4 ms·
Fair enough. The full AMD pipeline contains 4x 128-bit execution ports. But this is a bit complicated, so I was ignoring that part of the core. IMO, the more i
by dragontamer 9y ago
Fair enough. The full AMD pipeline contains 4x 128-bit execution ports. But this is a bit complicated, so I was ignoring that part of the core.
IMO, the more important bottleneck is the load/store unit. Intel's system has 2x 256-bit loads + 1x 256-bit writes to the L1 cache.
While AMD's system has 2x128-bit load+store units for its L1 cache.
Thus my numbers above: 3x256-bit for Intel (2 reads + 1 write), while 2x128-bits for AMD (2x read-or-write).