Comparisons
Two things routinely conflated, put side by side. Neither column is the winner — what decides is the workload and the machine.
ISA vs MicroarchitectureArray traversal vs Linked-list traversalLatency vs BandwidthCPU vs GPUCPU cache vs OS page cacheHardware thread vs OS threadPolling vs InterruptsBranch vs Branchless
Latency vs Bandwidth
Two different resources, routinely conflated. Latency is how long one access takes; bandwidth is how many bytes per second the system can sustain. A workload can saturate one while leaving the other almost untouched.
LatencyOpen lesson →
Strengths
Improved by locality, prefetching and overlapping independent misses
Costs
Cannot be hidden when accesses are dependent
Use when
Pointer chasing, tree and graph traversal, dependent loads
BandwidthOpen lesson →
Strengths
Improved by moving fewer bytes: better layout, smaller types, compression
Costs
Cannot be raised by adding cores — they share it
Use when
Streaming, scans, large copies, dense linear algebra
| Dimension | Latency | Bandwidth |
|---|---|---|
| Typical symptom | Low IPC, low bandwidth, high miss latency | Low IPC, bandwidth near peak |
| More cores help | Sometimes — more independent chains | No — they contend |
| Prefetching helps | Only if the address is predictable | Yes, up to the bandwidth limit |