DDR4 vs DDR5 latency: higher bandwidth doesn't always mean lower latency

How this is calculated

CAS latency (CL) numbers are in clock cycles, not nanoseconds. DDR5 runs at higher frequencies, so its cycles are shorter. CL30 at 6000 MT/s is 10 ns. CL16 at 3600 MT/s is 8.9 ns, giving the DDR4 kit lower first-word latency. The transfer-rate ratio is 6000 divided by 3600, or about 1.67x, so bandwidth rises from 28.8 to 48.0 GB/s per 64-bit module or channel. Two populated channels double both totals to 57.6 and 96.0 GB/s. DDR5's two 32-bit subchannels can serve requests more independently and improve scheduling efficiency. For bandwidth-bound workloads such as video editing, scientific computing, and AI inference, those gains matter more than the small latency difference.

Verdict

The bandwidth jump obscures the latency story. DDR5-6000 CL30 lands at about 10 ns true latency, while DDR4-3600 CL16 hits 8.9 ns, so the older kit is faster on this first-word calculation. DDR5-6000 runs at 1.67x the transfer rate and supplies 48.0 GB/s per 64-bit module or channel versus 28.8 GB/s for DDR4-3600. Its two 32-bit subchannels also improve request concurrency. Those gains serve bandwidth-bound workloads like video editing and AI inference, while higher capacity and better power efficiency matter more for servers than the small latency cost.