[SOLVED] Formula for comparing memory bandwidth?

box o rocks

Distinguished
Apr 9, 2012
750
10
18,985
Somewhere on this forum I remember seeing a poster present a short formula for comparing system RAM bandwidth (throughput?) based on the memory's clock speed and latency. The discussion as I remember was about comparing fast MHz RAM clock speed w/high latency -vs- slower MHZ RAM clock speed w/low latency to determine which would be faster at transferring data.
  1. can someone please show me this formula, if I'm presenting this question correctly?
  2. If so, does the formula work across DDR generations (ddr2, ddr3, ddr4...) or just within the same generation?
 
Solution
So I am playing with timings to see if I can bring the 2133 bandwidth equivalent to equal 3200 speed.
True latency is how you figure out how to get THERE.

So, let me share with you what was shared with me a long time ago, by my good friend Computronix, author of the Intel temperature guide.
Memory specifications and overclocking can be very deceiving. If you're unaware, here's the formula for "True Latency":

1 / Frequency (not DDR) x Latency = True Latency (nanoseconds).

Stock 3200 @ 14 is faster than Stock 3600 @ 16:

1 / 1.600GHz x 14 = 8.75nS
1 / 1.800GHz x 16 = 8.89nS

Stable Overclock with 3733 @ 16 is faster:

3733 @ 16 is 1 / 1.867GHz x 16 = 8.57nS

By calculating True Latency, when experimenting with memory...

box o rocks

Distinguished
Apr 9, 2012
750
10
18,985
Yeah, I'm aware that latency and true latency differ. Crucial calls it a latency paradox. So, I start out confused from the beginning...
Whichever one is the one shown in BIOS and in apps like CPU-Z, etc is what I would need to use to compare my RAM sticks. So... that one.
 

box o rocks

Distinguished
Apr 9, 2012
750
10
18,985
I didn't want to start out with a long story that no one would reply to, so I kinda left things unexplained...
I have this business motherboard (Asus Prime Q270M-C) and an i7-6700. The board has XMP and allows for faster RAM clock speeds. But maybe due to the Kaby Lake CPU or something else, I cant use faster RAM at its default speed. Everything either defaults to 2133 or fails to boot. I have some DDR4 3200. The XMP option shows up in BIOS, but doesn't seem to work. So I am playing with timings to see if I can bring the 2133 bandwidth equivalent to equal 3200 speed. That's why I wanted the formula. So I could see if I'm getting there.
Hope I haven't confused you too badly.
 
So I am playing with timings to see if I can bring the 2133 bandwidth equivalent to equal 3200 speed.
True latency is how you figure out how to get THERE.

So, let me share with you what was shared with me a long time ago, by my good friend Computronix, author of the Intel temperature guide.
Memory specifications and overclocking can be very deceiving. If you're unaware, here's the formula for "True Latency":

1 / Frequency (not DDR) x Latency = True Latency (nanoseconds).

Stock 3200 @ 14 is faster than Stock 3600 @ 16:

1 / 1.600GHz x 14 = 8.75nS
1 / 1.800GHz x 16 = 8.89nS

Stable Overclock with 3733 @ 16 is faster:

3733 @ 16 is 1 / 1.867GHz x 16 = 8.57nS

By calculating True Latency, when experimenting with memory overclocking, it becomes much easier to determine with greater consistency, where you might expect to find that elusive fuzzy-grey edge of stability.

Accept for applying excessive or prohibitive voltages above 1.35, which can potentially damage your processor's IMC, even with the best chips, any Frequency / Timing combinations that result in True Latencies below about 8.5'ish to 8.4'ish may be unstable, or unbootable.

The maximum memory speed supported by that motherboard chipset is 2400mhz, no matter WHAT CPU is installed (There is NO CPU you can install in that board that will support memory speeds above 2400mhz), so it shouldn't really take much to get past the default JEDEC true latency at 2400mhz by tightening the timings at 2133mhz, JUST BE SURE TO DO the extensive testing required to validate that the configuration is stable before moving on. The "XMP support" you see listed simply means that with the right processor it can run at 2400mhz AND regardless of processor it will use the XMP profile settings for primary, secondary and tertiary timings, as well as whichever clock speed is supported and the voltage settings. It does not mean that it supports higher than 2400mhz operations.

You can find the information on doing that testing towards the bottom of my memory guide located here:


What is the actual model number of the memory kit you are working with?

And no, I'm very much NOT confused.
 
Solution

box o rocks

Distinguished
Apr 9, 2012
750
10
18,985
Thank you for all your help and staying with me on this. I have gotten the (4x4GB) DDR4 2133 MHz RAM's timings down from their default of 15-18-36 to 10-12-30. No voltage increase needed. Seems stable running Prime95 for several hours, TechPowerUp MemTest, Intel Burn Test, and running several game bench marks.
I do see an numerical improvement in the Passmark Memory test. But I don't know how to relate to the score. It increased from 2988 to 3177. (?)
Shadow of the Tomb Raider demo for instance only shows a performance jump from 88fps to 91fps on highest settings (RX 5600 XT).
I wish I could try my 3200MHz at default timings for comparison, but that won't be possible on this platform. Thanks again.

Memory is:
2x4 Samsung M378A5143EB1-CPB
2x4 Kingston 9905678-024.A00G
Yeah, I know... mixed RAM. :giggle: But at least they are matching per set.
 
Yep. That mixed bag could be a problem, not only with incompatibilities that make it hard to actually get them tuned down to where they COULD be, but also because they are different don't be surprised if you see some random errors or weird glitches at some point. If you do, I suspect you'll know why that is.

Typically it's not as much of an issue at 2133mhz as it is at higher frequencies, so you might get a little extra leeway there.

As far as those performance gains are concerned, those are actually outstanding considering we're not talking about overclocking the memory, just tightening the timings. With a Z board, even with a locked CPU, you'd be able to use that memory at somewhere between 2933 and 3200mhz. Most systems that I've worked on in the Skylake generation, on a Z board, with a non-K SKU CPU model, have been able to do 3000mhz. On a really good board, possibly higher, so if you can find a good Z170 or Z270 board for the right price, depending on how important it is to you, might be worth doing a board swap for.