# Calculating memory speed CAS vs MHz.

#### don256us

##### Uber Folding Senior
I've been in the "industry" for more years than I will admit here. Recently I bought a new laptop and decided that I wanted to max out the RAM while prices are cheap. I did want to stay within a reasonable price range which started to show some interesting choices. Do I stay with the CAS 11 1600 MHz or get the CAS 9 or go up to 1866 CAS 11? I found this formula for checking which combination of speed and latency is best. While my machine is DDR3L, this should apply to all RAM.

CAS/MHz=speed?

E.G.:
CAS 11/1600MHz = .0069
CAS 9/1600MHz = .0056
CAS 11/1866MHz = .0059

In this example, RAM speed of 1600MHz and CAS 9 is the fastest of the bunch and may be cheaper than the 1866 MHz CAS 11.

What do you think? Is this even true?

Last edited:
Yes, it does have a relationship. Lower is better

yea, i cant find the page any more but it showed cas timings by speed. giving i think it was latency on paper of each speed/timings and where the "middle" was. being if you go with this speed get X timing to equal this lower speed/timings based on paper calculations. just as an example, there was many questions back in DDR2, if DDR2-1066cas5 was better/faster then DDR2-800cas4. the DDR2-1066cas 5 on paper was slightly higher latency vs DDR2-800cas4, so its really mixed on speed/timings which to go with. the bigger factor though is how much Y speed to X timings is going to cost you. if you can get a high speed with the lowest possible timings for a good price jump on it. if how ever you going to have to speed say another \$50 to get better timings even by say 1cas difference, is it really worth it if you dont notice it? speaking of say DDR4, look at say 3000mhz at cas 16/15/14. using newegg you can get 2*8gb DDR4-3000 cas14@\$89, cas15@\$64, cas16@\$69. all those numbers pulled from newegg using DDR4-3000, 2x8gig, choose your cas. you see a \$20 jump for one lower cas going from 15 to 14, in that case it might be worth it if you can spend that extra 20. its also interesting to note that on newegg cas 16 is slightly more then cas 15 but they are different brand sticks in the list. though as you increase in ram speed looking for lowest timings it starts to cost more.

the price gap does widen by a large margin, just looking i found Cas15-DDR4-3600 sticks,2x8gig for \$189, thats nuts to me!

anyway, with every generation of ram we have the "middle" ground, where you get the best performance for your system. being in ddr, what i found was 400-cas2 the best and DDR2-800cas4, DDR3-1600cas8. though i degress since there are articles out there showing newer gen i's using ddr3 do better with higher speed ram. though i contend that to a point. for X cpu speed you need Y ram speed/timings, if you dont go over X cpu speed getting the higher ram speed is a mute point(this is ignoring cost). as it has been pointed out to me in the past, you can get faster ram speed for the same price as the lower speed/tighting timing ram, so if paper math shows the same latency then go for it.

i feel like i kinda lost my point, so here is hoping someone understands what i am getting at.

I've been in the "industry" for more years than I will admit here. Recently I bought a new laptop and decided that I wanted to max out the RAM while prices are cheap. I did want to stay within a reasonable price range which started to show some interesting choices. Do I stay with the CAS 11 1600 MHz or get the CAS 9 or go up to 1866 CAS 11? I found this formula for checking which combination of speed and latency is best. While my machine is DDR3L, this should apply to all RAM.

CAS/MHz=speed?

E.G.:
CAS 11/1600MHz = .0069
CAS 9/1600MHz = .0056
CAS 11/1866MHz = .0059

In this example, RAM speed of 1600MHz and CAS 9 is the fastest of the bunch and may be cheaper than the 1866 MHz CAS 11.

What do you think? Is this even true?

Yes, CAS/MHZ X 1000 = access time in ns.

So your CAS 9 @ 1600Mhz is the fastest out of the box.

Theres ins't really any special DDR3 kits on the market anymore so I'm told, so your best bet is to either grab some regular 1600Mhz CAS9 stuff or hunt on classifieds and such for the better older DDR3 kits.

It's not really true as you count access time based on 1 timing while memory has a lot more settings. Especially in DDR4 higher clock is almost always better. Think that effective delay can be reduced by fast cache in new CPUs and all of that is related. You can count theoretical speed of memory but it's far from real performance.
For example how can you tell which memory is faster comparing single rank and dual rank modules based only on access time ? Access time will be about the same but dual rank modules will perform up to 20% better.
Other thing is that higher frequency = lower latency the same as lower timings = lower latency. At the same time higher clock affects both max bandwidth and lower latency while low timings affect mainly access time.

http://www.crucial.com/usa/en/memory-performance-speed-latency

^That what I noticed yesterday, doing some extensive testing on the DDR4 in sig.

It seems that frequency plays a bigger role in bandwidth/access time than latency with DDR4. More than in DDR3.

Wow. This is an informative thread for me. Thanks guys. For my application, the machine is a laptop so my choices are much more limited. Ultimately, I think that I will stick with the 1600 CAS 11. I'm trying to score another 8 GB to match my current 8 for around \$20 which will fit my need very well.

What do you think? Is this even true?

Yeah. It's short for 1/frequency * CL. But frequency has more uses than simply being a factor in how many nanoseconds you're gonna get. Otherwise 800 MHz CL4 would be just as good as 3200 MHz CL16, which I'm sure might be the case in a lot of applications, but I would expect some serious differences. (Within the same class, differences are usually likely to be greater in price than in performance.)

Also the problem right now is that most applications aren't taking advantage of the bandwidth you can get, and they won't for a while. But once they learn how to (more 4K? more VR?), all the flat lines on charts can get quite curvy. But it feels quite irrational to prepare for that eventuality years ahead by paying a significant premium for an advantage that physically exists but is most likely going to remain useless for several years. Meanwhile there are going to be new generations of CPUs with faster mem controllers and probably a bunch of new chipsets, all leading to much higher frequencies and lower CLs. In the end, unless some sort of a revolution happens (e.g. vertically stacked RAM & move from parallel to serial making production costs complicated and shaking the market up and down), affordable RAM purchased then will most likely be faster than expensive RAM purchased now.

Last edited:
Well. I tried to get my upgrade RAM from eBay but they are selling for over \$30. USED! I could have bought a new one from Newegg for around \$28. Instead I got a G.Skill kit of 2x8GB 1600 CAS9 for \$60. I'll sell my old stick for \$30 on eBay and will net a few dollars loss by upgrading my RAM by a factor of CAS2 and get some heat spreaders to boot. I learned a lot on this purchase.

It's not really true as you count access time based on 1 timing while memory has a lot more settings. Especially in DDR4 higher clock is almost always better. Think that effective delay can be reduced by fast cache in new CPUs and all of that is related. You can count theoretical speed of memory but it's far from real performance.

According to Aida64 when i overclocked my RAM 3200 14-14-14-34 2T -> 3600 15-15-15-30 1T i went from ~45500 MB/s to ~51500 MB/s so yay i suppose ?

Last edited:
According to Aida64 when i overclocked my RAM 3200 14-14-14-34 2T -> 3600 15-15-15-30 1T i went from ~47500 MB/s to ~51500 MB/s so yay i suppose ?

OK so .0044 to .0042 respectively thus showing a slight flaw in this calculation as it relates to actual performance. To me, this is saying that we are in the ballpark with this calculation but other factors may be at play when making a final selection.

I can test again at stock/overclocked and post if it helps, but has to be after im finished here (WoW HFC Mythic raiding)

I don't have a need for proof to that level. I had never heard of this before and wanted to know more about it. It's not an end all be all equation but it gets you close. That's all I needed.

To me, this is saying that we are in the ballpark with this calculation but other factors may be at play when making a final selection.
absolutely correct.

What's a good benchmark tool for RAM? I'll run a bench on my current 8 GB, then the new 8GB and then finally the new 16 GB kit together.

While not exhaustive, we will be able to see the difference between CAS 11 and CAS 9 and then we will be able to see the difference between single and dual channel.

AIDA64 has a good memory benchmark

What's a good benchmark tool for RAM? I'll run a bench on my current 8 GB, then the new 8GB and then finally the new 16 GB kit together.

While not exhaustive, we will be able to see the difference between CAS 11 and CAS 9 and then we will be able to see the difference between single and dual channel.
In a benchmark... real world results won't be much at all different.

In a benchmark... real world results won't be much at all different.

Oh no. You can't stop me that easily. I'm doing the tests.

compare it in HyperPi 32M

Replies
3
Views
844
Replies
3
Views
686
Replies
16
Views
6K
Replies
2
Views
434
Replies
4
Views
990