[SOLVED] Relationship between CPU-supported RAM speed and Mobo-supported RAM speed ???

[brokeBuilder2019]

Hello, I'm planning a custom build (components listed below) and I don't quite understand the relationship between the CPU-supported (max) RAM speed specification and motherboard-supported (max) RAM speed.

Ultimately, this boils down to the following practical question - if I buy DDR4 xxxxMHz RAM, for a system with abc processor and xyz motherboard, will it be a waste because the CPU will restrict its speed ?

Also, I have read a bit about XMP, but that just confused me even more.

My planned system components:

• Core i9-9900k CPU (up to DDR4-2666 supported ???)
• ASUS ROG Strix Z390-E gaming motherboard (all the way up to DDR4 4266(O.C.))
• Corsair Vengeance LPX DDR4 RAM (either 2666MHz or 3200MHz depending on the answer here)

Given that the Z390-E mobo was designed specifically for high end processors like the core i9-9900K, why would there be such a huge disparity (2666 vs 4266) in max allowed RAM speed ? Why would these mobos bother supporting such high speeds if CPUs like the i9-9900k don't allow anything even remotely close to 4266 ?

Is there some detail between the lines that I'm not seeing or understanding ? Thanks in advance.

 

[Phaaze88]

2666 is the implied memory standard - or, I should say 'up to 2666'. You still have to set the speed yourself. Exception: Some lower end non-Z mobos have memory frequency locked to 2666 for coffee lake cpus.

Any speed over the memory standard(it doesn't mean a cpu isn't capable of running faster) is considered an overclock, and with overclocks, there are no guarantees.

Whether your 9900k, or someone else's can actually run at 4266 depends on 2 things:
-The motherboard(small impact)
-The sturdiness of your individual cpu's Internal Memory Controller(greatest impact)

What can result from one aiming for such a high OC(just examples):
-Cpu successfully runs at that frequency with, say, 1.45v to the IMC
-Cpu runs at that frequency, but needs a little more voltage than 1.45 for stability @ 1.50v(not a good idea for 24/7 use)
-Cpu is actually able to run it with a little less voltage @ 1.40v
-Cpu isn't stable at anything over 3800mhz
-Cpu isn't stable at anything over the implied memory standard. It's possible, but rare.

Why do motherboard vendors bother?
Because some people are dumb enough to buy those kits that cost an arm.
The applications that can take advantage of the extra bandwidth from such expensive memory are very niche. For the majority of applications, just get a cheaper 3000 or 3200mhz kit and read up on how to tweak it for more performance.

There's a point of diminishing returns though. The higher you go up in frequency, the greater the penalty in latency, which offsets the gains in frequency.
-If you look at frequency alone, faster is ALWAYS better... sadly, it's not that cut-and-dry, as there's also the CL(core latency) and primary timings to take into account.
-The lower the CL, the better
A 3200 CL 18(cheaper) kit would actually perform worse than a 3000 CL 15(more expensive) kit.
-The tighter the timings, the better, but I'd suggest not getting hung up on timings as the first 2 have the greatest impact on memory performance.
Ex: 2x 3200 CL 16 kits, but kit 1 has timings of 16-16-16-36, and kit 2 has looser timings of 16-18-18-36.

Also, running 4 sticks in dual channel motherboards is akin to overclocking; they may require more voltage for stability over just 2 sticks at the same speeds.



TL;DR: the memory standard is more of an 'implied' one.

 

[brokeBuilder2019]

Any speed over the memory standard(it doesn't mean a cpu isn't capable of running faster) is considered an overclock, and with overclocks, there are no guarantees.

Whether your 9900k, or someone else's can actually run at 4266 depends on 2 things:
-The motherboard(small impact)
-The sturdiness of your individual cpu's Internal Memory Controller(greatest impact)

What can result from one aiming for such a high OC(just examples):
-Cpu successfully runs at that frequency with, say, 1.45v to the IMC
-Cpu runs at that frequency, but needs a little more voltage than 1.45 for stability @ 1.50v(not a good idea for 24/7 use)
-Cpu is actually able to run it with a little less voltage @ 1.40v
-Cpu isn't stable at anything over 3800mhz
-Cpu isn't stable at anything over the implied memory standard. It's possible, but rare.
...
TL;DR: the memory standard is more of an 'implied' one.

Wow, thanks a bunch for such a detailed answer !

Ok, just to make sure I understood you right ... basically, the CPU's stated memory speed spec (i.e. 2666) is just some kind of guideline saying "We guarantee stable RAM operation up to this speed ... anything above this means you're on your own (all bets are off)". If I choose to OC my RAM, the CPU will not forcefully prohibit it ... it will try to run it at my desired speed, but the resulting actual speed may not be what I wanted (due to the possible reasons you so eloquently listed).

Similar to CPU overclocking, then - Intel tests their CPUs up to a certain clock frequency, and based on the performance of a particular chip, they will categorize it and state its "max" OC'able freq. Anything above and the user is on his own - no guarantees.

So the "implied" spec is the upper bound with regard to their guarantee of stable operation.

Right ? Thanks again !

EDIT - Follow up question - So, for my particular system build, the next logical step in the RAM purchase decision making process would be to look at benchmarking sites and see what RAM speeds people with my exact same CPU/mobo have been able to actually achieve and (also looking at other factors which you mentioned, like my app usage pattern) accordingly decide what speed RAM to buy ... right ?

 

[Phaaze88]

Wow, thanks a bunch for such a detailed answer !

Ok, just to make sure I understood you right ... basically, the CPU's stated memory speed spec (i.e. 2666) is just some kind of guideline saying "We guarantee stable RAM operation up to this speed ... anything above this means you're on your own (all bets are off)". If I choose to OC my RAM, the CPU will not forcefully prohibit it ...

Correct.

it will try to run it at my desired speed, but the resulting actual speed may not be what I wanted (due to the possible reasons you so eloquently listed).

Slight correction here - It will try to run at the speed you set, but whether it can maintain it without crashing/freezing/BSOD...

Similar to CPU overclocking, then - Intel tests their CPUs up to a certain clock frequency, and based on the performance of a particular chip, they will categorize it and state its "max" OC'able freq. Anything above and the user is on his own - no guarantees.

So the "implied" spec is the upper bound with regard to their guarantee of stable operation.

Right ? Thanks again !

Yep.

 

[Endre]

Wow, thanks a bunch for such a detailed answer !

Ok, just to make sure I understood you right ... basically, the CPU's stated memory speed spec (i.e. 2666) is just some kind of guideline saying "We guarantee stable RAM operation up to this speed ... anything above this means you're on your own (all bets are off)". If I choose to OC my RAM, the CPU will not forcefully prohibit it ... it will try to run it at my desired speed, but the resulting actual speed may not be what I wanted (due to the possible reasons you so eloquently listed).

Similar to CPU overclocking, then - Intel tests their CPUs up to a certain clock frequency, and based on the performance of a particular chip, they will categorize it and state its "max" OC'able freq. Anything above and the user is on his own - no guarantees.

So the "implied" spec is the upper bound with regard to their guarantee of stable operation.

Right ? Thanks again !

Pretty much true.
So, don’t go a lot higher than 2666MT/s with your RAM speed.
3200MT/s CL16 is pretty much the “sweetspot”.
Besides, RAM speed won’t translate into direct perceived performance, since your CPU has 3 levels of cache, L3 being 16MB.

 

[Phaaze88]

Intel builds still benefit from faster ram... just not to the same extent as Ryzen ones.
-Intel skylake, and it's 14nm++++++++++infinity refreshes, continue to scale up with faster memory - it's nothing dramatic, but a plus is a plus
-Ryzen is a limited range from 3000 - 3600mhz. Going over that actually reduces performance.

Ryzen cpus have a fair bit of their performance 'gated' behind ram speed; Intel's don't.
Because of that, consumers can actually get away with a slower, cheaper kit on an Intel build. Do that on a Ryzen build, and it'll bite them pretty early in.

 

[brokeBuilder2019]

Pretty much true.
So, don’t go a lot higher than 2666MT/s with your RAM speed.
3200MT/s CL16 is pretty much the “sweetspot”.
Besides, RAM speed won’t translate into direct perceived performance, since your CPU has 3 levels of cache, L3 being 16MB.

^ Thank you guys for the great info ! I now have a much better idea of how to make my decision. You also gave me good pointers for further research (latency/timings).

Cheers !