Is there a benefit for syncing CPU, FSB, NB, HT, MEM frequency?

[GWARslave119]

I remember reading something about this but can't remember where, nor have I been able to find again, about having all or some of those frequencies matching, can anyone explain about that? Or if it makes no difference?

 

[Woomack]

It all depends on the platform. Sometimes it's better, sometimes it doesn't matter at all. All modern platforms are based on PCIE bus +internal connections, which are usually multipliers of the PCIE, so it's hard to run it asynchronously. On older platforms, FSB or HT could have different ratios and then one or another could bottleneck each other or simply waste time on additional clock cycles to transfer data. It's a quite wide topic to cover it with a single post. Anyway, on new platforms, you don't have to worry about that. The only exception can be Ryzen 3000 where infinity fabric can't run past some frequency so it switches the divider to 1:2 with memory. As an effect when you pass ~3600-3800 memory clock then the internal bus will run at lower frequency limiting whole memory/IMC performance.

 

[mackerel]

On Intel ring cache and mesh cache CPUs (pretty much everything since at least Sandy Bridge), it seems higher is better for the NB/cache/uncore or whatever you want to call it. Specifically for ring bus, that typically runs at least as fast as the mem although I've not had insane fast mem to try on it but it does seem to hurt if it is low compared to ram. This is from observations while overclocking k CPUs, but now I say that, I wonder how lower clocked locked CPUs cope with that. Or they're just so slow it doesn't really matter...

 

[Woomack]

On SL-X/KL-X it's more like a higher NB/memory controller clock that affects maximum memory bandwidth, not something that makes it work asynchronously when you touch it. It's more like a separated clock that still uses PCIE and its multipliers. On non-HEDT series so Skylake, Kaby Lake, Coffee Lake, ... ring clock or anything else doesn't affect performance in any significant way. Adjusting these clocks on new platforms is more often used to improve stability after OC but won't cause a performance increase or drop.
Since Sandy Bridge, everything is 100MHz PCIe bus and its multipliers so it's not possible to run it asynchronously. Last problems with performance because of FSB settings that I remember were on motherboards with LGA 775.

 

[mackerel]

On non-HEDT series so Skylake, Kaby Lake, Coffee Lake, ... ring clock or anything else doesn't affect performance in any significant way. Adjusting these clocks on new platforms is more often used to improve stability after OC but won't cause a performance increase or drop.

You saying you never saw a difference in ram bandwidth or benchmark scores from adjusting the cache on these? Or should I be looking at the word "significant". For normal use, I guess it isn't, but it isn't zero difference either and can give that slight edge when overclocking. Kinda like ram performance for lighter workloads.

 

[GWARslave119]

Sorry maybe it'd help if yall knew what I had lol...nothing 'new' persay but still very much not out of date ;p



And yes it's underclocked atm, just seeing how well it runs with games and stuff, which so far is fine, helps keep temps down

 

[Woomack]

You saying you never saw a difference in ram bandwidth or benchmark scores from adjusting the cache on these? Or should I be looking at the word "significant". For normal use, I guess it isn't, but it isn't zero difference either and can give that slight edge when overclocking. Kinda like ram performance for lighter workloads.

On most chips, the difference is so low that it doesn't really matter. I think Haswell was still showing performance gain in some tests because of additional busses or PCIE OC but I don't remember exactly. Some of these chips (pretty much all new) have cache clock tied to the CPU clock so overclocking any other bus is not helping at all. Anyway, on HEDT chips it counts a lot in some benchmarks but on lower parts not really.
There is a confusion because of naming and more or fewer options on various motherboard brands.

@GWARslave119

On FX there is 200MHz bus and all have multipliers around that value. If you run more than 2 memory modules then you don't really want to overclock memory or HT because it often causes weird behavior. It's not related to any asynchronous work but more to this platform architecture. I would recommend you to keep memory at DDR4-1600, FSB at 200MHz and play with the CPU clock to achieve higher performance.
The CPU clock is not affecting anything else. It's just FSB and multiplier.

 

[GWARslave119]

yeah i dabbled with things here and there but usually thats all i touch is the cpu multiplier, along with the cpu voltage if needed. i tend to keep the HT and NB freq both at 2400, does that matter if it's that or lower?

Oh and would underclocking the memory freq with faster timings add any benefit? You can see it start to list them in them parentheses on the pic before it gets cut off

 

[mackerel]

Some of these chips (pretty much all new) have cache clock tied to the CPU clock so overclocking any other bus is not helping at all.

How new is new? At least on Skylake - Coffee Lake consumer CPUs the cache ratio is not synced to CPU core ratio other than through base bus. Different multipliers. Under ram heavy loads (Prime95, y-cruncher) typically I'd see cache can be set higher with overclocking but typically a few hundred MHz below max core. Light workloads like Cinebench might not show much difference at all, again, kinda like how ram speed makes little difference with it.

 

[Woomack]

@mackerel, Eh, voltages, not clock ... this is why I shouldn't reply to posts when I'm at work. Thanks for pointing out that I said something stupid ... yes, the cache is typically ~300MHz lower than the CPU clock but can't go above the CPU clock. Depends on the CPU it's 100-300MHz lower so you can OC cache as long as you OC CPU clock. It's still on the same PCIE bus frequency but with higher multiplier and this actually affects memory performance. I don't remember now but there is another bus which is not really affecting anything. Every motherboard vendor is calling it different and later I don't remember how it's called

@GWARslave119
HT/NB affects performance but mostly NB affects memory performance. It's not so big difference so better keep it lower for stability.

 

[GWARslave119]

@GWARslave119
HT/NB affects performance but mostly NB affects memory performance. It's not so big difference so better keep it lower for stability.

Oki. Follow-up question which I've never actually questioned before until now...with all this talk of memory performance and timings and speed and whatnot, what difference would it realistically make, not just what benchmarks show. I assume for gaming, the video card is more important and the first to be used. When I'm doing audio editing with FL Studio or Audacity, any advantages there? I don't have much slowdown during realtime mixing or during encoding/rendering really to begin with... and the loads when i'm rendering with FL Studio look like:



I feel like it might not be much benefit, but you never know i guess..

 

[Woomack]

I don't know this software but it looks like it can't use all cores of your CPU. 21%+ is like 1 core + maybe additional service in the background. Or it simply doesn't need a faster CPU. As I said, I'm not sure how this software works and if it's optimized for multi-threading.

 

[GWARslave119]

I would assume that's what those options are supposed to do ;p

 

[Woomack]

Looks like yes

 

[GWARslave119]

well something's not working then caue thats what i had it set at already ha...i'll look into it some more