The AMD Ryzen All In One Thread /Overclocking/Memory Speeds & Timings/Tweaking/Cooling Part 2

I'm not telling myself anything. I don't use a 5900X on a X370 MB. If a 3950X runs fine on a X370 MB why won't a 5900X for that matter since they are both 105 watts and according to reviews, the total power output are very similar, between 3000 and 5000 series CPU's.
Besides I'm not the one saying the platform is obsolete you are. It's been proven over and over again that running 3200 MHz with tight timings and low CL numbers is just as effective on framerate as running 3800 MHz at least on the 3000 series CPU.
If that is still the case with 5000 series CPU's remains to be seen. Since a lot of features still isn't added to the AGESA - undervolting being one features that is missing and apparently will be significant compared to 3000 series,

 

Got a question for the guys with Ryzen 3xxx or 5xxx:
You guys have a limit on mem oc where bandwidth is split in half when passing a certain frequency (3733 or 4000 mhz).
Is this splitting done at a fixed total frequency of RAM or does it kick in at a certain multiplier / divider?

I ask because I thought of buying a new platform and doing bclk oc to pass these limits and still have full bandwidth.

Example:
100 mhz bclk = "limit" at 4000 mhz
110 mhz bclk + same RAM settings = 4400 mhz
Would it do the split then?

 

Kicks in if you change the divider applied to the infinity clock, 3000 series chips most do 1800MHz on IF meaning 3600 on memory at 1 for 1 divider, it's tough to run 3000 series chips at 1900MHz on IF clock is silicon lottery some will, most won't be stable or won't boot at all. It's said the 5000 series the 2000MHz IF is like trying obtain 1900MHz IF clock on 3000 series, going be down to quality of the silicon.

 

Personal experience is BCLK OC isn't worth it on Rysen. Others may disagree.
Initially on my ROG Crosshair VII HERO going above a certain BCLK meant you couldn't use a specific SATA port or the system wouldn't boot / become unstable (can't remember which SATA port)
However running with a 2700X in that board and using Auto OC in bios raised the BCLK to 102 MHz (or was it 101.2 can't remember) and it worked very well.
But currently PBO + undervolting is the way to go with 3000 series chips and by the sound of it undervolting will be big on 5000 series.
I have had no issues running 1900 MHz FCLK on a Ryzen 3700X, 3800X and a 3900X all on the same board. But I suspect it's related to the ram that was used.
G.Skill Trident Z RGB 3600 MHZ CL16 which runs 3800 MHz CL 16 at 1.35v (high quality Samsung B-die FTW) all though it also depends on the BIOS used (latest version doesn't run 1900 MHz FCLK)

 

Been using one for 3900x which based on reviews is a slightly hotter chip than both 5900x and 5950x so I dont see why it shouldn't be good for 5000 CPU's as well.

 

^^ Yup 3000 series don't like FSB OC at all
All i can do is 100.2 MHz, on 1700X i was able to push FSB @ 105MHz NP.
Some say that this is related to PCIe Gen.4, on Hero VI You'll need to change PCIe from Auto to Gen.3 or even Gen.2 to run FSB at higher rate.

Side Note:
Hero VI is very capable of runing Zen3 IMhO.
All You loose is PCIe Gen.4, rest will be plenty fast.

Zen, Zen+, Zen2 & now Zen3

 

I was able to get my 1600x over 4.0 thanks to FSB OC on my old X370 board but I could never get any memory to run past 2933.

 

On my current DDR4 i have 3400-3500 CL14,15 with Zen1 NP (B-die).
Now rocking 4x8GB at 3800 CL16 (~63ns) -> It's enough

 

G.SKILL Trident Z Royal RGB DDR4-4000MHz CL15 vs 4000 cl17 , the best speed and latency?
Cl15 less latency vs Cl 17?

 

Correct yes lowest Cl is always the best

 

To calulate timings in nanoseconds you would do : 2000 * timing / ddr_freq
Which gives ...:
CL17...: 2000 * 17 / 4000 = 8.5 ns
CL15...: 2000 * 15 / 4000 = 7.5 ns

So this means that ram running at 3000 MHz @ CL15 have the same latency as ram running 3200 MHz @ CL16

 

So is my assumption correct?

 

Same goes for Zen2

Note:
To put it simple, it is what it is cuz' of 2xDual Channel (|| + ||)*
Whole System bandwidth is just wider, |CPU<->PCIe<->RAM<->GPU|

* Im calling this that way.

 

DDR4-4000MHz CL15 vs 4000 cl17

To simplify (What is CL14 rated speed for it?):
4000/15 = 266*14 = 3733 (So this Kit is 3733MHz CL14)
4000/17 = 235*14 = 3294 (So this Kit is 3294MHz CL14) *

CL15 is just wayyy faster

* I have something like it, 3800MHz CL16 1.39v easy.

Note:
With CL15 You can easily do 3800MHz CL14-14-14-14 1T

 

The reason for increased performance is because by doubling the dimms, they are also doubling the memory ranks that the controller can address.
This means the memory controller can make more request for the same amount of time.
It doesn't mean it doubles memory performance, but it means that real world performance for the memory are closer to the theoretical numbers.

 

anyone got highter then 4.725ghz on a ryzen 5 5600x cant get it to go highter stable did get it to boot at 4,825 but it crash if did anything with it in secs thx for any replys i get

 

Do note that this is 2x8GB, it should have better FPS with 4x8GB.

 

Rank plays a role too does it not, 2x dual rank memory might even be better than 4x single rank in terms of potential but it's not quite so simple the CPU, motherboard and RAM qualities all factor in.
Seen a few videos and tests of this popping up lately as said on other forums it's interesting to see because this should have already been tested and checked much earlier like Zen or Zen+ for Intel.

Also it seems Zen3 here resolves the lower performance under certain instructions as a neat little thing.
https://github.com/dolphin-emu/dolphin/pull/9233

 

Yeah I've heard dual rank ram helps a bit (just have to remember that when I next get some!)