Which is the better RAM based on latency/timings?

[bengalih]

So I have two choices of RAM

a) G.SKILL - which is 2666 @ 15-15-15-35

and

b) PATRIOT which is 3000 @ 16-16-16-36

If my math is correct ((T = 1 /f) * CL), based on what I can gather the total latency for these is 11.25ns and 10.67ns respectively.
(Seemingly making the PATRIOT the better choice based on total latency)

The G.SKILL I have tested so far to 2933 @ 14 CL.
This would equate to a total latency of 9.54ns.

The PATRIOT I have tested so far to 3200 @ 18 CL.
This would equate to a total latency of 11.25ns.

I may be able to tweak that PATRIOT better, but it seems I would need to get it down to a CL15 (or a speed of 3800) to have a better latency than the G.SKILL.
I'm not sure that would be possible to get things that tight (though admittedly I haven't tested yet).

So here are my questions:

1) Based just on the math above - isn't the G.SKILL the better RAM in this situation? Based strictly on performance?

2) The G.SKILL gave me a lot more problems than the PATRIOT. The G.SKILL required changes in timing, voltage, and modification of some other settings (GDM) in order to boot above 2400 with my Ryzen 3. It seems stable now, but it took some work. The PATRIOT however worked fine at 3000 simply by selecting the XMP profile and seems to be working fine at 3200 with just a small increase in voltage.
Based on the fact that the PATRIOT seems more compatible out of the box, should that account for something? Is it worth a latency drop of almost 2ns?

I may be totally missing something here as this is all kind of new to me - so please feel free to school me

 

[caddi daddi]

what are you doing with the rig?

 

[bengalih]

Well as always I have grand designs

In reality it will be 75% Plex server and possible networks services backup (DC, DHCP, DNS).
It will also be where I let other encoding job runs (converting stuff for iPad, etc).

It is an HTPC and I always build one with the idea to do casual gaming - but quite honestly I never really get around to it.

Curious though as to when you would recommend one over the other based on usage (and why).

 

[Lochekey]

Why not download something like Aida 64 or Maxxmem and actually test one versus the other instead of calculating theoretical values. This would give you a better idea of what runs better in your given set up.

 

[bengalih]

What is the true latency of a module?

So let's say you have RAM rated as follows:

(a) 16-16-16-36. You would say the CL is 16, correct?

and if we had the following:

(b) 15-17-17-38. You would say the CL is 15, right?

My question is when using formula to calculate latency in ns:

T = 1 /f * CL

Is the CL value really all that is taken into the equation?

If that was the case then two modules with the same speed (say 3000), module (b) would calculate as lower latency:

(a) T = 1/1500Mhz * 16 = 10.67ns
(b) T = 1/1500Mhz * 15 = 10ns

However the tRCD, tRP, tRAS values on (b) are actually all looser than that on (a).

So on the basis of everything else being equal, how do we measure the CL of one module to another when the other values (tRCD, tRP, tRAS) are also unequal?

I realize that this is theoretical, and the true performance would need to be measured, but I would still like to know that answer.

 

[bengalih]

I'll give that a shot. I'm still learning about all this so don't/didn't even know what are best apps to test with. I've used AIDA64 for some other stuff in the past so I'll give that a whirl.

From a purely theoretical point though....if you were being offered only one of the choices with the info I posted in OP - and you couldn't swap after you chose - would I be correct in assuming the G.SKILL is the better option since the latency calculates less?

If so, I'm interested in experienced opinions as to whether the 1.5-2ns increase is worth the "peace of mind" that the other RAM seems to work without the need for tweaking.
I had such a strange problem with the G.SKILL when the timings were set to the XMP defaults (system was stable as long as Vega drivers weren't loaded) that I'm erring towards the PATRIOT ram.

I'll do some benchmarks on the two at the highest I've got them to run and see what the differences are.

 

[EarthDog]

Merged threads... this is essentially one subject bub.

Probably could have merged your 3rd thread too..

 

[caddi daddi]

you can broadly say that for gaming/surfing, you want ram with a low CL.
if you render video and things like that you want faster ram and CL is less important.

 

[Janus67]

Ultimately I'd say I'd get whatever is cheaper as you will likely never see a difference outside of memory-specific benchmarks

 

[caddi daddi]

could not have said it better, janus67.

 

[Woomack]

You shouldn't count performance or latency based on 2 variables. There are many timings which are affecting performance and other things like how memory controller acts. On Ryzen frequency gives more but at least looking at whole rig purpose, I would say it doesn't matter.
On the other hand if you can set 2933 CL14 on G.Skill then it should run at 3200-3600 CL16 or tighter (depends on the motherboard). Patriot probably too but I don't know what IC is there.

 

[Organik]

Why not download something like Aida 64 or Maxxmem and actually test one versus the other instead of calculating theoretical values. This would give you a better idea of what runs better in your given set up.

Yes and 9ms is not your latency my friend. More like 50's ns. Just check with Aida 64 or try a free app, but I don't know of any just Aida or Maxxmem. Aida however is the best diagnose app for hardcore people.

Go with the Patriot it will be fun. You can loosten timings to overclock it and tighten up timings and underclock it. If I were in your shoes I would underclock the Patriot to 3200Mhz and tighten those timings up like 14 or 15 CL