[SOLVED] 3600 cas 16 first word 8.889 or is something 3200 better?

testinglife

Distinguished
Dec 2, 2008
46
0
18,530
I am thinking about getting this ram but I have read that sometimes 3200s are actually faster than 3600. I have no clue how to calculate what is faster. Can someone make recommendations or send me a calculator to use ?

This is what I am looking at G.SKILL Ripjaws V Series 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Intel XMP 2.0 Desktop Memory Model F4-3600C16D-32GVKC

https://www.newegg.ca/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232907?Item=N82E16820232907&nm_mc=AFC-RAN-CAN&cm_mmc=AFC-RAN-CAN&utm_medium=affiliates&utm_source=afc-PCPartPicker&AFFID=2558510&AFFNAME=PCPartPicker&ACRID=1&ASID=https%3a%2f%2fca.pcpartpicker.com%2f&ranMID=44589&ranEAID=2558510&ranSiteID=8BacdVP0GFs-Qt4Mg2XhbbPhAfUENnoMqA

In canadian its 189 so use that as a pricepi
 
Solution
that's just a fraction at the end - true latency is expressed in nanosecs so a lot more values are in the calculation.
but the values cancel out so you know the relative figures. 16 ticks on 3200 MHz clock takes the same amount of time as 18 on a 3600 clock ....
(bear in mind that's also only one value from the spec) 16-17-17-54 for example so it's always a simplification.

and CL is the numerator so a smaller value is better

Interesting. Why doesn't anyone just say that then. So all I do is take the mhz and divide by the cl rating ?? Other than the stability issue.

What does that first word latency thing mean then? For example this one is 8.89
for latency it's one thing 3600 / CL16 means 16 clock ticks. simple fractions can allow you to compare true RAM latencies e.g. 3200/CL16 and 3600/CL18 have the same true latency - without have to go deep into the math. DDR4 3200 / CL14 for example has lower true latency e.g. 14 ticks / 3200 < 16 / 3600 but other factors need to be considered. you can tweak clocks etc but it means nothing if you're system doesn't run stable - some setups won't go past 2933Mhz on the older AM4 boards - so it's never an exact science.

I am thinking about getting this ram but I have read that sometimes 3200s are actually faster than 3600. I have no clue how to calculate what is faster. Can someone make recommendations or send me a calculator to use ?

This is what I am looking at G.SKILL Ripjaws V Series 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Intel XMP 2.0 Desktop Memory Model F4-3600C16D-32GVKC

https://www.newegg.ca/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232907?Item=N82E16820232907&nm_mc=AFC-RAN-CAN&cm_mmc=AFC-RAN-CAN&utm_medium=affiliates&utm_source=afc-PCPartPicker&AFFID=2558510&AFFNAME=PCPartPicker&ACRID=1&ASID=https%3a%2f%2fca.pcpartpicker.com%2f&ranMID=44589&ranEAID=2558510&ranSiteID=8BacdVP0GFs-Qt4Mg2XhbbPhAfUENnoMqA

In canadian its 189 so use that as a pricepi
 
Last edited:

testinglife

Distinguished
Dec 2, 2008
46
0
18,530
for latency it's one thing 3600 / CL16 means 16 clock ticks. simple fractions can allow you to compare true RAM latencies e.g. 3200/CL16 and 3600/CL18 have the same true latency - without have to go deep into the math. DDR4 3200 / CL14 for example has lower true latency e.g. 14 ticks / 3200 < 16 / 3600 but other factors need to be considered. you can tweak clocks etc but it means nothing if you're system doesn't run stable - some setups won't go past 2933Mhz on the older AM4 boards - so it's never an exact science.
Interesting. Why doesn't anyone just say that then. So all I do is take the mhz and divide by the cl rating ?? Other than the stability issue.

What does that first word latency thing mean then? For example this one is 8.89
 
that's just a fraction at the end - true latency is expressed in nanosecs so a lot more values are in the calculation.
but the values cancel out so you know the relative figures. 16 ticks on 3200 MHz clock takes the same amount of time as 18 on a 3600 clock ....
(bear in mind that's also only one value from the spec) 16-17-17-54 for example so it's always a simplification.

and CL is the numerator so a smaller value is better

Interesting. Why doesn't anyone just say that then. So all I do is take the mhz and divide by the cl rating ?? Other than the stability issue.

What does that first word latency thing mean then? For example this one is 8.89
 
Solution
for latency it's one thing 3600 / CL16 means 16 clock ticks. simple fractions can allow you to compare true RAM latencies e.g. 3200/CL16 and 3600/CL18 have the same true latency - without have to go deep into the math. DDR4 3200 / CL14 for example has lower true latency e.g. 14 ticks / 3200 < 16 / 3600 but other factors need to be considered. you can tweak clocks etc but it means nothing if you're system doesn't run stable - some setups won't go past 2933Mhz on the older AM4 boards - so it's never an exact science.

However, 'true latency' does not equate to max memory bandwidth. This is like comparing apples to salmon steaks.

Not very in depth but a good read, nonetheless - https://www.crucial.com/articles/about-memory/difference-between-speed-and-latency
You should ALWAYS go for faster MHz RAM first. Latency is a secondary thought.
 
that's patently wrong, you can probably run the "lower clocked" RAM at the higher rate to accomplish the same thing - of course neither clock nor timing is guaranteed but you're not taxing the RAM at the higher rate assuming it runs at the higher clock rate. neither are you guaranteed that higher clocked RAM will deliver as most 2nd gen didn't run any higher 2933Mhz except with b-die.

However, 'true latency' does not equate to max memory bandwidth. This is like comparing apples to salmon steaks.

Not very in depth but a good read, nonetheless - https://www.crucial.com/articles/about-memory/difference-between-speed-and-latency
You should ALWAYS go for faster MHz RAM first. Latency is a secondary thought.
 
that's patently wrong...
What part of my statement is patently wrong? If you run the "lower clocked" RAM at the higher rate (assuming you mean clock it up to the higher clocked RAM), you are increasing the Hz speed.

My statement, in simple terms, is that latency and bandwidth are very different things AND that, for overall performance, you should ALWAYS go for faster RAM over lower latency RAM.
 
Last edited:
I know of no case where someone knowledgeable would prefer the G.Skill DDR4 3600 CL19 over say their FlareX DDR4 3200 CL 14 given the choice and the prices reflect that.
What part of my statement is patently wrong?

example https://www.tomshardware.com/reviews/corsair-ddr4-3200-vengeance-rgb-pro-memory,5681.html, you'll note very often (not always) the RAM with lower true latency pushes better to higher stable clocks... in this the "recommended" Corsair RAM can't be pushed as far as two other RAM models
 
Last edited:
I know of no case where someone knowledgeable would prefer the G.Skill DDR4 3600 CL19 over say their FlareX DDR4 3200 CL 14 given the choice and the prices reflect that.

Oh, I wasn't referring to purchase preference. I was just bringing in some engineering.

I will admit the following -
If you run a particular task (let's say your favorite game), exclusively, and that game will NEVER, EVER bottleneck your memory bandwidth, no matter what you do in that game (mods, settings, resolution, everything, etc.), AND you've performed tests and found that with your 1st/2nd gen Ryzen you get better performance with the slower Hz memory and tighter timings then fine, get the slower MHz speed memory and enjoy your game.




The most important part of the above statement is "game will NEVER, EVER bottleneck your memory bandwidth." Replace 'game' with anything that your memory will ever do. That's why it's soooo important to think about bandwidth over latency. Every game is different and every program is different (including all the little things Windows is doing in the background). If any function ever gets up against your memory bandwidth limits then that's it, the CPU will be 'drumming its proverbial fingers on the table' while the RAM says 'Hold up'.

I hope I explained this well enough. I do understand your point but the benefits of 'true latency' can definitely be hit-and-miss whereas memory bandwidth limitation is a hard stop. Also note that much of this discussion, with respect to gaming, is academically limited to game benchmarking as performance differences in gaming is very, very small and well within the margin of error in most/all of the graphs I've seem.