Yeah since 1st gen, Ryzen prefers even CAS timings.
Above 2666 usually the odd CL you set will round up to closest even number with Geardown mode enabled and 1T.
Disabling GDM will allow you to run at odd CL and 2T.

CPU temperatures & power draws...

Realbench v2.56 (15 mins) - CPU Temperature

CPU	Idle (°C)	Load (°C)
10900K @ 5.1 GHz All-Core	40.9	89.4
5900X @ 4.5 GHz All-Core	44.8	88.6
5900X @ CO -30/25 +200 MHz	44.0	91.5

Realbench v2.56 (15 mins) - CPU Package Power

CPU	Idle (W)	Load (W)
10900K @ 5.1 GHz All-Core	29.5	210.2
5900X @ 4.5 GHz All-Core	63.4	209.0
5900X @ CO -30/25 +200 MHz	37.3	221.6

Realbench v2.56 (15 mins) - Total System Power Consumption (CPU+GPU+etc)

CPU	Idle (W)	Load (W)
10900K @ 5.1 GHz All-Core	132.0	744.0
5900X @ 4.5 GHz All-Core	144.0	692.0
5900X @ CO -30/25 +200 MHz	116.0	704.0

Battlefield V (10 mins) - CPU Temperature

CPU	Idle (°C)	Load (°C)
10900K @ 5.1 GHz All-Core	40.3	73.5
5900X @ 4.5 GHz All-Core	44.8	71.3
5900X @ CO -30/25 +200 MHz	44.8	77.0

Battlefield V (10 mins) - Total System Power Consumption (CPU+GPU+etc)

CPU	Idle (W)	Load (W)
10900K @ 5.1 GHz All-Core	132.0	596.0
5900X @ 4.5 GHz All-Core	148.0	556.0
5900X @ CO -30/25 +200 MHz	124.0	564.0

My setups:
10900K @ 5.1 GHz All-Core: vCore auto | LLC6 | SVID Trained | C-state off
5900X @ 4.5 GHz All-Core: vCore BIOS set 1.35v | LLC auto | C-state off
5900X @ @ CO -30/25 +200 MHz: PBO Mobo | CO - ve 25,25,25,25,25,25,30,30,30,25,30,25 | OCOV +200MHz| C-state off
Custom WC: by EK-Supremacy EVO RGB + 2x RX360 + 12x GT + 1x D5
ROG Strix B550-I, BIOS 2007 | Maximus XII FORMULA, BIOS 2005
2x16GB TridentZ Royal 4000C17 B-die @ 3800 14-15-15-28, cooled by 2x 60mm fans
TUF 3080 BIOS V2, OC +150/1100 | driver 461.81
AX1200i (digital power draw monitoring)
Dimastech EasyDual V2.5
Ambient 30C A/C off

This post has been edited by owikh84: Apr 24 2021, 10:11 AM

ROG CROSSHAIR VIII / STRIX B550 latest AGESA BIOS

1. Improved system stability
2. Update AGESA ComboV2PI 1203_PatchA

📌 ROG CROSSHAIR VIII HERO 3601 BETA BIOS
https://drive.google.com/file/d/1s7ITDqsxVi...iew?usp=sharing

📌 ROG CROSSHAIR VIII HERO WI-FI 3601 BETA BIOS
https://drive.google.com/file/d/1FNA1_zB4B_...iew?usp=sharing

📌 ROG CROSSHAIR VIII FORMULA 3601 BETA BIOS
https://drive.google.com/file/d/1PZ9cmIvzjo...iew?usp=sharing

📌 ROG CROSSHAIR VIII IMPACT 3601 BETA BIOS
https://drive.google.com/file/d/1-59vYb8eQW...iew?usp=sharing

📌 ROG CROSSHAIR VIII DARK HERO 3601 BETA BIOS
https://drive.google.com/file/d/16vu2oIqmxY...iew?usp=sharing

📌 ROG STRIX B550-F WI-FI GAMING 2401 BETA BIOS
https://drive.google.com/file/d/1ezZ7BGFbc7...iew?usp=sharing

📌 ROG STRIX B550-E GAMING 2401 BETA BIOS
https://drive.google.com/file/d/1Byddq-15gh...iew?usp=sharing

📌 ROG STRIX B550-I GAMING 2401 BETA BIOS
https://drive.google.com/file/d/1LVy2Uwtx80...iew?usp=sharing

📌 ROG STRIX B550-XE GAMING WI-FI 2401 BETA BIOS
https://drive.google.com/file/d/1UTrYxR8f12...iew?usp=sharing

MSI X570/B550 New AGESA BIOS

1. Improved system performance
2. Update AGESA ComboV2PI 1203_PatchA
3. Added support for Ryzen 5000G APU

▪️MSI MEG X570 GODLIKE
▪️MSI MEG X570 ACE
▪️MSI MEG X570 UNIFY
▪️MSI MEG B550 UNIFY
▪️MSI MEG B550 UNIFY-X
▪️MSI MAG B550M MORTAR
▪️MSI MAG B550M MORTAR WIFI
▪️MSI MAG B550 TOMAHAWK

BETA BIOS DL：
https://drive.google.com/drive/folders/16uL...bnF?usp=sharing

Holla Senpai, with 1.5K budget range, which mobo is the best in term of OC and ecstatic?

For OC I always prefer boards that come with a debug LED for ease of error troubleshooting.

MSI X570 Unify or ACE
ASUS ROG Strix X570-E WI-FI

Unify is a solid board for OC, but if you want better performance efficiency then go for ASUS.
Nice VRM temp, high resale value and constant BIOS support.

Anyone tried the latest ClockTuner 2.1 RC5 already?
I recently tested the PX, P1 and P2 profiles recommended by the software. CB R20 results as below:

Cinebench R20

CPU	MC (cb)	SC (cb)
5900X @ CO -30/25 +200 MHz	8922	622
5900X @ CTR PX	7780	624
5900X @ CTR P1	8733	584
5900X @ CTR P2	8972	600

From my results, CTR P2 profile yields slightly better result than my curve optimizer method in terms of MT score (+0.6%) but lower by -3.5% in ST performance.
However, CTR PX profile gave the best ST score while registering the lowest MT score.
In conclusion, I still think that my CO still have the advantage in perfect balance of MT and ST performances.

My setup:
CPU: 5900X
CO: PBO Mobo | Per Core - ve 25,25,25,25,25,25,30,30,30,25,30,25 | OCOV +200MHz| C-state off
MB: ROG Strix B550-I, BIOS 3401
RAM: 2x16GB Klevv CRAS XR RGB 3600C18 DJR DR @ 3800 16-21-20-32-1T @ 1.40v | IF 1900 Synced
GPU: TUF 3080 BIOS V5, OC +150/1100 | driver 466.63
SSD: WD SN850 1TB + Crucial MX500 2TB
PSU: CM V850 SFX
Cooler: EK-AIO D-RGB 360mm + 6x EK Vardar-S 120 Fans
Case: Jonsplus i100 Pro

Thanks to thankyou I have a Cezanne APU to test a bit on a budget board.

Setup:
MB: MSI B550M A-PRO (Bios 2.40) [attached some small aluminum heatsinks to mosfets]
APU: AMD R7 5700G
RAM: 2x8GB 3733 22-23-23-55 HP37D4U1S8ME (weird HyperX Fury kit made for HP) (XMP 1) / changed to a average B-Die Kit later on
M.2: MP510 480GB
GPU: RTX 3080Ti / removed at later part for iGPU testing
PSU: AX850
Cooling: Custom Loop (1x 360mm slim rad) / changed to Sythe Fuma 2 after removing GPU
OS: Win 10 21H1 using windows balanced power plan
latest AMD drivers as of June 17th, 2021

CB R20 MC

Package Power	Temp CPU	Temp Mosfet	Average Score	Points Per Watt
95W	77°C	70°C	5505	57.95
88W (Stock)	75°C	68.5°C	5365	60.97
57W	58°C	56°C	4824	84.63
40W	50°C	50.5°C	4240	106
96W (1.275V load, 1.3V Bios LLC4) 4.5GhZ all core	76.5°C	70.5°C	5842	60.85

CB R20 MC/SC Scores [incl. mild boost clock override. (4.725GHz single core boost) 95W Package Power]

MC	SC
5529	592

CB R20 MC/SC Scores [incl. CO (-11,-6,-11,-12,-13,-11,-11,-10)] (based on above settings)

MC	SC
5628	600

Aida64: (changed to 4000 1:1:1 16-17-17-34 B-Dies) (can boot up to 4200 in sync but the required vSoC is getting to high for my tastes)

Read	Write	Copy	Latency
~61.2k	~61k	~55.5k	~50ns

Furthermore the CB20 scores also improved by... ~1% not gonna bother putting that down here.

Quick Gaming Benchmark:
SoTR CPU Game/Render 1080p Highest

Min	Max	Average	95%
131/163	252/419	187/247	135/182

~15% slower than a 5800x with similar optimization.

iGPU time now (switched to a more realistic 3600 1:1:1 16-17-17-34 GDM enabled which should be easily possible for anyone with b-dies and a little time)
SoTR 3600 16-17-17-34


SoTR 2400 JEDEC



Conclusions:
Firstly I'd like to point out that this board despite being low in spec handles the top end Cezanne pretty good.
There are overclocking limitations such as on the IF part or even on the APU itself but this hardly matters for regular users.
If the connectivity is sufficient, this board will do just fine for any Cezanne APU.

Cezanne is a bit of a mixed bag. I have just pushed until IF 2100 1:1:1 and stopped there as the vSoC requirements were too high for what I'd consider a daily driver.
Even at this point it can't compete with the 5800X when it comes to gaming. The cache cut half is just too much. You'll end up looking at ~15% performance loss with similar system configurations whenever being CPU limited.
However this is also promising news with a Zen3 refresh with added cache around the corner.

Cezanne behaves a little different to Vermeer. It boosts roughly 150-250MHz lower in different load states and sticks to a single core for single core loads rather than cycling between two cores like Vermeer is doing. Latency didn't drop just as much as it did using Renoir over Zen2, I guess that's down to AMD optimizing Zen3 quite a bit there.

So should you consider a 5700G at ($359 USD) over a 5800X ($449 USD) or a 5600G ($259 USD) over a 5600X ($299 USD)? If you use a GPU I'd rather stick to their 5600X/5800X despite the top up in price. Alternatives would be Rocket Lake CPU's obviously such as the 11400F instead of the 5600X or 11700 instead of the 5800X. Hopefully prices on the Vermeer's will come down slightly with the Cezanne launch anyways.

But lets come to where Cezanne really shines. Power draw. This is the perfect APU for a GPU less system for casual 1080p low graphics gaming and regular workloads. It doesn't lose much efficiency going down to 57W package power. This 57W is also where this APU matches a 3700x (at a fraction of the power draw!) which should also be reasonably close in pricing upon launch. So imagine 3700x performance in a small system which is barely audible. Enough to get your work down and for some casual gaming with sacrifices on resolution and graphics.
I just reran the CPU at 55W package power, air cooling (Fuma 2) with the Curve Optimizer + Boost clock override settings above. CB R20 MC score is stable ~@4940 points over 10 runs. And the system is hardly audible on a silent fan profile running at roughly 30K above ambient.

It'll also be easily possible to run this chip using a PicoPSU after limiting power draws.
Take a look at the total system (excluding monitor) power draw of the above system (AX850) in idle.


That's 13.2W for Mainboard, APU, RAM, Peripherals, 2 Fans, active iGPU with connected 1080p monitor ... with a PSU that will have quite some switching losses at this load. Take a PicoPSU and you'll be at ~10W. Possibly just below. Your Vermeer CPU can't even power its IO Die with that...

At this point I'd like to give a thanks to [thankyou] yet again for letting me test his 5700G. I'll probably be buying the 6 core myself if it is priced close to the USD msrp. Cheers.

This post has been edited by nrw: Jun 19 2021, 08:06 PM

So difficult to get one.

2 questions for you

1. How did you observe the idle W of the APU?

2. 3733 CL22? Possible to go tighter timings?

1) so far using HWiNFO64, from my past experience (Renoir APU) with the digital reading from my AX1200i and a watt meter it's accurate. After I'm done with all my stuff I will slap a Fuma 2 on and show wall power draw.
2) For sure! That's just the XMP. SPD not accessible and RAM not mine so I can't take off heatspreaders to see what memory IC it is using. Anyways once Curve OC is done RAM is next on the list. Will figure with timings/frequency which memory IC I am likely to be dealing with.

This post has been edited by nrw: Jun 18 2021, 05:00 PM

Thanks. I never took note of 3733 kit like this.

I thought CL20 is like for 4000MHz (just as 18 for 3600, 16 for 3200) so seeing 3733 at CL22 very really low latency.

hey guys, just wondering is it ok to set my 3600 OC to 4.2 and 1.25volt(can go lower) for daily use? if I use default or PBO the core voltage is fluctuating sometimes over 1.4. stable on cinebench(temp <75c) and prime(<85c)

Hi Guys, i just switch from my Intel to AMD platform yesterday.

5900X + Asus RS Gaming-XE Wifi, the rest (including RAM) just migrated from Intel rig as in my siggy.
I have alot of confusion and question would like to ask but will do my reading in here as well:

1. HWinfo, CPU-Z and AIDA constantly shown one of my DIMM is missing (although the one that reported missing is different), however all can report 64GB in total.


2. The temperature sensor only on 'CPU' and 'CPU package', not per die? Maybe be i too used to Intel platform already.

3. I never ever see any single core reach 4.95Ghz using Ryzen Master even with PBO. Maybe i need a suitable benchmarking tools?
4. Looks like the idle temperature of this 5900x is much higher than my 9900KS. Seems like all the SW report different temperature...i don't know who to trust.
5. The memory latency of same memory setting yield 70ns which is almost 20ns slower than my 9900KS, after search online looks line memory latency is higher in AMD platform.

In overall, i'm drown in the sea of parameters. Still learning. Would be better if someone can give me some insight on my question/observation above

This post has been edited by coolkwc: Jul 2 2021, 11:54 AM

1) Something wong
2) use only hwinfo and watch the tdie which also includes the IOD sometimes, dont trust anything else including ryzen master.
3) you're probably on older bios.
4) yes AMD idle and average powerdraw and temp is crap.
5) yes AMD latency is crap, but that is compensated by the L3 cache. However usually only some games will suffer noticeably.

Sure you can do that. Especially if you're using it for lighter loads only.
I'd probably do the same on a 3600 nowadays if its a solid one.

coolkwc in addition to what Bonchi suggested.
Make sure you only have one monitoring tool installed. That should be hwinfo64. Uninstall others including that Asus one.
Use the power plan TristanX shared in the AMD Hardware brand thread some time ago regarding idle optimizations.
Set your PBO to advanced, key in the stock limits, and see how far you can push the boost clock override. You might get away with +200.
If it crashes just go back 25Mhz each time.

This post has been edited by nrw: Jul 3 2021, 12:59 AM

After some crash course tonight, finally understand the PBO and curve optimizer. This is my Cinebench R23 result with the setting below:

1. Curve optimizer - All core -20
2. Boost override - +200MHz
3. Scalar - Auto

Multicore running



Singlecore



Any comment?

This post has been edited by coolkwc: Jul 3 2021, 02:25 AM

I switched from Intel to AMD platform, from Gigabyte Z390 Aorus Master to Asus Rog Strix B550-XE Gaming, both should be same range.
So if Gigabyte Aorus Master in AMD follow the same design language as in Intel platform, i would say ASUS no match in terms of build and ecstatic.

Yesterday when i unboxing the Asus board, immediately i felt a huge downgrading.
Cheap VRM heatsink, no back plate, cheap RGB effect, cheap metal appearance, and despite a 1.5k board, there is NO clear CMOS switch...oh come on, i need to short the clear CMOS using jumper, what era is that.

Just hope the reliability is better.

remember to monitor your VRM temps as you seem to be drawing 200W at load especially on an upper mid range board which is not exactly designed for max out OC.

pretty much that's it since youre almost hitting the 90C thermal limit. maybe try reducing voltage to lower the temps and find more headroom.

I feel that 5000 series dont seem to benefit much from RAM speed and latency as much as 3000 series and performance seems to stagnant after 3600mhz. (maybe nrw can clarify) but the next step should be playing with memory OC.

Thanks @Bonchi

How to reduce the voltage when using PBO + curve optimizer? The curve optimizer is reduce the voltage opportunistic right?

Anyway i try CPU voltage offset, -0.05V, seems able tonshake off 15W during multicore, boost clock still maintain 4.6Ghz. is that the right way to reducr the voltage?

This post has been edited by coolkwc: Jul 3 2021, 11:58 AM

Yeap through the offset.