4 ms·
Intrigued by the single-thread main memory bandwidth being a multiple of what you get from a single SKX. We also see this with Graviton 2. The latency is not te
by jeffbee 6y ago
Intrigued by the single-thread main memory bandwidth being a multiple of what you get from a single SKX. We also see this with Graviton 2. The latency is not terrible, either. How would this much available bandwidth change your choices when optimizing your algorithms?