News for hackers. Stuff that matters.

A comment on Apple's M4 Max chip is the fastest single-core performer in consumer computing. Parent comment

Re:Apple's M4 Max chip is the fastest single-core performer ... by jrockway on Friday November 01, 2024 @05:17PM (#42021640)
It depends on the game. If there are a lot of game simulation calculations to do for every frame, then you're going to be CPU constrained. If it's a storybook that you're walking through and every pixel is raytraced using tons of 8000x8000 texture maps, then it's going to be GPU constrained.

Most games are combinations of the two, and so some people are going to be CPU limited and some people are going to be GPU limited. For games I play, I'm often CPU limited; I can set the graphics to low at 1280 x 720, or ultra at 3840 x 2160 and get the same FPS. That's CPU limiting.

[ Parent | on HN ]
Re:Apple's M4 Max chip is the fastest single-core performer ... by magicalhippo on Friday November 01, 2024 @05:59PM (#42021977)
I recently swapped out my AMD 3800X with a 5900X as an alternative to a full platform upgrade. I got it mostly for non-gaming workloads, however I do enjoy games.

Paired with my aging but still chugging 2080Ti, the max framerates in games I play did not significantly increase.

However I did get a significant improvement in 99-percentile framerate, and the games feel much smoother. YMMV, but it surprised me a bit.

[ Parent | on HN ]
Re:Apple's M4 Max chip is the fastest single-core performer ... by rbanffy on Friday November 01, 2024 @05:53PM (#42021924)
> If there are a lot of game simulation calculations

Why not move at least some of that into the GPU as well? Lots of different branchy code paths for the in-game objects?

[ Parent | on HN ]
Re:Apple's M4 Max chip is the fastest single-core performer ... by TinkersW on Friday November 01, 2024 @07:01PM (#42022389)
Latency of communication with GPU is too large, and GPUs suck at things CPUs are good at.
[ Parent | on HN ]
Re:Apple's M4 Max chip is the fastest single-core performer ... by rbanffy on Saturday November 02, 2024 @06:57AM (#42025546)
Wouldn’t the latency be because of no shared memory and game context data having to go over the PCIe bus?

GPUs suck at branchy code, but is that the case with this simulation data? I can see why it’d suck for NPC simulation, but not physics.

[ Parent | on HN ]