InferenceMAX

Hi,

Not sponsored or affiliated, but wanted to cross-pollinate InferenceMAX ([benchmark page](https://inferencemax.semianalysis.com/), [GitHub](https://github.com/InferenceMAX/InferenceMAX), [blog post](https://newsletter.semianalysis.com/p/inferencemax-open-source-inference)). Just wanted to share here because it'd be great for developers like myself and companies shopping for hardware (like mine) if Intel Arc systems would be included in this open-source benchmarking effort (currently Nvidia and AMD, Google TPU and Amazon Trainium in the works). It appears to take a holistic approach, viewing systems as the whole hardware + software stack, factoring in performance for price, and running benchmarks nightly to capture rapid progress. That sounds to me like an apt stage for Intel's work, based on the online discussions I've seen, which have been along the lines of "very promising, hope it succeeds, it's rapidly improving" but also "it's early days, not sure what support is like across the stack from Intel developers and the community."

The first rounds of their benchmark results are focused on systems with prohibitively high entry points, so I'd love to see B60 Pro systems (and future Intel systems) on there. No idea who at Intel needs to talk to the InferenceMAX team but would love for it to happen. If there's a better channel to share this, please let me know!

PS Already, Nvidia's "CUDA halo" may be questioned by the InferenceMAX team's suggestion that [AMD had less bugs](https://newsletter.semianalysis.com/p/inferencemax-open-source-inference#:~:text=On%20the%20AMD%20front%2C%20we%20ran%20into%20fewer%20bugs%20while%20developing%20InferenceMAX%E2%84%A2%2C%20and%20these%20bugs%20were%20easier%20to%20fix.) in their testing. Once Intel systems are in the benchmarks, we may all benefit from a clearer view of the different ecosystems.