Do RAM speeds now match CPU/are L3 misses less expensive?

[oxf77]

For years CPUs were much faster than memory.

The latest DDR5 seems to have 4800/5200/5600 MHz speeds.

Does this mean memory speeds have finally caught up with CPU speeds and therefore L3 cache misses are less-costly?

 

[Mr Evil]

L3 will always be faster than main memory, regardless of clock speed. This is because main memory is physically further away, so the latency is higher, typically by a factor of 10 in current systems.

 

[oxf77]

L3 will always be faster than main memory, regardless of clock speed. This is because main memory is physically further away, so the latency is higher, typically by a factor of 10 in current systems.

That wasn't my question

When memory wasn't as fast as the CPU L3 cache misses cost ~200 ns, I.e this was the latency of RAM (assuming the address is in RAM).

Do the DDR5 speeds I quoted now suggest L3 cache misses are not as expensive as 200ns?

Or in other words, is memory now able to keep up with the CPU?

 

[Tetras]

Or in other words, is memory now able to keep up with the CPU?

As far as I know, DDR5 (even fast DDR5) didn't improve latency on fast DDR4 (see table), so cache is still very important. If AMD didn't think so, I'd imagine they would not be releasing an AM5 X3D.

 

[VinylRichTea]

Not entirely true, I bought the Team Group 7600MHz from OCUK for my 13900k and tweaked the secondary timings (I've not even touched the primary yet) and I've got a latency of 58ns down from 65ns. With a bit more tweaking I think I can get it to 53-54ns. In contrast, with my 10900k, I had some GSkill DDR4 at 4133, CL16 at around 48ns.