Philip Turner

Likely next summer when I will work on researching the architecture for a living. Until then, the best documentation out there is Apple's developer videos.

https://github.com/philipturner/metal-benchmarks/commit/e83e8280ee6f8c0c16264beb10c0c4580cfc16fc

Bank accesses is granted per-warp. If threads all access the same location in threadgroup memory, the data transfer is fused. This should be the case across all GPU architectures. There's...

Closed due to being stale.

Make an Xcode project and copy the source code to `main.swift`. Also, create a Metal file and copy any shader code you need.

Closed due to being stale.

I'm not sure the clock speeds are correct for this architecture yet. Could you run `sudo powermetrics gpu_power --sample_rate 500` and report the results in a text file?

Have you double-checked whether it's in low-power mode? These clock speeds are news to me: CPU: 3.6 GHz -> 4.05 GHz GPU: 1.398 GHz -> 1.380 GHz

Also, following the Apple developer video, I need to correct this generation's TFLOPS and IPS to "unknown".

@schwa would you mind running the test with the latest version of `applegpuinfo`?