In the search for the "Ram rule of thumb"

[mnahit]

Throughout the hundreds of types/brands of RAMs within the market, as a rule of thumb, can we qualify them by dividing the ram hertz speed by the CAS number?

Eg, not for all the brands but the renowned ones, can we compare a 2666 MHz CL13 ram with a 3000 MHz CL16, calculating
2666/13 = 205
3000/16 = 188
and as 205 > 188,
concluding 2666/CL13 smokes 3000/16 ???

This is my rule of thumb for a few weeks..
If i'm not mistaking??

 

[Tradesman1]

Then too there's DDR4 3200/14 which equates to 4.375. You also need to factor in the higher data rates handle more DRAM at a time than lower data rates so in theose slower ms of latency the higher data rate DRAM is move more MT/s (mega transfers) per seconds

 

[Pinhedd]

Throughout the hundreds of types/brands of RAMs within the market, as a rule of thumb, can we qualify them by dividing the ram hertz speed by the CAS number?

Eg, not for all the brands but the renowned ones, can we compare a 2666 MHz CL13 ram with a 3000 MHz CL16, calculating
2666/13 = 205
3000/16 = 188
and as 205 > 188,
concluding 2666/CL13 smokes 3000/16 ???

This is my rule of thumb for a few weeks..
If i'm not mistaking??

Nope. Although there are thousands of amateurs out there that would like to think so, DRAM doesn't work that way. Column Latency and Data Rate are two different characteristics. For a more detailed explanation, please read my tutorial stickied at the top of this forum.

In general, higher data rates are better regardless of accompanying latency. However, dealing with high latency requires a more complex memory controller and accompanying microprocessor architecture. There's lots of scheduling and state tracking going on inside of a memory controller of which enthusiasts are generally completely ignorant. From a bird's eye view, peak performance is limited by the memory controller's internal architecture and capabilities, the number of ranks installed on the channel, and the data rate. The oft repeated "Data rate divided by Column Latency = relative performance" is nothing more than hogwash concocted by someone who was too proud to admit that they had absolutely no idea what they are talking about.

Memory performance and optimization is properly the subject of a textbook, not a forum post.

 

[Tradesman1]

Throughout the hundreds of types/brands of RAMs within the market, as a rule of thumb, can we qualify them by dividing the ram hertz speed by the CAS number?

Eg, not for all the brands but the renowned ones, can we compare a 2666 MHz CL13 ram with a 3000 MHz CL16, calculating
2666/13 = 205
3000/16 = 188
and as 205 > 188,
concluding 2666/CL13 smokes 3000/16 ???

This is my rule of thumb for a few weeks..
If i'm not mistaking??

Nope. Although there are thousands of amateurs out there that would like to think so, DRAM doesn't work that way. Column Latency and Data Rate are two different characteristics. For a more detailed explanation, please read my tutorial stickied at the top of this forum.

In general, higher data rates are better regardless of accompanying latency. However, dealing with high latency requires a more complex memory controller and accompanying microprocessor architecture. There's lots of scheduling and state tracking going on inside of a memory controller of which enthusiasts are generally completely ignorant. From a bird's eye view, peak performance is limited by the memory controller's internal architecture and capabilities, the number of ranks installed on the channel, and the data rate. The oft repeated "Data rate divided by Column Latency = relative performance" is nothing more than hogwash concocted by someone who was too proud to admit that they had absolutely no idea what they are talking about.

Memory performance and optimization is properly the subject of a textbook, not a forum post.

__________________

Some good points raised, however reading and reallity are two separate things. One can read all there is about the specs and theories of DRAM and how it, the mobo and the MC (memory controller) all interact, as well as what any given specs 'should' end up resulting in. According to the material (which is often presented as 'this is how it is', it just doesn't wash. One can take 10 identical sets of DRAM and run them on the same system and get a variety of different results from the sets. You can take a set of DRAM we'll say a 3200 set of DDR4 and run it with a CPU (say a 6500), on a given rig and it may run fine under XMP at full 3200. You could then change ONLY the CPU to a different 6500 and may not even be able to run the DRAM at spec under XMP, under manual settings, even manually with looser timings and higher voltages

That's the thing with DRAM, it's the least understood component in a rig, there are few things that are truly hard set because too many other things play in and effect how DRAM runs. So pretty much no matter what you see or read it becomes theoretical as it may apply in one instance, and not in the next few.

 

[mnahit]

Whoauwwwf. I was expecting some counter measures but not thought it would be so complicated! Anyway, what you two say actually coincides. One trying to explain by theory/rules and one by practical usage. Both are so valuable and increase my desire to bring a solid viewpoint. Thank you very much.

Now i will locate the stickied tutorial first (sorry for not making this earlier) and try not to get lost there, while requesting both of your valuable remarks..

 

[mnahit]

As i try to read the "hightly technical" text, i started to see how brave we are to select a ram/configuration. Actually i'm losing my long-lasting courage to play my ram-mobo opting games.. ;(

 

[Tradesman1]

The old CAs/data rate * 1000 can give you a rough idea when comparing individual sets/sticks but there are so many variables that can play in is the problem. I did an article on 2400/32GB sets a while back

http://www.tomshardware.com/reviews/dram-benchmark-fluctuations,4080.html

and in some places 2400/10 sticks were topped by 2400/11 though primarily the piece was aimed at the MCs of AMD vs Intel