Going Forward Will Graphics Cards Still be Used for Chess

[daniel71]

Today I'm reading the next Nvidia RTX 4090 will cost $2999 and will offer up to 21,000 Cuda Cores, will be 100% faster than RTX 3090 graphics cards. With the switch to CPU computation again will Graphics cards make a comeback for chess software? The GPU clocks expected are now 2.2GHz and above, which will result in around 81 TFLOPs of compute performance (35.5 TFLOPs on the GA102 or RTX 3090).

Read more: https://www.tweaktown.com/news/80778/nv ... index.html

[Modern Times]

I wonder if the next big breakthrough in chess will be a hybrid engine that uses both CPU and GPU effectively. Sure, Lc0 needs a CPU to run, but the GPU is the dominant part.

[towforce]

Going Forward Will Graphics Cards Still be Used for Chess

Some possible future scenarios:

1. "Yes". To be competitive, you have to spend a small fortune on graphics cards

2. Computers get so close to the upper limit of chess skill that further improvement is no longer worthwhile (this is similar to the "death by draw" scenario)

3. Equivalent TPUs become a lot cheaper than GPUs (for those who just want AI and not actual graphics). This seems quite likely to me.

4. A mathematician finds a way to quickly discover whether a given position in won, lost or drawn (doable IMO)

5. A mathematician finds a way to discover a small number of deep patterns in chess which will result in accurate evaluations with only a small amount of processing (such patterns exist and are discoverable IMO. I provided evidence for this POV in another thread). If this happens, a cheap SOC will be literally all that's needed - bye bye expensive GPU!