I see.. so it’s still called a B when the density and fabrication node is different.

owell guess i got a not so ideal stick. nowonder it’s so cheap (32gb for rm390 used) i probably shouldnt complain since im able to keep it under 75ns in aida64 bench with 45gb/s read

This post has been edited by Bonchi: Aug 5 2020, 12:21 PM

Maybe ur messing things up in your mind with Samsung B-Die. Every Micron rev. B (also called Micron B-Die frequently) has the same density 16Gb(2GB as you noticed) and is done on the same node. Whats different is just if they slap 8 on a stick (single rank like urs) or 16 (dual rank) like crucial does for their 2x32GB kits.

Just try what I suggested with 3733MHz. Save a profile before doing and if it doesnt work you hardly wasted time...

micron b pretty much max out @ 3600 @ 18 22 22 (16gb kit) from what i tested b4. i'd expect slightly lower for 32gb kit but worth a shot @ slightly higher frequency.

This post has been edited by ah_khoo: Aug 5 2020, 02:42 PM

i thought the micron b-die is refering specifically to the D9TBH which is a 8gb density on a 20nm node.

My bios is a msi lite bios because it’s a non max version.. tends to not boot at all if the setting is not right which will require to use a cmos jumper.

Currently work from home so i need my pc to be reliable and up at all times. Will definitely push this kit when this is over although not too optimistic about it haha.. probably not going beyond 1.37v

Edit:
Well i had an MC today so i just tried to push a little bit since im bored Did a 3 hours memtest86 with 0 errors.





I have lowered dram voltage to 1.35, vddp to 0.92 and vddg to 1 at the moment..and will be doing memtest overnight. Are these settings ok?

This post has been edited by Bonchi: Aug 6 2020, 02:09 AM

pretty much what can be done already, but no shame at all for a pair of 32gb kit bought @ RM3xx. overachieved i'd say.

Did the overnight run and it has 0 errors. Seems like this is good to go .. i guess teamgroup really did stick to their claims of using the best binned chips.

probably can hit 50+gbps read on aida's benchmark if i try to lower the trfc but i heard that this could potentially damage data without showing errors in memtest.. Is that true?

Thanks for everyone's opinions and tips here

This post has been edited by Bonchi: Aug 6 2020, 09:43 AM

RYZEN 5 3600
Noctua NH-U12F
ASROCK B450M PRO 4
XPG SPECTRIX D41 RGB 16GB (8GBX2) 3200MHZ DDR4
SSD XPG SX8200 PRO 256GB NVME
GALAX GEFORCE RTX 2060 6GB OC
SEASONIC 620W S2II

Hai. This is my firsttime to OC. Already install with msi after burner. Can i know what setup can OC with this spec? Forget about these id coz this is my bro old id. Hehe

https://youtu.be/W872lQcy65I

Can't wait for this program to be out. By 1usmus.

After OC, really become fast? Anyone burned your CPU already?

Update: Final result, this kit doesn't like lower tRCD's and tRFC. Other than that it's pretty much perfect.
Now just gotta see if I can drop voltage a tiny bit without losing stability.

done.


This post has been edited by nrw: Oct 19 2020, 09:49 PM

is msi tomahawk b550 + ryzen 5600 + ddr4 adata ok?
or any recommended mobo? high end too pricy 2k crazy

cyberloner wrong thread (you should go hardware section) but ... yeah more than OK, if you don't need any special connectivity/featureset you could even be getting a cheaper board than the tomahawk.

edit: there's another thread with the same question just today: https://forum.lowyat.net/topic/5046814
can refer to chocobo.

This post has been edited by nrw: Oct 24 2020, 11:45 PM

What rams are these?

can see from ZenTimings (F4-3200C14-32GVK)

This post has been edited by nrw: Oct 25 2020, 02:57 PM

Alright, just got a 5900x & 5600x installed in my rigs.
Both are retail samples so no binned/selected review samples.
(special thanks to the two guys helping me to source these chips. much appreciated!)
Not so special thanks to AMD who put a 10% mark up on international pricing.


To be updated whenever I have done more testing.

Rig 1: 5900x | B550i Strix | 2x16GB B-Dies | RTX 3080
Rig 2: 5600x | X570 Unify | 2x16GB B-Dies | RTX 2080

CPU OC Rig 1 (CB20):
Ambient: 28°C, Custom Loop, Fans&Pump silent
BIOS Settings (other than those directly connected to RAM):
1) Pure Stock + [+200MHz Clock Override]
2) Pure Stock with adjusted Voltage & Multipliers
3) Stock + [PPT 300 | TDC 250 | EDC 250] + [Manual Scalar 1x] + [+200MHz Clock Override] + [Global C-State Control (Enabled)] + [L1 & L2 HW Prefetecher (Enabled)] + [(Power Supply Idle Control (Low Current Idle)] + [CPPC & CPPC Preferred Cores (Enabled)]

Score for MC is rounded 5 run average, Score for SC is single run. HWinfo64 was active during all runs for monitoring.

Above Setting	Voltage (V) actual	Temperature in °C	Clock in GhZ	Score MC (Score SC)
1	1.112	64	~4.17	8220 (620)
2	1.325	85	4.6	9158
2	1.325	85	4.7/4.6	9244
2	1.325	85	4.725/4.625	9293
3	1.275	77	~4.35	8704 (631)

The adjusted stock setting with unlocked PPT/TDC/EDC is my clear favorite. Not static so I am not losing any single core boosts and the stability is flawless and temp on the safe side even on heavier loads than CB20.
Do note the static overclocks were just stable in CB20.



Boosting:
AMD defined 4650MHz, 4825MHz, 4950MHz, 5050MHz in the 1.1.0.0 v2 AGESA. (5600x, 5800x, 5900x, 5950x).
On my Rig 1 it looks like this though:
5600x will boost to 4650MHz when running stock.
5900x will boost to 4950MHz when running stock.

So they are over delivering as these are boosts that are utilized under actual (light) loads and not in idle.
They have learned this lesson!

Once you set Max CPU Boost Clock Override to +200MHz you can see up to that being added to your chip. My 5900x maxed out at +125. My 5600x goes all the way up to 4850.
5900x:

Core	CCD	Max. Clock GHz (single)
0	1	5.075
1	1	4.975
2	1	4.875
3	1	5.075
4	1	4.875
5	1	4.950
6	2	4.675
7	2	4.775
8	2	4.775
9	2	4.675
10	2	4.625
11	2	4.650

5600x:

Core	CCD	Max. Clock GHz (single)
0	1	4.850
1	1	4.850
2	1	4.850
3	1	4.850
4	1	4.850
5	1	4.850

Much improvement on this side. These max clocks are like a dream coming from Zen2.
Binning scheme of CPU's same as Zen2. One great CCD (4.875GHz worst cores) one average CCD (4.625GHz worst core).
If you guys want to compare boosts of your chip: run THIS tool for 5 mins and check clocks during this time using HWInfo64.

CURVE OC
This feature basically allows you to add voltage to your weaker cores (while optionally taking off voltage of your better ones).
Taking the cores up there, you could add 15mV to core 1, 30mV to core 2, 30mV to core 4, and 20mV to core 5 in order to make them boost ever so slightly higher.
It's a super useful feature that I am surprised not to find much on just yet.

Temperature Stability / Spikes during Desktop activities:
You surely remember your Zen2 being really nervous on Idle or Low load states, very quickly boosting up to be responsive.
All to the expensive of temperature and especially noise on stock coolers. Power plans were able to fix it but not many were aware.

So have we seen any improvements here?
YES. On true idle its rock stable. The voltage doesn't spike.
My temperature sticks on 33 and never changes. Regardless of being in high performance or balanced power plan.

On low load it takes much higher loards to trigger boosts and those boosts don't last long.
While I can't comment for stock cooling (I use a beefy custom loop) I can still see that the window of fluctuation is just 5°K.
This was more like 15°K on Zen2.
Great job, AMD. Can't wait to see a more budget 5600.
This will be a crazy "hot" selling chip.

RAM/IF Teaser:
Be cautious when you see Aida64 Memory Benchmarks on Zen3 (not optimized yet)

For older versions like v6.25.540 results look like this:

For the new v6.30.550 results are like this:

I think this difference is due to the newer version not including cache in those numbers any longer.
Use THIS tool to check your timings.
For Zen3 there is only a working beta version that they did not put on the page yet.

** RAM OC is postponed due to the current AGESA on Rig 1 is not allowing for the FLCK to exceed 1900. Hurry up AMD & ASUS **

This post has been edited by nrw: Nov 16 2020, 08:51 PM

Gaming Section:

Gaming benchmarks will be on request.
If you have a game and wonder what my rig will do just drop me a detailed PM (game settings etc.) and if I have the game I can test.
That's because most benchmarks by good reviewers out there already are quite good.

SoTR:
1080p - 5900x 2 ccd's (stock with 3800c14 2x16GB) - 3080 2100MHz locked


1080p - 5900X 2ccd's (stock with 3800c14 2x16GB) - 3080 1950MHz locked


1080p - 5900X (5600x simulated) 1ccd (stock with 3800c14 2x16GB) - 3080 1950MHz locked


1440p - 5900X 2ccd's (stock with 3800c14 2x16GB) - 3080 1950Mhz locked


1440p - 5900X (5600x simulated) 1ccd (stock with 3800c14 2x16GB) - 3080 1950Mhz locked


To those saying RAM OC is a waste of time.
Have a look at what the benefit is like between a 3800 setting with adjusted sub timings and a 3200C14 one running XMP.
1080p - 5900x 2 ccd's (stock with 3200c14 (XMP) 2x16GB) - 3080 2100MHz locked

1x proper Ram OC on maxed out settings in 1080p:

+30% CPU Game Render (Min)

+14% CPU Game Render (Max)

+18% CPU Game Render (Avg)

+10% Raw FPS

CPU Bound: from 61% -> 50% CPU Bound

1080p the 5900x still has quite something of offer over the 5600x in SOTR.
For high graphic settings, 1440p however is where things change and the simulated 5600x (albeit on the quick CCD which clocks higher) is just as fast as the 5900x.
SOTTR? 1440p with 3080/3090 or equivalent? 5600x will be just fine.

This post has been edited by nrw: Nov 9 2020, 09:47 AM

Which is why you got a honest thumbs up.
Mentioning its a review sample, being clear about what you do and why you do it.
Like that everyone can decide if it's worth looking into it and spending time, or not.

And one more thing regarding GN and your facebook post.
I'm well aware that their reviews are super detailed, and with all the money coming in at their side and the manpower they got this better be this way. Otherwise they wouldn't deserve all those subscribers.
But that doesn't change that anytime they get deeper into the matter (especially on overclocking) they do lots of mistakes themselves.
So last time I pointed out they didn't understand how Nvidia cards clock.
Another example.
https://www.youtube.com/watch?v=-UkGu6A-6sQ
They just now only discovered that 4 way memory interleaving exists.
And the way he talks they didn't get that either. Otherwise he would have tested 4x8 vs 2x16 and would have had an AHAAAAA event taking place.
So when I say "their reviews are fine for basic stuff" I don't mean that what they do is basic (you get spammed with tables that any1 not having studied statistics must get annoyed by from time to time) but that anytime they go deep they do mistakes, just like we all do.
So now ... because of this review ppl will think wow +10% performance, lets go and buy 4x8GB kits. Which is exactly the wrong conclusion.

This post has been edited by nrw: Nov 8 2020, 06:08 PM

Thanks, glad to hear that.

LOL don't get me started on their AIO video.

That 4 way interleaving even i myself get some very confusing results. I did a benchmark with 4 DIMM and 2 DIMM 8GB sticks. Same timings. I lost some read speed, but gain some copy speed if i go from 2 DIMM to 4 DIMM. I thought I'm supposed to see improvements on both read and copy.





This post has been edited by k!nex: Nov 8 2020, 11:05 PM

Actually in Aida64 Read will be about the same, only copy will see ~2% gains.

But be careful with Aida64... it is highly CPU clock dependent which lets scores go WILD.
Same goes for background tasks or even a Windows that is not clean.
Their new version is better btw, they not taking cache into numbers anymore. So results are more stable.
However easiest way to compare results is to lock CPU speed to some safe value (3.8GHz on Zen2 e.g.) and then run tests to get rid of fluctuations.