I’m not sure comparing benchmarks of yet to be released hardware is super useful, especially given intels recent track record. I’m more likely to believe nvidia

I’m not sure comparing benchmarks of yet to be released hardware is super useful, especially given intels recent track record. I’m more likely to believe nvidia’s shipping estimates there. I’ll given them both props for not pulling that ludicrous bullshit graph that apple had in the original M1 announcement. An interesting feature of SPEC that strongly disadvantages Intel is the a number of SPEC tests use ‘long double’ which every platform except Intel is always 64bit double running on modern execution units. For Intel (and AMD) the ABI for 32bit, and 64bit on linux and macOS is to use 80 bit floats. That necessarily means x87. So no SIMD, a vastly slower execution engine, and higher precision. There is no case where x87 performance comes close to the primary floating point pipelines.