Not sure if this topic is in the right forum subfolder, but,
Would it be more beneficial for a large GPU to disable some cuda cores when not needed, and run the remaining cores at a higher frequency?
Currently, most GPUs run all their cores all the time when gaming.
In less demanding games where shader count isn’t as important, data is sent to passive cores when active cores are still busy, and active cores are then left idling when their work is finished, resulting in shaders constantly being triggered into a sleep and wake state, which results in an increased heat output to the cooler, which will make it harder for the GPU to hit higher frequencies.
The passive cuda cores that aren’t disabled, still contribute to heat generation towards the cooler, and more heat means lower boost frequency, and higher Watts necessary to reach the same boost frequencies.
If a GPU would have 2 Cuda core clusters totalling 2k cuda cores, with for instance 1000 faster cores (at 2,5Ghz) and 1000 regular cores at 2Ghz boost,
And the game running wasn’t very GPU demanding, one could technically run higher fps, if only the performance cores were enabled (at the faster 2,5Ghz speeds), while still staying within the GPU’s TDP.
If the game is very graphics intensive, like DX12, 4k, high LOD, the additional cores can be activated, and the overall boost frequency lowered to match the low speed cores. (say: up to 2Ghz).
Under heavy load, there should not be any difference between this newer GPU/driver and the currently used version.
For games that require less graphics power (eg: DX9, 1080P),
The GPU could run a game at the first cuda core cluster with the 125% core frequency of 2,50Ghz, while utilizing for instance 80% of TDP.
Currently used GPU and driver model at less intensive games, will have all cores active at for instance, an 80% core frequency, and 80% of TDP.
So while the TDP values are similar, the frequency of one block of shaders (cuda cores) can be increased, resulting in higher FPS. This is good if you want to play a game at 90, 120, or 144fps compatible monitors.
Performance wise, the whole concept is basically similar to running a 3060 as primary GPU, which overclocks to 3070 speeds when games don’t require much processing power, but when the game becomes very demanding it would perform as a fictive 2x 3050 GPUs in SLI configuration.