SPD CAS reporting by MemTest86

[byte99]

I understand that MemTest86 obtains the CAS timing by reading SPD chip details. Does this show the actual timing at which the chip is operating, or is it just reading info. from a lookup table?

For instance, suppose you have a DDR4 PC4-21300 2666 MHz chip, which is nominally CL19, but this particular model is listed by the vendor as CL15. Can you tell from the SPD chip details if it is actually running at CL15 in your machine?

Or consider these screenshots from CPU-Z (not my machine), which show that the actual timing (5-5-5-15) doesn't correspond to any of the available SPD timings:

[Richard (PassMark)]

MemTest86 doesn't report the RAM timing values currently set in the BIOS but only reports the timing capabilities stored in the DIMM's SPD module (i.e the values programmed into the RAM stick itself by the vendor when the stick was manufactured).

[byte99]

Thanks for your reply. On my current RAM, MemTest86 only reports one set of timing values: 19-19-19-43. Does this mean there is only that one set of CAS timings stored in the SPD module?

Conversely, if I were instead using the RAM shown in my original post, would MemTest86 report all three sets of timings shown in the Timings Table?

Or if it only reports one set regardless, how does it choose which one to display?

[David (PassMark)]

There isn't space on the screen to display all the SPD data while the test run is taking place. Normally there are multiple profiles in the SPD data. It displays the best looking one, which is normally what is used in the marketing material as well.

If you wish to look at all the SPD data, you can use RAMMon.
But MemTest86 can also display this from the main menu, when not running a test.

[byte99]

Unfortunately it appears RAMMon is Windows-only, and I use a Mac. But I did go into the config section of MemTest86 (that was fun to look through) and, for my DDR4 PC4-21300 2666 RAM, it said:

CAS latencies supported: 10 11 12 13 14 15 16 17 18 19 20
Timings @ Max Frequency (JDEC): 19-19-19-43

What MemTest86 displays when you run the program is just the 19-19-19-43, so it looks like it chooses the nominal timing, rather than the best-looking one (from a marketing perspective), since the latter would (I assume) have lower latencies (e.g., CL 15), no?

It sounds like if I want to pursue this further I should install Boot Camp so I can run CPU-Z (I tried it in Parallels, and it doesn't work, since in that case CPU-Z shows what the VM sees).

[David (PassMark)]

The lowest latency timings aren't the best if to get those low latencies you need to run at a very low clock speed. Need to look at both clock speeds and latency.

[byte99]

Well, I'm thinking of this screenshot I saw (from CPU-Z) of the SPD Timings Table for Kingston-branded CL15 DDR4 PC4-21300 2666 MHz RAM made by SK Hynix. According to the first screenshot, it's capable of running at 15-17-17-35 with same 2666 MHz clock as seen with the nominal 19-19-19-43 timing:



And I'm assuming (perhaps incorrectly) that this isn't special customized RAM, but rather that it's standard Hynix RAM that Kingston configures to run in the BIOS with CL 15 timing (which it does--see 2nd screenshot). If that's the case, then my SK Hynix DDR4 PC4-21300 2666 RAM should also be capable* of running 15-17-17-35 @2666, which is what you'd choose to show instead of 19-19-19-43@2666 if you wanted to select, from the available timing options, what looks better. Or do I have that wrong? [*I'm referring to the capability of the RAM; it might not actuallly run with those timings in a particular machine.]

So in summary (and, again, maybe I have this wrong), it looks like what MemTest86 is showing is not the best timing set available from the SPD table, but the nominal timing.

[David (PassMark)]

SK Hynix & Kingston produce a huge array of different memory sticks with different timings and clock speeds. I don't think you can assume that your PC4-21300 is the same as some other random PC4-21300 stick you found screen shots of on the internet.

Also there are normally two XMP profiles. CPU-Z might be not be displaying them all in that screen shot.

Logic in Memtest86 is that we pick the profile with the higher clock speed and there are several profiles with the max clock, we pick the second one.
But it makes zero difference to the actual memory test and we display the full set from the main menu. So it isn't that critical for MemTest86.