3 ms·
Or memory bandwidth (800GB/sec). Or iGPU perf, which of course needs the above mentioned bandwidth. Or ML inference performance (5 tokens/sec with llama 65B).
by sliken 3y ago
Or memory bandwidth (800GB/sec).
Or iGPU perf, which of course needs the above mentioned bandwidth.
Or ML inference performance (5 tokens/sec with llama 65B).