases and run it on Sabaki.
Currently, it is the strongest Go program that the public can use. It has a 87% win rate over Leela 11 and has a 9d rank on OGS, Tygem and Fox. This week on Tygem, AQ has won against Li Xiangyu 3P and Chen Hao 5p so it will be an interesting game to follow this Saturday!
“What the heck is 3pm PST?” you might ask.
o 12 pm - midnight - WET (London) corrected
o 12 pm - midnight - CET (Berlin)
o 1 am EET (Helsinki) - technically sunday
o 2 am EST (Moscow) - technically sunday
“AQ has been officially running at Yugen no ma, which is a Go server managed by Nihon Ki-in, since last week and the result of 100 games against Japanese professional players was 95-5 (win-loss). The hardware is dual GPUs (GTX 1080Ti) and i7-8700K. If you use a single GTX 1080Ti, the elo rating will be reduced by 100-120.” – Author of AQ
The 95-5 win against Japanese pros is from hardware that is about 4 times stronger than the GTX 970.
The ELO between two GTX 1080 Ti and one GTX 970 (what you will be playing) is about 200. So it will be weaker (Not 4 times weaker, about 200 ELO) than the games you see on the Yugen no ma server.
A computer with 2 GTX 1080 Ti’s typically costs $3,000 USD to buy
A computer with my hardware typically cost $1,200 USD to buy
The good thing about Hajin playing against the hardware I have is that people typically don’t spend $3,000 USD on a computer however $1,200 is much more reasonable. Therefore you can expect the strength of the AQ program Hajin will be playing is the strength of the AQ program you will get if you download the program and play it on your own home computer.
Earlier in the day I wanted to test ZenLeeBot ahead of the game with Haylee, so I had run ZenLeeBot against a high-end computer, that was running AQ with GTX 1080 and an Intel Xeon E5-1650 v4 @ 3.60GHz . ZenLeeBot had won 2 out of 3 games. Time was 5 minutes + 5*30
More powerful cards have been used. If you want to see top hardware against top pros, go to the Yugen no ma go server. You will get 100 games to see, not just 1 against a pro. You have seen that 2/3 games have already been won against top hardware by the GTX 970 system, showing that the 970 system is not weak. The majority of AQ’s strength comes from the program, not the hardware.
AQ has been officially running at Yugen no ma, which is a Go server managed by Nihon Ki-in, since last week and the result of 100 games against Japanese professional players was 95-5 (win-loss). The hardware is dual GPUs (GTX 1080Ti) and i7-8700K. If you use a single GTX 1080Ti, the elo rating will be reduced by 100-120.
So, from 2x1080Ti, to 1x1080Ti it’s 100-120 Elo less
From 1x1080Ti, to 1xGTX970, it’s another … Elo less
From i7-8700K, to i5-6500, it’s yet another … Elo less
The point you are making is that some of AQ’s strength comes from hardware. I agree, if you want to see enthusiast strength hardware against pros, there are hundreds of games available online. This is not going to be one of them. This is going to be high end hardware, that is more readily obtainable to the public, which HAS NOT been tested against pros before.
You can see here https://i.imgur.com/FVyKGUR.png The video-card prevalence (taken from Steam Survey February 2018) compared to the card strength of the GTX 970 (taken from User benchmark). You can see the vast majority of cards that are owned are actually weaker than the GTX 970. The ownership rates for the GTX 1080 is 1.05% and GTX 1080 Ti 0.47%. This enthusiast hardware is NOT COMMON amongst households and is not the aim of this match-up. The aim of this is to show the strength of the AQ program on hardware that doesn’t cost thousands to obtain.