X670E Ace Bios Problems

[Michele P]

some feedback after few days with the latest 184 bios 1.0.0.7a

6200C28 at 1.24V SOC (I won't go over 1.25 even if the safe point is considered to be 1.3V "now" )

Other voltages have been further lowered from stock:

VPP 1.95V -> 1.8V
VDDP 1.15V -> 1.10V
unlinked VDD/VDDQ/VDDIO
VDDIO 1.35V -> 1.20V
VDDQ 1.35V -> 1.34V

Only voltage increased is the VDD for the ram, but that's on totally separate circuit, should be ok as long as the ram is stable and doesn't overheat (1.35V -> 1.44V)
Everything stable after 20ish Testmem5 1usmus cycles
Tested also 1 hour y-cruncher VST for fclk 2167 stability (check latest buildzoid video about infinity fabric stability)
For every day use I will stick to 2133, still a bit unclear how to detect properly fclk instability without an oscilloscope, better stay safe.
PBO CO -15 all cores, no problems, infamous aida sha3 test doesnt crash (some unlucky cpus crash @ stock settings xD )
There's still the same old bug with fmax boost override PBO option that breaks entirely CPPC on X3D cpus... but that I think it's not a bios bug, it's more like AGESA's fault.
I know this option is half useless on dual mixed CCDs, but it breaks many things and need fixing from amd. No hurry, it will be completely useless if the cpu burns before XD

 

[RemusM]

some feedback after few days with the latest 184 bios 1.0.0.7a

6200C28 at 1.24V SOC (I won't go over 1.25 even if the safe point is considered to be 1.3V "now" )

Other voltages have been further lowered from stock:

VPP 1.95V -> 1.8V
VDDP 1.15V -> 1.10V
unlinked VDD/VDDQ/VDDIO
VDDIO 1.35V -> 1.20V
VDDQ 1.35V -> 1.34V

For 6200 VDDIO should be in the 1.30-1.35V range (and VSOC at least 1.25V).
But if your system is rock stable with those voltages, that's great!

 

[Michele P]

For 6200 VDDIO should be in the 1.30-1.35V range (and VSOC at least 1.25V).
But if your system is rock stable with those voltages, that's great!

I know at default voltages should be more conservative to secure stability on many different machines.

Have tested everything, from testmem 1usmus to y-cruncher and aida

and also few hours of multiplayer gaming, that usually triggers the smallest instability from my experience

More voltage is not always better in terms of stability (I remember another recent buildzoid video about that)
Maybe this is valid if you keep all the voltage linked together
Im not a OC guru but from what I understand the principle is that if you un-link everything and respect some proportions you can run lower voltages because it improves signaling between the different components.

 

[skitu157502df]

some feedback after few days with the latest 184 bios 1.0.0.7a

6200C28 at 1.24V SOC (I won't go over 1.25 even if the safe point is considered to be 1.3V "now" )

Other voltages have been further lowered from stock:

VPP 1.95V -> 1.8V
VDDP 1.15V -> 1.10V
unlinked VDD/VDDQ/VDDIO
VDDIO 1.35V -> 1.20V
VDDQ 1.35V -> 1.34V

Only voltage increased is the VDD for the ram, but that's on totally separate circuit, should be ok as long as the ram is stable and doesn't overheat (1.35V -> 1.44V)
Everything stable after 20ish Testmem5 1usmus cycles
Tested also 1 hour y-cruncher VST for fclk 2167 stability (check latest buildzoid video about infinity fabric stability)
For every day use I will stick to 2133, still a bit unclear how to detect properly fclk instability without an oscilloscope, better stay safe.
PBO CO -15 all cores, no problems, infamous aida sha3 test doesnt crash (some unlucky cpus crash @ stock settings xD )
There's still the same old bug with fmax boost override PBO option that breaks entirely CPPC on X3D cpus... but that I think it's not a bios bug, it's more like AGESA's fault.
I know this option is half useless on dual mixed CCDs, but it breaks many things and need fixing from amd. No hurry, it will be completely useless if the cpu burns before XD

What is your pyprime result with your previous CL30 settings?
I'm not sure if pyprime scales with CPU. Since the ACE has an external clock generator I'm curious what your scores would be if you used asynchronous eclk 105 MHz instead of a CO-15.
105MHz is easy to run, I have been running it for 2 weeks now on stock voltages except with 1.21v vsoc.

 

[Michele P]

What is your pyprime result with your previous CL30 settings?
I'm not sure if pyprime scales with CPU. Since the ACE has an external clock generator I'm curious what your scores would be if you used asynchronous eclk 105 MHz instead of a CO-15.
105MHz is easy to run, I have been running it for 2 weeks now on stock voltages except with 1.21v vsoc.

I have this old screenshot, but I used different (wrong) timings and a step lower fclk
EDIT: found the correct screenshot, it was posted on my pcpartpicker build page in sign, d'oh !
look at the bottom lists of benchmate saved results
I was around 7.94x with decent 6000C30 timings

PYPrime is very fclk and cas sensitive
6200C30 pyprime result can be easily worse than a well tuned 6000C28

example:

 

[skitu157502df]

I have this old screenshot, but I used different (wrong) timings and a step lower fclk
EDIT: found the correct screenshot, it was posted on my pcpartpicker build page in sign, d'oh !
look at the bottom lists of benchmate saved results
I was around 7.94x with decent 6000C30 timings

PYPrime is very fclk and cas sensitive
6200C30 pyprime result can be easily worse than a well tuned 6000C28

example:

Interesting, I was wondering why my I get similar results with worse timings. Its the eCLK overclock
ECLK 100MHz (Auto/Stock)

ECLK 105 MHZ

The max effective clock is higher than what I can see on your 7950x3D

 

[Michele P]

Every fclk mhz matters in this pyprime test... the test also benefits a lot from the cache CCD, few mhz more on that CCD have a big impact, no surprise.
I don't have any eclk applied, that's only way to increase the clock of cache ccd... even pbo fmax override has no effect on that CCD, it only works on the frequency CCD and at the moment is broken because it breaks entirely CPPC.
I only have a tiny undervolt (CO -15 all cores), so stock clocks... given my room is always pretty hot, I'm more incline to undervolt rather then overclock
If you see my post with 6200C28 screen only voltage above the standard is the VDD for the ram, every other voltage has been drasticaly reduced from stock (or at least the expo defaults)
I have also prepared the same config for 6000C28 with even lower voltages, ready for summer

 

[jfrobinso156502de]

I have not updated to the latest beta BIOS. So, I was wondering, for those that have, how is it working for you?

 

[Michele P]

I have not updated to the latest beta BIOS. So, I was wondering, for those that have, how is it working for you?

for me it's working good, just read my latest posts with tests and findings, all on 184 bios (start from #639, #661, and follow ups)
With this bios you have to pay attention only to one thing: the fTPM bug, well not really a bug, maybe it's more like a GUI glitch.
Don't try to enable/disable fTPM from the toggle on the main bios page or It will bug something about a fan failure alert.
But you shouldn't disable fTPM anyway so... it's not a problem

 

[frozen_12157b02de]

i just had a quick scroll thru the recent posts. seems mostly ram related. just curious if the network controller dropping issue ever got resolved? are people still having that problem? im still on 170, and still relatively problem free, though that issue would be nice to resolve.

 

[Michele P]

network droppin is a marvell driver issue
install 3.1.7
disable low power ethernet and segmented coalescing and you are good, not a single network drop out
at least for me, I play a lot of multiplayer and I definitely noticed the network dropping before making those changes.
Also check the event viewer for aqnic lost connection #14 events
I only get one single event 14 during windows booting process, I think that's normal.

 

[frozen_12157b02de]

oh thats good, somehow i missed that info. just making sure, this is the one, right? cant say iv ever downloaded marvel drivers before...

wnloaded marvel drivers before...

 

[Michele P]

yes, I use that driver
with the mentioned fixes in adapter properties
I got rid of the random event 14 lost link events, no more dropouts

in the recent posts you find some ram info ok, but also the fTPM bug and some notes I added about the "unnecessary" high signaling voltages the mb is applying by default (VDDIO, VDDP, ecc)
if the system is stable with even lower voltages than default should mean bios/hardware is working good.
Nothing wrong using the default "high" voltages, some of them don't even deliver power because they are used for signal transmission

SOC is different, that can melt cpus, so be delicate with that xD
but not too much, even if stable a too much low value can handicap your latencies

 

[frozen_12157b02de]

@Michele P , i did the marvel update. went into device properties, into the advanced tab, to make the adjustments you mentioned. i found "Energy-Efficient Ethernet" (not low-power as you described it) and disabled that. i found "Recv Segment Coalescing" for IPv4 and 6. I disabled them both. Now ill do a restart and see if i get any more drop outs!

 

[Michele P]

The name can change between different manufacturers, EEE, green, low power, etc, it’s always the same energy saving option.
I suggest to disable it, just in case, but I think the dropouts are caused by the coalescing options

 

[BiscuitBomber]

Is there a particular reason why 7D69v184 and 7D69v182 are still BETA? I'm still using 7D69v17 on my 7800X3D which is working perfectly.

 

[frozen_12157b02de]

i know this is slightly off topic for this particular thread, but since there has been so much discussion on memory here, i thought this was a good place to ask this.

so im going to build an itx travelling/work system. not going balls to the wall with the components. so far i've picked out...

MSI MPG B650i Edge Wifi
R5 7600x

Im trying to decide on memory. At 1st i was thinking id just get 32gb of something moderate, but since memory prices have come down a bit since i build this 7950 system, and 6000 memory now seems to be pretty stable, maybe i take the memory in this system, put it in the itx system, and get some 6000 dimms for this, my main work system. Oddly, when checking the memory compatibility chart on the MSI site, I cant find my current kit in there for either this board, or the itx board, though i know i picked it based on it being compatible back in January. I currently have Kingston KF552c40-32. So i would move this kit to the itx board.

I went thru all the Hynix enteries for the Meg and picked out a few 6000 kits:
-Fury Beast 64 cas36
-Fury Beast 64 cas36 RGB (not sure if the rgb is the only diff here?)
-Fury Beast 64 cas40
-Trident 64 cas30
-Ripjaws 64 cas30

I would lean towards the last one as the price is a bit lower than the others and there is no RGB (not an rgb fan), plus its also the lowest CAS. I know alot of other users here have done alot more memory testing than I have, as iv just been content to sit here with my 5200 kit, so if anyone has any insight here on which might be a good kit to upgrade to for 6000, id love some input. Iv always had good luck with Kingston, and i feel like i recall other users here saying they were having issues with G.Skill...?

thanks!

 

[BiscuitBomber]

i know this is slightly off topic for this particular thread, but since there has been so much discussion on memory here, i thought this was a good place to ask this.

so im going to build an itx travelling/work system. not going balls to the wall with the components. so far i've picked out...

MSI MPG B650i Edge Wifi
R5 7600x

Im trying to decide on memory. At 1st i was thinking id just get 32gb of something moderate, but since memory prices have come down a bit since i build this 7950 system, and 6000 memory now seems to be pretty stable, maybe i take the memory in this system, put it in the itx system, and get some 6000 dimms for this, my main work system. Oddly, when checking the memory compatibility chart on the MSI site, I cant find my current kit in there for either this board, or the itx board, though i know i picked it based on it being compatible back in January. I currently have Kingston KF552c40-32. So i would move this kit to the itx board.

I went thru all the Hynix enteries for the Meg and picked out a few 6000 kits:
-Fury Beast 64 cas36
-Fury Beast 64 cas36 RGB (not sure if the rgb is the only diff here?)
-Fury Beast 64 cas40
-Trident 64 cas30
-Ripjaws 64 cas30

I would lean towards the last one as the price is a bit lower than the others and there is no RGB (not an rgb fan), plus its also the lowest CAS. I know alot of other users here have done alot more memory testing than I have, as iv just been content to sit here with my 5200 kit, so if anyone has any insight here on which might be a good kit to upgrade to for 6000, id love some input. Iv always had good luck with Kingston, and i feel like i recall other users here saying they were having issues with G.Skill...?

thanks!

With my build I had better luck with Kingston than G.Skill, but I'm sure your mileage may vary.

 

[frozen_12157b02de]

i guess the other question is will i see that much of a difference going from 5200 cas40 to 6000 cas 36? with the hynix memory id be able to use the High Efficiency mode, which as far as i know is hynix only, or at least that was the case a while back.

EDIT: just out of curiosity i opened my case and took the part # of my current 5200 ram, and put it into the QVL again incase i was getting the wrong info from the store website. this time it did show up, and it showed up as Hynix? im looking at HWinfo right now and it says SDRAM Manufacturer: Micron. So whos lying?

 

[Michele P]

qvls can be inaccurate, many vendors, including g.skill are selling same exact models with different chips
because manufacturer costs and more availability of certain parts they can decide to make changes, and more time passes more inaccurate/outdated the list can become
example, my kit is listed as m-die in msi qvl list, but in fact it's a-die
sticker on the ram says a-die
timings also confirm a-die, because all m-die refuse to boot under 400 tRFC and I can confortable run 3xx even at 6200c28
don't trust any qvl: look the stickers, serials, codes, if you want to be 101% sure only way is look under the heathsink or find the actual limits of the ram on a running system.
DDR5 is new, expect many different vendors/revisions coming in the near future, manufacturers will just keep the old model number while changing the internal components, silently