5 ms·
Is that available to any core or just a sum-all-up and has latency penalties when going around?
by consp 3y ago
Is that available to any core or just a sum-all-up and has latency penalties when going around?
- hypercube33 3y agodepends on NUMA node config so I believe this is combined on the whole chip if all cores are working on threads with their local 16GB HBM, theoretically.