3 ms·
Isn't the bottleneck in any sequential access case the memory bandwidth? IOW: Are the scalar instructions slower than memory bandwidth?
by Peaker 8y ago
Isn't the bottleneck in any sequential access case the memory bandwidth?
IOW: Are the scalar instructions slower than memory bandwidth?
- obl 8y agoyou can saturate memory bandwidth without SIMD, since you can issue at least 2 8-byte scalar loads per cycle. it does not leave much room for actual processing though