CL16 vs CL18 vs...

[frozensun]

Guys can you please teach me about RAM,which RAM is faster (from 2 kits below) and why,or which RAM would give me higher performance and approx how much in %?

Does lower CL means better performance,no matter of RAM speed frequency?

Can 3000MHz RAM give better performance compared to 3666 MHz and why?

I always thought only frequency matters,higher=faster=better performance.

I hope you get the point.

[Eigenvektor]

The right one (3600 MHz) is faster than the one on the left (3200 MHz).

 

The RAM with 3600 MHz technically has higher latency (18 vs 16 clock cycles). But the important part is that they are measured in clock cycles and a clock cycle at higher frequency is shorter so in reality they actually have the same absolute latency (measured in nanoseconds, 10 ns here).

 

So in this case the 3600 MHz one would be faster overall. Whether that is measurable in games is up for debate. In most cases the 3200 MHz one should be fast enough.

 

~edit: To give a bit more technical info: Higher frequency means the RAM can transfer data faster. The latency one the other hand means how long it takes before the RAM can start to transfer data after it has been requested by the CPU. So higher frequency = better and lower latency = better. How much each of these matters depends on the application(s) you're using. Are they doing lots of tiny transfers? Latency more important. Are they doing few large transfers? Frequency more important.

[frozensun]

1 minute ago, Eigenvektor said:

The right one (3600 MHz) is faster than the one on the left (3200 MHz).

 

The RAM with 3600 MHz technically higher latency (18 vs 16 clock cycles). But the important part is that they are measured in clock cycles and a clock cycle at higher frequency is shorter so in reality they actually have the same absolute latency (measured in nanoseconds).

 

So in this case the 3600 MHz one would be faster overall. Whether that is measurable in games is up for debate. In most cases the 3200 MHz one should be fast enough.

ooh okay but what if it would be the opposite, 3200MHz CL 18 vs 3600 MHz CL 16,which would give better performance?

[gal-m]

5 minutes ago, frozensun said:

which would give better performance?

I'm not going to get technical, but say this. 

In the real world, you will NEVER notice the performance difference between CL16 and 18. 

There, I said it.

 

Here @frozensun, watch this: 

 

[Eigenvektor]

1 minute ago, frozensun said:

ooh okay but what if it would be the opposite, 3200MHz CL 18 vs 3600 MHz CL 16,which would give better performance?

It depends. As I said in my edit, if you do lots of small transfers then latency becomes more important, because the time you have to wait for a transfer to start is going to happen more often. On the other hand if you have very few very large transfers then latency will matter less and frequency (bandwidth) will matter more. So it really depends on the characteristics of the software you're using. For games, I'd expect latency will "win" most of the time, but obviously at some point lower latency will not be enough to counter the effects of lower bandwidth.

 

But in most cases we're talking about a small percentage difference here.

[gal-m]

15 minutes ago, frozensun said:

Does lower CL means better performance,no matter of RAM speed frequency?

Can 3000MHz RAM give better performance compared to 3666 MHz and why?

Best performance is, in general: LOW CL (latency) and HIGH frequency. The prob is, you'll need a decent CPU and board to get there + RAM like that is EXPENSIVE.. really really expensive

[frozensun]

Hmmm so CL 18 is slower then Cl16....but 3600 MHz>3200MHz..

So I assume the performance is around the same.

But for ryzen sweet spot is 3200MHz,right?

It's hard to get Ryzen to work with 3600MHz and gain performance from it,or I am wrong?

[Eigenvektor]

1 minute ago, gal-m said:

Best performance is, in general: LOW CL (latency) and HIGH frequency. The prob is, you'll need a decent CPU and board to get there + RAM like that is EXPENSIVE.. really really expensive

Pretty much. But there's also a sweet spot in terms of performance per dollar. At some point going even higher frequency/lower latency is so much more expensive and the performance difference is so marginal that it is simply no longer worth it.

[gal-m]

3 minutes ago, frozensun said:

So I assume the performance is around the same.

 

Kind of, probably, yeah. Around the same. It can also depend on the platform though. AMD can favor higher frequency and maybe, possibly not mind higher latency.. but, in the real world, you wont notice it at all.

 

You planning to use your PC for gaming or more specialized tasks is the real question?

[gal-m]

6 minutes ago, Eigenvektor said:

At some point going even higher frequency/lower latency is so much more expensive and the performance difference is so marginal that it is simply no longer worth it.

Exactly.

 

Here's that, on a graph: 

[Eigenvektor]

1 minute ago, frozensun said:

Hmmm so CL 18 is slower then Cl16....but 3600 MHz>3200MHz..

So I assume the performance is around the same.

No, at least not in these general terms. 3200 MHz CL18 would have higher latency than 3200 MHz CL16. But 3200 MHz CL16 has the same absolute latency as 3600 MHz CL18. Both have a latency of 10 nanoseconds.

 

As I said above, they're measured in clock cycles so higher frequency "counteracts" the latency to some extend. A latency of 18 clock cycles at 3600 MHz takes the same amount of time as 16 clock cycles at 3200 MHz.

[Vishera]

7 minutes ago, frozensun said:

Hmmm so CL 18 is slower then Cl16....but 3600 MHz>3200MHz..

So I assume the performance is around the same.

 

3600MHz CL18 has similar performance to 3200MHz CL16:

 

3600/16 = 225

3466/16 = 216.625

3600/18 = 200

3200/16 = 200

3466/18 = 192.5555555555556

3000/16 = 187.5

3200/18 = 177.7777777777778

3000/18 = 166.6666666666667

[porina]

8 minutes ago, Vishera said:

3600MHz CL18 has similar performance to 3200MHz CL16:

If looking only at latency. There's two major factors, latency and bandwidth. So the 3600C18 has more bandwidth and similar absolute latency compared to 3200C16. If all else is equal, the 3600C18 would be the logical choice.

[frozensun]

32 minutes ago, porina said:

If looking only at latency. There's two major factors, latency and bandwidth. So the 3600C18 has more bandwidth and similar absolute latency compared to 3200C16. If all else is equal, the 3600C18 would be the logical choice.

but isn't the ryzen infinity fabric locked at 3200MHz memory?

[porina]

1 minute ago, frozensun said:

but isn't the ryzen infinity fabric locked at 3200MHz memory?

Not unless you do it yourself. At least on boards I've had, it will try to maintain 1:1 ratio with ram up to 3600 (1800) for best performance.

[frozensun]

Just now, porina said:

Not unless you do it yourself. At least on boards I've had, it will try to maintain 1:1 ratio with ram up to 3600 (1800) for best performance.

Oh okay,how much performance is worth if I'd put this 3600MHz kit instead of my 3200MHz currently now?

[gal-m]

14 minutes ago, frozensun said:

how much performance is worth if I'd put this 3600MHz kit instead of my 3200MHz currently now?

Nothing. You won't gain a single thing. It's not worth it. In other words: If you're getting e.g. 120 FPS in a game, you might get 121FPS after the upgrade.

Unless you like looking at benchmarks.

 

I'll edit this real quick: it might depend a bit on your system (for that we'd need more info, of course), but @frozensun my man... it's 10000000% not worth spending $$$ on ram for such a TINY performance gain. It's literally better to buy anything else with that money, like more videogames for example.

[frozensun]

thank u guys.

no, I already have this 3600 kit,thought to jkust swap.

[gal-m]

4 minutes ago, frozensun said:

no, I already have this 3600 kit,thought to jkust swap.

Ohhh, if you already have the kit, I'd prob swap it.

 

You running an AMD rig @frozensun?

 

Btw, greetings from Slovenia  

[porina]

1 hour ago, frozensun said:

I already have this 3600 kit,thought to jkust swap.

You could do your own testing of how much difference it makes in your actual use cases and settings. Other benches only give more of a general indication. Some things may be better or worse than average.