With every new generation and increase in frequency, the latency has gotten higher and higher.
Is this something to worry about?
Would it not be more beneficial to focus on making timing improvements instead of just increasing the speed?
Changing command rate to 1T has a bigger impact (~5%) than slightly lower cas and pretty much all ram can do it without even bumping the voltage.
>>59048666
>With every new generation and increase in frequency, the latency has gotten higher and higher.
Incorrect.
http://www.gskill.com/en/product/f4-3600c15d-16gtz
>>59048995
I haven't checked for ddr4, but when I was running 2133@cl9 for ddr3, 1T seemed more.. unstable, after changing it to 2t then everything appeared to be more stable (no weird time changing error when running 3dmark firestrike or random ass stuttering on anything running).
>>59048666
>the latency has gotten higher and higher
No it hasn't, you tech-illiterate filth. Latency is a factor of both clock speed and timings. A stick of 16-16-16-48 3200MHz DDR4 would have the exact same latency as a stick of 8-8-8-24 1600MHz DDR3 (minor timing fluctuations aside). Plus double the bandwidth on top of that. There are literally no downsides.
>>59049471
This.
http://www.crucial.com/usa/en/memory-performance-speed-latency
The fluctuations with timings on DDR4 once you start getting into the 3000MHz+ range are so minimal as to be irrelevant. A few nanoseconds at most, given there's a very small range of timing sets actually being sold on the market (almost all 3200MHz kits are now 16-18-18-3x).
>>59049554
>ddr3 cl7
11ns ime
>>59049554
It always astounds me how fast computer memory can be.
In the time it takes for a computer to write a few hundred bytes to RAM light can only travel a little over 4 meters.
That just seems insane to me.
>>59048666
Cycle access time is more important the metric than just speed over timings, Satan.
15cas ddr4-2666 is about 11.25ns latency, on par with 9 cas ddr3-1600. 17cl 3200 is 3% slower than cl8 (10.38 ns versus a flat 10) 1600 yet provides twice the bandwidth.
>>59048666
It's normally offset by the higher clockspeeds.
When does the RAM become a bottleneck?
>>59049471
So, price aside, I should buy the highest bandwidth DDR4 my motherboard supports regardless of timings?
>>59050995
When it fails to swap data in time to apply it or does not have enough space to accommodate what you're doing