[Baio73]

R23 don’t really care about ram.

And RAM OC is the next step ahead!
I've started over with a new kit but I'm struggling a little:

A guide to ram overclocking on Zen 3 | Page 4 | Overclock.net

Baio

 

[Baio73]

Try +100 instead of +200.
With your cooling might benefit from even lower tdc and edc. R23 isn’t really heavy on amps.

Thanks, I'm gonna try +100.
Any tip for lower TDC and EDC?

Baio

 

[Luggage]

Thanks, I'm gonna try +100.
Any tip for lower TDC and EDC?

Baio

You just said you ran with lots of processes in the background - so no then I don't have any tips because there is nothing to compare against. You can't optimize any limit or value if you have a billion random parameters dirtying up the results.

This is a tip going forward if you wan't to OC RAM - clean up bg apps and processes or you will be in for a world of hurt trying to figure out latency and bandwidth

edit: looking at your system you really want to disable icue and g-suite(?) while messing with yourr ram.

 

[flyinion]

Did you guys lose ability to load saved setting from USB after updating to 4201? Sooo not happy right now. All settings from scratch cause they won’t load. “Not for this cpu family”

 

[xeizo]

Did you guys lose ability to load saved setting from USB after updating to 4201? Sooo not happy right now. All settings from scratch cause they won’t load. “Not for this cpu family”

I noticed it on some of the AGESA 1.2.0.7 upgrades I did, for me it's not a biggie though as by now I know most if not all bios settings from memory.

Different thing with my Z690 rig, a lot of memory settings I'm not familiar with have to thread carefully LoL

 

[flyinion]

I noticed it on some of the AGESA 1.2.0.7 upgrades I did, for me it's not a biggie though as by now I know most if not all bios settings from memory.

Different thing with my Z690 rig, a lot of memory settings I'm not familiar with have to thread carefully LoL

Yeah the problem is I had to redo my fan profiles in BIOS. Luckily fan xpert didn't eat them this time (though I have backups). I've had it reset its base profiles before. Well, it did this time as well but still saw my saved ones. I just had to set the temp source back to the water temp sensors again. I really didn't want to have to redo profiles for the fans on my loop.

 

[des2k...]

I don't understand exactly which point you are referring to.

If you disable all the CPPC settings, you will get less performances because the OS doesn't know anymore which core relates to which CCX cache, so it can't group threads that work on the same set of data to the cores that use the same cache, and this will inherently slow down cpu heavy algorithms as accessing the cache of a different CCX has a big performance penalty.

The processor performance boost mode is a setting that defines the speed with which a core will increase its frequency up to the maximum turbo boost mhz when a thread with a heavy workload requires it.
When CPPC is OFF, the CPU will use the P-states to manage each core frequency, and the P-states allow only two possible frequency states: Turbo boost ON, and Turbo boost OFF. So the "Aggressive" setting will be mapped to "Turbo boost ON immediately" and the "Efficient" setting will be mapped to "Turbo boost ON only if the thread specifically ask for the turbo frequency".
When CPPC is ON, the CPU has more granularity to choose the target core frequency, so the "aggressive" setting will be mapped to "Turbo boost ON immediately" and the "Efficient" setting will be mapped to "Start at an OS-calculated frequency and progressively increase up to theTurbo boost frequency".

All the options in that list are always mapped to one of the available choices following this table:

View attachment 2560466

There isn't one of those listed choices that is not "functional" if CPPC is ON or OFF. In Windows 11 the default for both the "balanced" and "high performance" profiles is to use "Aggressive" as boost mode and I didn't get any performance increase by using a different value (I got performance increases by changing other power profile settings unrelated to CPPC).

But the performance penalty that you get if you disable CPPC is not related to the frequency of the cores. It's related to the ability of the cores on the same CCX to access the same cache instead of having to do an extra roundtrip to access the cache on the other CCX. Without knowing which cores are in which CCX , the OS can't optimize this cache access strategy.

Glad it works for you. On my side after we got cppc support in bios. The default aggressive only worked with cppc off. Eff aggressive with cppc on boosted based on ranking and scheduled threads properly with my 3900x and 5900x.

Always have hwinfo open on my 2nd screen, aggressive (default) doesn't boost very well on dual ccds. It's only 1 to 4 cores randomly (ccd0 or ccd1) being loaded vs having ccd0 boosted on ranks then moves to 1-2 cores from ccd1.

This is with normal light games, heavy threaded games will trigger all cores to boost.

Here's good scaling for core boost / thread scheduling(~86w with Destiny2 at 4k) which I don't get with aggressive default.

 

[des2k...]

With my 5950X, I finally can boot at RAM 4000MHz/2000MHz.
This was impossible with my former 5900X.

I also set my best optimized primary and secondary timings in order to boot.

Aida64 (L1 + L2 HW Prefetcher disabled, which provided better latency but lower Memory Read/Write/copy values) :
View attachment 2560435

I could pass 1 hour KArhu's (1800%), as a first shot for stability.
Of course this is not enough and I will test it overnight to confirm stability.

BUT...This is Whea 19 fest....

I installed Whea suppressor service from @ManniX-ITA.

Let's see how things go in the next days.

L1,L2 prefetch off reporting better latency in Aida64 is not accurate or representative of real usage.

here's what it came out with when I tested aida64 yesterday.
l1,l2 on
4000cl16 with no apps open 53.2ns, open a few apps and latency climbs to 53.4ns on repeated runs


l1,l2 off
4000cl16 with no apps open 53.0ns, open a few apps and latency climbs to 54ns on repeated runs
and -700MB/s bandwith loss

this makes sense because once you have more than one program/process active dealing with memory you need prefetch active to get good bandwhith / latency response from the memory side

this also shows as less smooth fps in heavy games

 

[GRABibus]

Have you tried DDR4 4000 - GDM off, CL15, T2 ?

L1,L2 prefetch off reporting better latency in Aida64 is not accurate or representative of real usage.

here's what it came out with when I tested aida64 yesterday.
l1,l2 on
4000cl16 with no apps open 53.2ns, open a few apps and latency climbs to 53.4ns on repeated runs


l1,l2 off
4000cl16 with no apps open 53.0ns, open a few apps and latency climbs to 54ns on repeated runs
and -700MB/s bandwith loss

this makes sense because once you have more than one program/process active dealing with memory you need prefetch active to get good bandwhith / latency response from the memory side

this also shows as less smooth fps in heavy games

@J7SC
CL15, 2T => no boot, even with [email protected]
I don't want ot tweak again all timings and voltages



With L1 + L2 auto :

But with this memory OC, I had to tweak again all my CO curve as it crashes or WHEA 18 like hell in gaming.

conclusion : if I use this nice memory OC in gaming with L1 + L2 HW auto, as I had to add voltages through CO to all my cores, I have now poor single core performance, but very nice memory and cache Bandwith and latencies.

 

[neikosr0x]

DELETED

 

[GRABibus]

Have they fixed (4201 the EDC limit yet? Just tried 145EDC and Ryzen master is showing that limit.

No, still Vid=1.425V if EDC>140Amps.

 

[GRABibus]

@J7SC
CL15, 2T => no boot, even with [email protected]
I don't want ot tweak again all timings and voltages



With L1 + L2 auto :
View attachment 2560532

But with this memory OC, I had to tweak again all my CO curve as it crashes or WHEA 18 like hell in gaming.

conclusion : if I use this nice memory OC in gaming with L1 + L2 HW auto, as I had to add voltages through CO to all my cores, I have now poor single core performance, but very nice memory and cache Bandwith and latencies.

I fixed those crashes in games by raising PLL from 1.9V to 1.95V.
I could come back to much lower CO voltages, as I had with [email protected] (see in sig for details).

Of course, 1.95V PLL is an absolute max voltage for me to not go beyond.
My SOC is at 1.2V.

 

[J7SC]

Glad it works for you. On my side after we got cppc support in bios. The default aggressive only worked with cppc off. Eff aggressive with cppc on boosted based on ranking and scheduled threads properly with my 3900x and 5900x.

Always have hwinfo open on my 2nd screen, aggressive (default) doesn't boost very well on dual ccds. It's only 1 to 4 cores randomly (ccd0 or ccd1) being loaded vs having ccd0 boosted on ranks then moves to 1-2 cores from ccd1.

This is with normal light games, heavy threaded games will trigger all cores to boost.

Here's good scaling for core boost / thread scheduling(~86w with Destiny2 at 4k) which I don't get with aggressive default.

View attachment 2560521

Probably makes more sense to look at effective core speeds...below is an oldie on nominal ones which at the end of the day don't mean that much...

As mentioned before, FS2020 is a really good test as it continuously stresses one core more than any other, and effective speeds are typically around 5025 - 5040 max, with L3 effective speed also just over 5000...because of that, I actually introduced a small negative offset for the CPU - don't like to see 1.5v or higher ever, even for just a brief moment on one core - old school I guess...

 

[Kelutrel]

this makes sense because once you have more than one program/process active dealing with memory you need prefetch active to get good bandwhith / latency response from the memory side

This statement makes no sense. Explain, if you think you know what you are talking about.

 

[KedarWolf]

L1,L2 prefetch off reporting better latency in Aida64 is not accurate or representative of real usage.

here's what it came out with when I tested aida64 yesterday.
l1,l2 on
4000cl16 with no apps open 53.2ns, open a few apps and latency climbs to 53.4ns on repeated runs


l1,l2 off
4000cl16 with no apps open 53.0ns, open a few apps and latency climbs to 54ns on repeated runs
and -700MB/s bandwith loss

this makes sense because once you have more than one program/process active dealing with memory you need prefetch active to get good bandwhith / latency response from the memory side

this also shows as less smooth fps in heavy games

Whenever you have any apps running, benchmarks are gonna get worse. Try R23, y-cruncher, anything, open apps slow things,

 

[escoltajuverf]

bios with agesa starting from version 1.2.0.5 including the latest 1.2.0.7 have reboot (warm boot) hang qcode 01, 1f or 0d when using 4x16 gb micron sticks, going back 1.2.0.3 Patch C for now

 

[tommyd2k]

Anyone using liquid metal with an EK monoblock? I know it isn't recommended and I know the risks. I'm not asking because I expect miracles from thermal compound either. The HDMI output on my Powercolor 6900xt is broken and to fix it I need to use my reflow setup and replace the port. While I have it apart I am gonna try out liquid metal. It's got a Byski waterblock I installed a year ago, and I think I'll see a noticable difference going to LM. So while I am at it, I wanna give it a try on the CPU.

It's also because I just got a replacement 5950X from AMD. I gotta say it was the best decision I have made in a while in regards to computer parts. One look at the before and after results you can see why.

The 5950X I returned

Results CO testing
CORE| CPPC | START | RES | TEMP | CO | UNSAFE
C01 | 215 | 4950 | 4950 | 75 | 258 | False
C02 | 220 | 4825 | 4850 | 74 | 20 | False
C03 | 206 | 4875 | 4875 | 68 | 158 | False
C04 | 211 | 4750 | 4900 | 74 | 200 | False
C05 | 220 | 4900 | 4900 | 68 | 56 | False
C06 | 201 | 4900 | 4875 | 70 | 181 | False
C07 | 197 | 4800 | 4800 | 74 | 155 | False
C08 | 192 | 4800 | 4800 | 74 | 154 | False
C09 | 178 | 4675 | 4775 | 69 | 153 | False
C10 | 187 | 4900 | 4900 | 63 | 178 | False
C11 | 173 | 4775 | 4775 | 68 | 191 | False
C12 | 183 | 4650 | 4725 | 69 | 96 | False
C13 | 155 | 4825 | 4825 | 66 | 228 | False
C14 | 169 | 4650 | 4825 | 70 | 225 | False
C15 | 164 | 4575 | 4750 | 72 | 210 | False
C16 | 159 | 4625 | 4750 | 74 | 195 | False

The one I just got back from AMD

Results CO testing
CORE| CPPC | START | RES | TEMP | CO | UNSAFE
C01 | 215 | 5050 | 5025 | 63 | 235 | True
C02 | 220 | 4900 | 4925 | 63 | 6 | False
C03 | 206 | 4975 | 4975 | 60 | 216 | False
C04 | 211 | 4825 | 4950 | 61 | 162 | False
C05 | 220 | 4875 | 5000 | 62 | 96 | False
C06 | 201 | 4775 | 4950 | 61 | 196 | False
C07 | 197 | 4725 | 4900 | 65 | 205 | False
C08 | 192 | 4875 | 4875 | 66 | 179 | False
C09 | 178 | 4725 | 4850 | 57 | 146 | False
C10 | 187 | 4975 | 4975 | 60 | 216 | False
C11 | 173 | 4850 | 4850 | 58 | 178 | False
C12 | 183 | 4700 | 4800 | 57 | 92 | False
C13 | 155 | 4900 | 4900 | 58 | 236 | False
C14 | 169 | 4700 | 4875 | 59 | 193 | False
C15 | 164 | 4625 | 4800 | 61 | 191 | False
C16 | 159 | 4675 | 4825 | 60 | 187 | False

And efficiency rating??

Results CCD testing
CORE#1 CO: 24 CORE#9 CO: 33
CORE#2 CO: 9 CORE#10 CO: 15
CORE#3 CO: 47 CORE#11 CO: 86
CORE#4 CO: 35 CORE#12 CO: 58
CORE#5 CO: 25 CORE#13 CO: 62
CORE#6 CO: 49 CORE#14 CO: 47
CORE#7 CO: 78 CORE#15 CO: 96
CORE#8 CO: 65 CORE#16 CO: 67
Energy Efficiency CCD#1 4.1 | PLATINUM sample
Energy Efficiency CCD#2 4.05 | GOLDEN sample

Old one was a double Bronze. So I was pretty psyched when I got those diagnostic results.

So I am probably not going to be removing the processor again anytime soon. I just ordered a 5g tube of liquid metal to try out on my GPU and CPU. I have the EK CH8 monoblock. So anyone who is or has used liquid metal on this cooler, want to let me know how it worked out? I will remove the mobo from the case to do the application and take any steps I can to prevent shorts. That being said, any tips would be great. I have never used liquid metal b4. I don't normally concern myself with trying different TIM products, an 8G tube of MX-4 is a good as any to me . But since I don't expect to have to remove the cooler again, if it can shave off a few degress, I'll give Conductonaut a try.

Of course if there is some reason why it will not work correctly at all I would like to know. What steps I should take to protect the chip and the socket to avoid a catastrophe would be helpful.
At the end of the day if it makes a noticeable drop in temps then great. If not, as long as it doesn't perform worse then at least I tried right?

 

[Theo164]

Same efficiency ratings here 5900X bought November 2020

I've used Conductonaut LM many times with delided intel cpus on the die under heat spreader after insulating other nearby components like capacitors e.c.t. and the result's was excellent

LM is a risk to use on non isolated surfaces, an accident can kill whole pcb. This thing can get under electronic components like inductors / mosfets and any kind of surface mounted components and create permanent short circuits...

Check this video as an example before make any decision

 

[stimpy88]

I am having problems with fTPM...lots of stuttering when enabled but I can't play games such as Valorant if I disable fTPM because they need to have fTPM enabled...

Well, the latest BIOS is supposed to fix fTPM issues... You should take screenshots of your 3302 BIOS settings, and update to the latest BIOS, reset everything to optimized defaults, and see how you get on.

You can always use BIOS Flashback to go back to 3302 if the later BIOS causes other issues.

 

[stimpy88]

Thanks, I'm gonna try +100.
Any tip for lower TDC and EDC?

Baio

220
145
145

This is the way.