Question RAM Speed/Timing/"Work" Calculator? Comparison?

[nasch007]

Greetings,

I've tried searching but didn't find much of a result, except people with sort of similar questions (no solid answers though).

I'm wondering if there is a calculator out there that can show how much "work" your RAM can do @ what speed and timings. I'm looking for a comparison in numbers, and not colloquial or anecdotal evidence. I've seen general arguments on both sides, and they mostly rely on logic or circular logic, so I'm curious to see if there is anything I can plug numbers into and see?

My hunch is that there's a point of diminishing returns, or rather an intersection and overtake... the thought process I had was:
You can move more cars on an 8 lane highway with a speed limit of 65mph than you can on a 4 lane highway with a speed limit of 70mph. But I figure, at some point, when the speed difference between the two is large enough, they will move the same amount of cars or the 4 lane highway may even be faster. This initial assumption may be wrong, but in any event, that's what led me here.

Specifically I'm looking to answer the following questions:

Which is "faster" or "more efficient" or can do more "work" in less time (or, compare how much work they do in the same amount of time, say 0.1, 1, or 10 seconds):
32GB (8x4GB) of DDR4 running in Quad Channel mode @ 2133MHz with CL14, or, the same sticks @ 2400MHz CL15?

Or, same question: 32GB (8x4GB) of DDR4 running in Quad Channel mode @ 2133MHz with CL14 vs the same sticks, speeds, and timing in Dual Channel mode.

Is there anywhere that will show us the numbers? The bits/bytes/MB per time calculation?

Thanks for your consideration,

 

[13thmonkey]

It depends on the application as to the change in performance, and performance is analogous to work in this instance.

https://www.tomshardware.com/reviews/best-ram-speed-x399-colfax,6164-6.html

There are similar articles for Ryzen.

 

[nasch007]

It depends on the application as to the change in performance, and performance is analogous to work in this instance.

https://www.tomshardware.com/reviews/best-ram-speed-x399-colfax,6164-6.html

There are similar articles for Ryzen.

I see. That is very informative and I see that 4k pretty much just consumes everything across the board lol. It gives me a general sense but sort of misses the question... I'm interested in the calculator, like how did they arrive at those numbers, what data was generated and how to make the graphs. I'll give it another day or two and if no one else bites mark yours as the best answer.

Thanks!

 

[13thmonkey]

I see. That is very informative and I see that 4k pretty much just consumes everything across the board lol. It gives me a general sense but sort of misses the question... I'm interested in the calculator, like how did they arrive at those numbers, what data was generated and how to make the graphs. I'll give it another day or two and if no one else bites mark yours as the best answer.

Thanks!

There's no formula (well there is but we'll come to that),

Those graphs were literally created by setting the ram speed, running a set of tests, getting a result. Change the ram speed, run tests, etc. etc. So what this shows is the 'real life' impact of speed vs latency. All apps will vary in how they use Ram, how sensitive they are to it etc. So whilst you can do calculation to determine how much 'bandwidth' ram has for a given frequency, it's not that useful to know the answer as apps will perform differently.

In theory the highest speed lowest latency is best, BUT that costs 2-3x something that is a bit slower in speed and latency, and only performs 2% better in some apps, 5% in another and 0% in everything else.

 

[nasch007]

There's no formula (well there is but we'll come to that),

Those graphs were literally created by setting the ram speed, running a set of tests, getting a result. Change the ram speed, run tests, etc. etc. So what this shows is the 'real life' impact of speed vs latency. All apps will vary in how they use Ram, how sensitive they are to it etc. So whilst you can do calculation to determine how much 'bandwidth' ram has for a given frequency, it's not that useful to know the answer as apps will perform differently.

In theory the highest speed lowest latency is best, BUT that costs 2-3x something that is a bit slower in speed and latency, and only performs 2% better in some apps, 5% in another and 0% in everything else.

I get that. I guess I'm just looking for real world specifics to back up my own thoughts 'in theory'.

See I have a X99 board and Xeon processor. The Xeon does quad channel ram but tops out at the 2400 MHz speed. I thought this is better than dual channel because hey more sticks more lanes... made sense using the freeway analogy.

Now I'm wondering if I didn't shoot myself in the foot by doing this though. I know the board will support an i7, so should I have gone dual channel ram if it maxes out at higher speeds... aka 2600, 2800 3000 or 3200 MHz?

I think I am just looking to see how much I potentially lost out on or whether I made the right choice since for the work I do and speeds I'm at... yeah. Anyway I agree, like you said high speed low timings (especially 8 sticks worth) can get rather expensive.. for my current setup I found the lowest timing 2400MHz I could.

Thank you for entertaining my question.

 

[Calvin7]

You have a X99 board and Xeon processor. Which model numbers are something you omitted.

The X99 chipset supports both Intel Core i7 Extreme and Intel Xeon E5-16xx v3 and E5-26xx v3 processors, which belong to the Haswell-E and Haswell-EP variants of the Haswell microarchitecture, respectively. All supported processors use the LGA 2011-v3 socket.

The X99 chipset was released in late August 2014, while the supported processors were released in late August 2014 (Haswell-E) and early September 2014 (Haswell-EP).[3][5] In May 2016, X99's processor support was extended to the Broadwell variants of the Intel Core i7 Extreme and Intel Xeon E5-16xx v4 and E5-26xx v4 processors, which belong to the Broadwell-E and Broadwell-EP variants of the Broadwell microarchitecture, respectively.

Usually the listed memory specifications on the memory you buy are adequate in determining memory clock speed and timings. Some CPUs are more responsive than others to higher speed memory when overclocking to speeds higher than rated specifications.

The Core i7 6800K, for example, is a year 2016 model processor selling for under $200 on eBay. It uses Broadwell Microarchitecture with the Broadwell-E Processor Core. There are many reviews about overclocking such a processor and examples of what memory to use. What processor you have, I don't know anything about, other than it is a Xeon.

 

[nasch007]

You have a X99 board and Xeon processor. Which model numbers are something you omitted.

The X99 chipset supports both Intel Core i7 Extreme and Intel Xeon E5-16xx v3 and E5-26xx v3 processors, which belong to the Haswell-E and Haswell-EP variants of the Haswell microarchitecture, respectively. All supported processors use the LGA 2011-v3 socket.

The X99 chipset was released in late August 2014, while the supported processors were released in late August 2014 (Haswell-E) and early September 2014 (Haswell-EP).[3][5] In May 2016, X99's processor support was extended to the Broadwell variants of the Intel Core i7 Extreme and Intel Xeon E5-16xx v4 and E5-26xx v4 processors, which belong to the Broadwell-E and Broadwell-EP variants of the Broadwell microarchitecture, respectively.

Usually the listed memory specifications on the memory you buy are adequate in determining memory clock speed and timings. Some CPUs are more responsive than others to higher speed memory when overclocking to speeds higher than rated specifications.

The Core i7 6800K, for example, is a year 2016 model processor selling for under $200 on eBay. It uses Broadwell Microarchitecture with the Broadwell-E Processor Core. There are many reviews about overclocking such a processor and examples of what memory to use. What processor you have, I don't know anything about, other than it is a Xeon.

I thought my full specs were listed somewhere, I thought I filled them out, my apologies.

Here is my build:
Case- Phanteks Enthoo Pro M SE
PS- Thermaltake Smart Pro RGB 750W Zero Fan
MOBO- MSI X99 Gaming Pro Carbon LGA 2011-3 (refurb)
RAM- Corsair Vengeance LPX 32GB (8x4GB) Quad-Channel DDR4 2400MHz C14
CPU- Intel Xeon E5-1620 v4 Broadwell-EP Quad Core CPU @ 3.5GHz
Cooler- Enermax Liqmax II 240 "Front/Pull" config
NMVe- MyDigitalSSD 240GB 80mm BPX Pro m.2 PCIE
GPU- MSI GeForce GTX 1070 Armor OC 8GB
SSDs- 1x240GB Patriot, 1x240GB HyperX Cloud 9, 1x120GB Kingston SSDNow! v300
Storage - WD Passport USB 3.0

Although the BIOS on the motherboard does support overclocking both with simple multiplier and more advanced BCLK settings, I didn't think the CPU did. Even with the liquid cooling, I'm not certain I should try overclocking. From what I understand the OS/CPU determines when it needs to kick in and the first core will bump to 3.9GHz, the second to 3.8, and the rest to 3.6. I think that's what I found when I researched it although it has been quite some time. In any event, thanks again for your dialog!

 

[nasch007]

I'm not sure why the guy asked for details about my machine like he needed them, and then never responded. LOL!

Anyway this was all in theory, and like I said I was trying to find a model or line or graph that fit... I think it all boils down to it being a linear equation and relationship. Quad Channel will always be precisely twice the width of Dual Channel. So if my thinking on that is correct, then one "pull" from a dual channel setup can only be 128-bits wide at best (2x64) whereas one pull from a quad channel setup will always be twice that (4x64) assuming all sticks are used evenly/memory is allocated across all sticks.

Again this is all theory and I do not claim any truth for real world application.

This is what I came up with based on the logic above... if I am off anywhere, correct me:

Seems legit, eh?

So basically I think I answered my own question in that dual channel would have to run at 2x the speed of the quad channel set to be "worth it"... I've seen DDR4 at very high speeds but I don't plan on running them like that.

Thank you everyone who entertained/corrected me on this. If no one disagrees then we can mark best answer or close it up.