4 ms·
Bets aren't worth much. It depends on the workload. Throughput-wise the Xeon Phi could handle 4-way SMT, so we ran the 68-core KNL node with AVX512 workloads
by marmaduke 6y ago
Bets aren't worth much. It depends on the workload. Throughput-wise the Xeon Phi could handle 4-way SMT, so we ran the 68-core KNL node with AVX512 workloads and did quite well, compared to equivalent conventional Xeon nodes.