MAJOR issue: Zen 3 Motherboard BIOS issues (UPDATE: FIX FOUND! 1933MHz IF Stable)

[Guest]

5 minutes ago, terminalinfinity said:

This is true.  My general advice right now would be just make sure you run a stability test and don't assume its stable because it boots.  Treat even previously stable XMP profiles if you're doing a drop in upgrade as unknown.

For certain! 

 

I just wonder if anyone has played around in the sub timings yet for a resolve. There's so much more to memory than setting XMP.....

[terminalinfinity]

1 hour ago, ShrimpBrime said:

For certain! 

 

I just wonder if anyone has played around in the sub timings yet for a resolve. There's so much more to memory than setting XMP.....

I messed around a little bit.  I think I might do some more testing along w/ decoupling infinity fabric to see if the rumors of that helping have any merit.

[terminalinfinity]

Alright, it's preliminary but it definitely seems like it has something to do with the way the infinity fabric is handled in Zen 3/New BIOS

 

I ran some very loose baseline timings at 3600 (16-20-20-20-40), WHEA errors immediately.

 

Uncoupled the infinity fabric, no errors.  Perfect stability.

 

Going to do some more experimentation, see if I can feed more voltage to the fabric to make it stable

Also keep in mind that AMD specifically mentioned infinity fabric stability for the new AGESA update coming down the pipeline.

[Eajimoba]

22 minutes ago, terminalinfinity said:

Alright, it's preliminary but it definitely seems like it has something to do with the way the infinity fabric is handled in Zen 3/New BIOS

 

I ran some very loose baseline timings at 3600 (16-20-20-20-40), WHEA errors immediately.

 

Uncoupled the infinity fabric, no errors.  Perfect stability.

 

Going to do some more experimentation, see if I can feed more voltage to the fabric to make it stable

Also keep in mind that AMD specifically mentioned infinity fabric stability for the new AGESA update coming down the pipeline.

Isn't your testing a moot point?

 

There really is no benefit to go beyond 3200Mhz if the FCLK can't keep up right? Or are you doing testing just to try and diagnose where the specific problems come from?

Not being argumentative just curious as im learning things as I go as well

[terminalinfinity]

14 minutes ago, Eajimoba said:

Isn't your testing a moot point?

 

There really is no benefit to go beyond 3200Mhz if the FCLK can't keep up right? Or are you doing testing just to try and diagnose where the specific problems come from?

Not being argumentative just curious as im learning things as I go as well

Im trying to figure out what the problem is so I can figure out if it 1) sounds like its being addressed (it is) and 2) if it can be worked around

 

Also you can overcome the latency hit from desyncing clocks with high enough ram frequency.

[terminalinfinity]

With that being said, none of the voltages I tweaked seemed to reduce or increase WHEA errors.  I would put this squarely as a BIOS issue that has to be rectified through a patch.  I highly doubt there's a genuine massive range of infinity fabric quality on zen 3 dropping all the way down to not being able to do 1600.

 

Also, if it genuinely was a infinity fabric quality issue, voltages would have some effect on the error rate in all likelihood.

[Eajimoba]

14 minutes ago, terminalinfinity said:

Im trying to figure out what the problem is so I can figure out if it 1) sounds like its being addressed (it is) and 2) if it can be worked around

 

Also you can overcome the latency hit by desyncing clocks with high enough ram frequency.

 

I never knew about this..

 

Once the XMP issues through bios factors are fixed, my cl14 3600Mhz with no no additional overclock as FCLK at 1933 to be most effective?

 

I thought it needed to be the same for that 1:1 efficiency

 

Edit: Oh and 1800 FCLK as well I see. Wait im so confused now

[terminalinfinity]

12 minutes ago, Eajimoba said:

 

I never knew about this..

 

Once the XMP issues through bios factors are fixed, my cl14 3600Mhz with no no additional overclock as FCLK at 1933 to be most effective?

 

I thought it needed to be the same for that 1:1 efficiency

I think you're reading the graph wrong.  You don't increase FCLK if its 1:1 w/ your ram.  This is showing latency at certain FCLK frequency.  You should only think about decoupling if your RAM can handle some very high frequencies your FCLK cant handle.

 

Unsyncing clocks gives you a latency penalty

Higher MEMORY (MCLK) frequencies reduces latency

Thus there is a point that you can offset an unsyncing penalty w/ higher memory frequency.  Generally the rule is about 300MHz above the desync point while keeping the same CAS latency.  (So if you desync at 3800 leaving Fclk at 1900, you need to get in the range of 4066-4100 MCLK to offset the hit and get back to where you were latency wise)

 

That last part is a bitch though.  Most Infinity fabric can do 1900.  (3800 MHz Mclk)  So you'd need 4100MHz or above memory to see any gains if your FCLK cant go any higher, WITHOUT increasing CAS latency.
 

There are very few that can do 4100+ CAS 16. Even the best Samsung B-die kits are going to have trouble.  I can manage just above 4000 at CAS 16 before I hit a brick wall and have to move to CAS 17/18

[Eajimoba]

4 minutes ago, terminalinfinity said:

I think you're reading the graph wrong.  You don't increase FCLK if its 1:1 w/ your ram.  This is showing latency at certain FCLK frequency.  You should only think about decoupling if your RAM can handle some very high frequencies your FCLK cant handle.

 

Unsyncing clocks gives you a latency penalty

Higher MEMORY (MCLK) frequencies reduces latency

Thus there is a point that you can offset an unsyncing penalty w/ higher memory frequency.  Generally the rule is about 300MHz above the desync point while keeping the same CAS latency.  (So if you desync at 3800 leaving Fclk at 1900, you need to get in the range of 4066-4100 MCLK to offset the hit and get back to where you were latency wise)
 

Ok so generally speaking if I can hit 1:1 mCLK and FCLK I don't need to worry about much else in terms of IF tuning? [ram timings aside]

Or would it be better to see how far my FCLK can go based on silicon lottery and try and OC my mCLK to the optimal number? [300Mhz more than FCLK]?

[terminalinfinity]

Just now, Eajimoba said:

 

Generally as a rule, you should push your ram as far as it can go where the FCLK can still keep up.  But NEVER raise your FCLK above MCLK.

[terminalinfinity]

2 minutes ago, Eajimoba said:

 

Also I made some edits to that post to give some examples.

[Eajimoba]

1 minute ago, terminalinfinity said:

Generally as a rule, you should push your ram as far as it can go where the FCLK can still keep up.  But NEVER raise your FCLK above MCLK.

 

Thanks! Let's hope for this BIOS update to allow 2000Mhz FCLK

My ram is supposedly getting really good OC so I'm hoping for some nice clocks/timings.

 

[Guest]

5 hours ago, terminalinfinity said:

 

 

Uncoupled the infinity fabric, no errors.  Perfect stability.

 

 

There you go then. 

 

Good testing. 

 

What performance loss? 1%? 2 maybe?

[terminalinfinity]

25 minutes ago, ShrimpBrime said:

There you go then. 

 

Good testing. 

 

What performance loss? 1%? 2 maybe?

I didn't do performance testing, I was just trying to nail down the reason.  Given that it's already been announced that part of the next AGESA is trying to get the infinity fabric stable on Zen 3 at speeds up to 2000MHz, I'm going to hold out on the next BIOS before doing any tuning, since it's likely I'd need to start over anyway and it won't be until I get one of the new GPUs that I can take full advantage of it anyway

[Guest]

3 hours ago, terminalinfinity said:

I didn't do performance testing, I was just trying to nail down the reason.  Given that it's already been announced that part of the next AGESA is trying to get the infinity fabric stable on Zen 3 at speeds up to 2000MHz, I'm going to hold out on the next BIOS before doing any tuning, since it's likely I'd need to start over anyway and it won't be until I get one of the new GPUs that I can take full advantage of it anyway

What I meant is you did something. Most everyone else has a complaint and that's about as far as it goes. 

[Eajimoba]

9 hours ago, ShrimpBrime said:

What I meant is you did something. Most everyone else has a complaint and that's about as far as it goes. 

 

https://forum-en.msi.com/index.php?threads/msi-x570-b550-beta-bios-update.348919/

 

They released new BETA bios that fixed the FCLK instability.

[Guest]

48 minutes ago, Eajimoba said:

 

https://forum-en.msi.com/index.php?threads/msi-x570-b550-beta-bios-update.348919/

 

They released new BETA bios that fixed the FCLK instability.

Beta is testing phase of the bios pre-release of full updated "fixed" version. They are collecting data.

 

And your link doesn't work by the way.

[Eajimoba]

1 hour ago, ShrimpBrime said:

Beta is testing phase of the bios pre-release of full updated "fixed" version. They are collecting data.

 

And your link doesn't work by the way.

 

Thanks for defining BETA for the users here.

 

The site is really slow atm, don't know why. In any case for those with x570 or b550 boards go to the link.

 

I am able to use my XMP profile 3600mhz CL14 as opposed to being limited to 3200mhz previously

[Guest]

1 hour ago, Eajimoba said:

 

Thanks for defining BETA for the users here.

 

The site is really slow atm, don't know why. In any case for those with x570 or b550 boards go to the link.

 

I am able to use my XMP profile 3600mhz CL14 as opposed to being limited to 3200mhz previously

You where limited by frequency, or just getting errors at that frequency? 

[Guest]

This thread is a good read, but with a 3800X.

https://www.overclock.net/threads/ryzen-3800x-whea-logger-event-19.1774309/

 

Hoping for 5000 resolves soon for you guys though. Soon enough I gather. Problem is computing is 50% HW and 50% OS. 

What I mean by that, is I'm trying to determine if the error reporting is just being picky while the WHEA errors are not really showing me people actually have an issue with stability. 

 

[Eajimoba]

2 hours ago, ShrimpBrime said:

You where limited by frequency, or just getting errors at that frequency? 

Manual overclocks and XMP profiles past 3200Mhz resulted into no boot, I couldn't even get into windows to check on the WHEA errors  

 

At 3200Mhz and below there were no issues, had looping 3DMark Timespy tests back to back for like 2 hours.

 

Now at my 3600Mhz CL14 [Manually tuned off of the XMP profile] I'm not seeing any Whea errors, it's weird though how some were getting them and some weren't

[Guest]

5 hours ago, Eajimoba said:

Manual overclocks and XMP profiles past 3200Mhz resulted into no boot, I couldn't even get into windows to check on the WHEA errors  

 

At 3200Mhz and below there were no issues, had looping 3DMark Timespy tests back to back for like 2 hours.

 

Now at my 3600Mhz CL14 [Manually tuned off of the XMP profile] I'm not seeing any Whea errors, it's weird though how some were getting them and some weren't

Not all hardware is the same even though manufactured the same. 

Looks better at 3600 CL14 with no errors. Good.

[terminalinfinity]

BREAKTHROUGH!  I managed to push past 1600MHz IF by DROPPING voltages.  A lot.  Doing some more testing, will post results when I have parameters confirmed

[terminalinfinity]

Im up to 1800MHz stable.  Here is what I've discovered so far:

1. The issues with infinity fabric stability in this BIOS will show themselves irrespective of your memory frequency and timing.  Thus, you can leave your memory on 2133 at loose settings to eliminate any chance of a random memory error from timings mucking up things.

2. The normal rules regarding voltage with this new setup seem to be changed a little bit.  In addition to voltage too low making things unstable, it now seems voltage too high, even something as minor as a 1.1 SOC voltage when it really wants 1 volt exactly might make things unstable.  You're now looking for a stable "range" rather than a stable minimum with all SOC related voltages including CLDO VDDGs (Infinity fabric voltages)

3. These are my current voltages at 1800MHz IF stable.  Some MAJOR differences from Zen 2.

Edit: I'm REALLY zeroing in on the fact that the default SOC value for many XMP profiles is 1.1.  Zen 3 REALLY doesn't seem to like 1.1 SOC voltage unless the frequency demands it.  In addition this is the jumping off point for many people when manually OCing.   I'm really thinking this is causing all the headaches right here.

[terminalinfinity]

1900MHz Stable Infinity Fabric.  Voltages are very beta and I haven't tried reducing.  My theory on voltage "ranges" is looking more confirmed.  Will try for 2000 FCLK soon.