What is the formula Buildzoid uses to calculate memory speed?

[Nena Trinity]

Lets say how do I calculate which of the following is faster? No I do not wanna know which is faster but rather how to calculate it...

A) DDR4-3200MHz CL14-14-14-34

B) DDR4-3600MHz CL16-18-18-38

C) DDR4-3800MHz CL18-19-19-39

[Naijin]

31 minutes ago, Nena Trinity said:

Lets say how do I calculate which of the following is faster? No I do not wanna know which is faster but rather how to calculate it...

A) DDR4-3200MHz CL14-14-14-34

B) DDR4-3600MHz CL16-18-18-38

C) DDR4-3800MHz CL18-19-19-39

Stumbled across this topic

https://www.overclockers.com/forums/showthread.php/773720-Calculating-memory-speed-CAS-vs-MHz

 

You could divide the CL latency (since those are ticks on the actual frequency) by the rated frequency.

14/3200 = 0.004375

16/3600 = 0.0044444444444444

18/3800 = 0.0047368421052632

 

Multiply by x1000 to get your actual latency at rated frequency in nanoseconds.

3200 CL14 = 4.375 ns

3600 CL16 = 4.444 ns

3800 CL18 = 4.737 ns

 

Lower would be better (faster timing compared to clock) so 3200 at CL14 would be a better option, but only very slightly. It's also not representative in real-world performance but only theoretical. The actual transfer rates are usually better with higher frequency memory.

[boggy77]

13 minutes ago, Naijin said:

3200 CL14 = 4.375 ns

3600 CL16 = 4.444 ns

3800 CL18 = 4.737 ns

so in addition to this, you need to take into account the frequency as well.

so the 3200CL14 kit has 7% faster latency than the 3800CL18 kit, but the 3800 kit is 19% faster based on frequency.

so in this case, I'd say the 3800CL18 kit is better.

[camieabz]

Crude way to calc it. With 3200 with C16, divide 3200 by 16 = 200. 3200 with C18 would be 177.8 (slower). So higher is better (in theory).

 

3200/14 = 228.6

3600/16 = 225.0

3800/18 = 211.1

 

However, this crude calc is best used for comparing RAM timings of RAM of the same or similar speeds. When the speeds are different, it's not so pertinent. There's also the real world fact that some people won't get their RAM to run at 3800, so the theoretical speed of it is useless. The simple answer is to choose the lowest Cas RAM for the fastest RAM speed you can actually run at. The other real world answer is to pick the fastest stable solution, so you don't spend the next 2-3 years chasing your timings.

[Nena Trinity]

5 hours ago, Naijin said:

Stumbled across this topic

https://www.overclockers.com/forums/showthread.php/773720-Calculating-memory-speed-CAS-vs-MHz

 

You could divide the CL latency (since those are ticks on the actual frequency) by the rated frequency.

14/3200 = 0.004375

16/3600 = 0.0044444444444444

18/3800 = 0.0047368421052632

 

Multiply by x1000 to get your actual latency at rated frequency in nanoseconds.

3200 CL14 = 4.375 ns

3600 CL16 = 4.444 ns

3800 CL18 = 4.737 ns

 

Lower would be better (faster timing compared to clock) so 3200 at CL14 would be a better option, but only very slightly. It's also not representative in real-world performance but only theoretical. The actual transfer rates are usually better with higher frequency memory.

 

5 hours ago, boggy77 said:

so in addition to this, you need to take into account the frequency as well.

so the 3200CL14 kit has 7% faster latency than the 3800CL18 kit, but the 3800 kit is 19% faster based on frequency.

so in this case, I'd say the 3800CL18 kit is better.

 

1 hour ago, camieabz said:

Crude way to calc it. With 3200 with C16, divide 3200 by 16 = 200. 3200 with C18 would be 177.8 (slower). So higher is better (in theory).

 

3200/14 = 228.6

3600/16 = 225.0

3800/18 = 211.1

 

However, this crude calc is best used for comparing RAM timings of RAM of the same or similar speeds. When the speeds are different, it's not so pertinent. There's also the real world fact that some people won't get their RAM to run at 3800, so the theoretical speed of it is useless. The simple answer is to choose the lowest Cas RAM for the fastest RAM speed you can actually run at. The other real world answer is to pick the fastest stable solution, so you don't spend the next 2-3 years chasing your timings.

What about the 3 other sub-latencies? I know that he had a way to calculate all 4 into it...