Cheating has been a problem in multiplayer titles on PC and Valve’s CS: GO is no exception, however, Valve has been cracking down on the cheater in the game and according to the Valve, the company uses 1,700 CPUs for its VACnet that work non-stop just to find cheaters in the game.
This is according to Valve programmer John McDonald, who spoke about how Valve managed to detect and ban cheaters as the community for CS: GO increased significantly.
He noted that when VAC bans became obsolete in the game, he and Valve looked to automate the process through deep learning instead of hiring hundreds of curators and VACnet was born.
He added that VACnet is not a new version of VAC but is an entirely new system that uses deep learning to “analyze players’ in-game behavior, learn what cheats look like, and then spot and ban hackers based on a dynamic criteria”.
However, to bring VACnet to life, Valve and John McDonald had created a server farm to monitor millions of players in CS: GO. To evaluate players, Valve needed “four minutes of computation, amounting to 2.4 million minutes of CPU effort per day” which required 1,700 CPUs to do that.
Valve had to spend at least a few million dollars on the VACnet hardware which included 64 servers blades with each blade having 54 CPUs and 128 GB of RAM.
This represents one of the most expensive and beefiest anti-cheat measure for a single game. He further added that VACnet is working flawlessly and can potentially help non-valve games and other stuff on Steam.
Deep learning is this sea-change technology for evolutionary behaviour. We think that it is really helping us get developers off of the treadmill without impacting our customers in any way. Our customers are seeing fewer cheaters today than they have been, and the conversation around cheating has died down tremendously compared to where it was before we started this work.
Cheating has been a problem in video games a recent example of that is PUBG and if VACnet works the way Valve has been describing it, then it will work wonders for the game.