Sign in with
Sign up | Sign in
Your question

Latency vs Frequency

Last response: in Memory
Share
October 31, 2010 10:55:13 PM

Greetings,

Another "age old question": Latency factors into RAM modules in a relationship that is promotional to frequency, what exactly is that relationship?

Let me illustrate:

CL (CAS latency, or Column Address Strobe latency) is rated/measured in clocks cycles, commonly referred to as ‘ticks.’

So a CAS value of 4 means 4 cycles. As RAM increase in performance, the higher the speed, the faster the cycle time. Thus cost effective lower memory speed coupled with a lower CAS value can provide tighter latencies than a RAM with higher memory speeds and “looser” timings.

Here is a chart that someone put together to illustrate the relationships:

http://www.thetechrepository.com/attachment.php?attachm...

Although, the information of this chart is outdated. I would love to see more information on this with perhaps an updated chart... Does anyone know of such a source that talks about the CAS/Frequency relationship and timings?

More about : latency frequency

November 1, 2010 12:05:05 AM



Do mean the link contained in that article that links to http://en.wikipedia.org/wiki/CAS_latency ?

I mean first, that chat is out-dated and offers no update beyond the one I shared. Also I am looking for something similar to the chart I shared which shows ranked relationship between various latencies of each Speed...

Thank you for the link though, I am pouring through subsequent links to see if I can find anything else.
m
0
l
Related resources
November 1, 2010 12:12:30 AM

After re-reading that post you linked to I also find it a bit misleading to viewers. The poster neglects to mention that various frequencies of RAM (1333, 1600, 2000) have different choices for latencies. Just because a stick of RAM runs at 8-8-9-24 1333mhz does not mean a second manufactures 7-7-7-13 1333mhz will be the same. My question relates to the comparison of (for example) 7-7-7-13 1333mhz vs 8-8-9-24 1600mhz... See the issue? Well by using the chart in my original post one could calculate the true timings of the RAM by relating the Ticks (CS) to the Frequency (MHz)...
m
0
l
a c 347 } Memory
November 1, 2010 12:22:09 AM

The Q&A is/was 10 hours ago. The image link doesn't necessary = reality. There are 1600 MHz RAM with 7/8 CAS and 1333 MHz with CAS 9.

I didn't want to verbatim paste that last post, but...

The "expense" or "quality" of RAM is determined by: 1. Type DDR3/DDR2/DDR/etc, 2. the IC used on the DIMM, 3. CAS uniformity {8-8-9-24 vs 8-8-8-24}, 4. Speed {Faster Rated Speeds}, 5. Heat sinks used, 6. Mfg BIN {i.e. Matched sets: Hand vs Machine BIn/Sorted}.

As, I tried to illuminate it's all about the IC the RAM uses. There is no chart that covers the different IC, there is no direct CAS per Speed correlation with different IC, and so to answer your question would "MASK" the truth of CAS vs Speed. In principle ONLY "As Speed increases the CAS {latency} increases {slower}: Example 1333 8-8-8-24 -> 1600 9-9-9-24."

BTW - there at least 12+ {dozen} different current IC that RAM Mfg/Assemblers use.
m
0
l
November 1, 2010 12:41:29 AM

jaquith said:
The Q&A is/was 10 hours ago. The image link doesn't necessary = reality. There are 1600 MHz RAM with 7/8 CAS and 1333 MHz with CAS 9.

I didn't want to verbatim paste that last post, but...

The "expense" or "quality" of RAM is determined by: 1. Type DDR3/DDR2/DDR/etc, 2. the IC used on the DIMM, 3. CAS uniformity {8-8-9-24 vs 8-8-8-24}, 4. Speed {Faster Rated Speeds}, 5. Heat sinks used, 6. Mfg BIN {i.e. Matched sets: Hand vs Machine BIn/Sorted}.

As, I tried to illuminate it's all about the IC the RAM uses. There is no chart that covers the different IC, there is no direct CAS per Speed correlation with different IC, and so to answer your question would "MASK" the truth of CAS vs Speed. In principle ONLY "As Speed increases the CAS {latency} increases {slower}: Example 1333 8-8-8-24 -> 1600 9-9-9-24."

BTW - there at least 12+ {dozen} different current IC that RAM Mfg/Assemblers use.


That was what I was afraid of... if it is just theoretical based on principle only then I will use the formulation in the chart above to calculate the theoretical timings of RAM following the principles of the CAS/MHZ set by the original creator.

This would help me make a decision on 1333 8-8-8-24 vs. 1600 9-9-9-24 timings.... I am less interested in the "quality" of ram, IC, CAS uniformity, matched sets and so on... I am more focused on CAS vs MHz as I am a firm believer that these two elements can help narrow down my choices on what (at least theoretically ) is the best RAM to aim for. THEN elements such as CAS uniformity, matched sets and so on can be considered.
m
0
l
a c 347 } Memory
November 1, 2010 1:00:23 AM

Then you'll be making a mistake. In addition, a BIGGER mistake unless your verify that you RAM is compatible with you MOBO {Certified or Tested}.

Comparison Module #1: Mushkin Redline 996805 6-8-6-24 1600MHz
Comparison Module #2: G.Skill Pi Series 7-8-7-24 1600MHz
Comparison Module #3: Mushkin 998687 8-9-8-24 1866MHz
Comparison Module #4: Corsair Dominator TR3X6G1600C8D 8-8-8-24 1600MHz
Comparison Module #5: Kingston HyperX T1 9-9-9-27 1600MHz
Comparison Module #6: Mushkin Blackline 998677B 7-7-7-20 1333MHz
Comparison Module #7: Patriot Viper II Sector 7 9-9-9-27 1800MHz
m
0
l
November 1, 2010 1:27:21 AM

jaquith said:
Then you'll be making a mistake. In addition, a BIGGER mistake unless your verify that you RAM is compatible with you MOBO {Certified or Tested}.

Comparison Module #1: Mushkin Redline 996805 6-8-6-24 1600MHz
Comparison Module #2: G.Skill Pi Series 7-8-7-24 1600MHz
Comparison Module #3: Mushkin 998687 8-9-8-24 1866MHz
Comparison Module #4: Corsair Dominator TR3X6G1600C8D 8-8-8-24 1600MHz
Comparison Module #5: Kingston HyperX T1 9-9-9-27 1600MHz
Comparison Module #6: Mushkin Blackline 998677B 7-7-7-20 1333MHz
Comparison Module #7: Patriot Viper II Sector 7 9-9-9-27 1800MHz
http://i1013.photobucket.com/albums/af254/Jaquith/DDR3__Bench.jpg


Seeing my Ram will be 5 to 10 times the cost of the MOBO, I will be selecting my RAM first, then finding a compatible MOBO to go with it.

Ty for the benchmark chart... it is a bit confusing as what were they tested on? The same board? different boards? What was the CPU? What amount was tested a single 1GB module? Also I notice it says "overclocked" does that mean each module was OC'd to 1866MHZ?

That is why benchmarks are not very useful, they only illustrate items in a very specific condition. By getting the bigger picture you can make better sense of said benchmarks.
m
0
l
a b } Memory
November 1, 2010 4:20:16 AM

Bandwidth than latency.
m
0
l
a c 347 } Memory
November 1, 2010 10:54:14 AM

highcommander said:
Seeing my Ram will be 5 to 10 times the cost of the MOBO

Even an SR-2 won't meet that 10X ratio with 48GB RAM.

highcommander said:
...what were they tested on? The same board? different boards? What was the CPU? What amount was tested a single 1GB module? Also I notice it says "overclocked" does that mean each module was OC'd to 1866MHZ?

Testbed is identical {i7 920 & MSI X58 Eclipse SLI} the OC used the same BCLK 200 w/Multiplier 10, but the CPU Multiplier was turned-down ≈ Stock speed with a slight QPI Voltage increase {Identical for ALL RAM}. The max Memory controller speed for the i7 920 ≈ 2190 MHz. In addition, I'm adding the stock speed test; I reused a graph from a memory OC post.
100% Stock CPU + Stock RAM per spec:


highcommander said:
That is why benchmarks are not very useful, they only illustrate items in a very specific condition. By getting the bigger picture you can make better sense of said benchmarks.

I completely and emphatically disagree with your statements; here you have a Corsair Dominator 8-8-8-24 1600MHz outperforming Mushkin Redline 6-8-6-24 1600MHz ; this illustrates (2) of my very clear points: 1. a higher CAS beating a lower CAS RAM, 2. CAS disparity -> concluding the IC as I've stated all along.
m
0
l
a b } Memory
November 1, 2010 1:22:40 PM

In the end, you won't won't be able to tell 1 iota of difference in memory speeds. You can see a difference in benchmarks, but not in actual use. DDR memory running at 1333mhz is effectively fast enough that all for the most severely demanding users it will never, ever be a bottleneck, at any latencies. When talking about making your PC perform, as far as the RAM goes, just buy decent name brand memory, and enough of it to do what you need to do.
m
0
l
a c 347 } Memory
November 1, 2010 6:51:05 PM

I agree in the 'Benchmarks' to determine the 'fastest' RAM, but not with the '1333 vs 1600 vs 2000' - each jump on bench equates to ~2-8 FPS. I don't feel like doing the Google'ing, but I know Tom's is one of many that did these similar tests.

Also, DDR3 over ~ 1600 requires the CPU and/or BCLK to be increased ~ 160~190 to be effective. 2000 MHz RAM will either not function or becomes somewhat unstable with a huge Memory Multiplier w/o the BCLK increase. Never always & never Never; Most.
m
0
l
November 7, 2010 11:09:29 PM

ty all for the advice so far. Still researching this in greater depth.
m
0
l
October 5, 2012 3:28:48 AM

....so are you still researching?
m
0
l
!