Ok guys so referencing this post.
Above should be sticky imo
He first off states only raise frequency if latency isnt compromised.
Ironically the combination for me a speed/latency combo that is barely mentioned anywhere.
For reference I have samsung B die.
Which is 3000mhz and CAS 12
3000/12 = 250
Other combinations are all inferior. Some dont even post.
3000/12=250, posts and passes stability tests, used as daily driver config for many months. Needs 1.45v
3200/14=228, XMP spec of my ram, posts and stable (on this 9900k, wasnt on my 8600k, which was the reason I started using the 3000/12 config), this is significantly inferior on latnecy on AIDA64 and interestingly caues one of my favourite games to stutter much more than 3000/12. Much lower voltage than 3000/12 needed.
3200/13=246, this would have been great, but even at 1.45v doesnt even post.
3333/14=238, have managed to boot into windows, no stability test carried out yet. Really want to get closer to the latency of 3000/12.
3400/14=242, doesnt post
3400/15=226, posts but as crappy as 3200/14 for latency
Not tried anymore, as I feel getting close to 3000/12 latency at very high frequencies is not going to happen, 3000/12 seems godly yet noone uses it?
So the question is, if you was in same situation, would you favour 3000 with cas 12? Also 3000 with Cas 12 has almost same bandwidth on AIDA64 as 3000CL14 XMP, but If I tune the 3200 config a bit it pulls away somewhat, same with 3333mhz, however if I do a % calculation, the % increase in bandwidth is a fraction of the % decrease in latency, and latency seems to be benefit software more?
Latency as reported in AIDA64
3200 47-48ns tuned 45-47ns