AMD Ryzen 5000 Builders Thread

[MrTeal]

X570-P is out as well. Have you tried the new BIOS yet?

 

[repoman0]

PRIME X570-PRO BIOS & FIRMWARE | Motherboards | ASUS Canada

AMD AM4 motherboard with PCIe Gen 4 for ultra-high transfer speed, and ASUS OptiMem for improved memory overclocking

www.asus.com

Weird, I was looking at the same page before and now it’s there. Anyway thanks, will try later. I rolled back from 3402 to 3202 because the beta gave me lots of WHEA errors at 1800 fclk at the same settings.

 

[Kenmitch]

Testing out the MSI x570 ComboAM4PIV2 1.2.0.0 uEFI for my MB to see how it'll work out in the end.

Like the last uEFI version I'm unable to boot up at 3800/1900 which is what I was after. I can boot lower and higher, but 3800/1900 is a no go for some reason or another.

Yesterday I did some experimenting with 3866/1933 with some really loose timings to see if it would generate WHEA errors while running the Aida64 stability test.

Very loose, but the results look to give some hope at least. VDDG IOD voltage looked to eliminate the WHEA errors on my rig at least. I had to slowly bump it up till I got no WHEA errors with the loose timings unchanged.



I started tightening up the timings today to see if I can make 3866/1933 work without WHEA errors and am currently using the Aida64 stability test to see if it generates any. WHEA errors usually come pretty quick so I'm only going to run the stability test for 1/2hr or so before tightening more timings. I haven't made a whole lot of progress so far, but it's looking like 1933 FCLK is doable on this uEFI version.

 

[repoman0]

Just updated to 1.2.0.0 final on my Prime X570 Pro. Old IF clocks and voltages again generated lots of WHEA errors in HWInfo and Event Viewer, same as the beta. I had it set for 1800 MHz Fclk and DRAM clock, 1.025V SoC, 0.975V VDDG IOD and CCD, 0.95V VDDP. It's (maybe?) stable now at 1.1V SoC, 1.025V VDDG IOD/SOC, 0.9V VDDP. I'm starting to suspect that maybe the 1.1.9.0 beta just didn't have the capability to report its errors to Windows and it was just correcting errors silently that whole time. Who knows though, going to stick with this for now and see if it generates errors eventually.



Updated so I remember: reduced VSoC to 1.05, tried VDDG CCD 0.95 / IOD 1.025 but didn't work, raised to 1.0 / 1.025 and working so far. Did not work with both VDDGs at 1.0V ('Auto' now results in 1.0 for both ... it was definitely 0.95 in earlier releases, so ASUS knows that these need to be messed with to make things stable with Zen 3). No errors in 10 min which is the best I've done since 1.1.9.0 which probably just ignored corrected bus errors.

 

[Makaveli]

Just updated to 1.2.0.0 final on my Prime X570 Pro. Old IF clocks and voltages again generated lots of WHEA errors in HWInfo and Event Viewer, same as the beta. I had it set for 1800 MHz Fclk and DRAM clock, 1.025V SoC, 0.975V VDDG IOD and CCD, 0.95V VDDP. It's (maybe?) stable now at 1.1V SoC, 1.025V VDDG IOD/SOC, 0.9V VDDP. I'm starting to suspect that maybe the 1.1.9.0 beta just didn't have the capability to report its errors to Windows and it was just correcting errors silently that whole time. Who knows though, going to stick with this for now and see if it generates errors eventually.

View attachment 39266

Updated so I remember: reduced VSoC to 1.05, tried VDDG CCD 0.95 / IOD 1.025 but didn't work, raised to 1.0 / 1.025 and working so far. Did not work with both VDDGs at 1.0V ('Auto' now results in 1.0 for both ... it was definitely 0.95 in earlier releases, so ASUS knows that these need to be messed with to make things stable with Zen 3). No errors in 10 min which is the best I've done since 1.1.9.0 which probably just ignored corrected bus errors.

Thanks for the mini review this will be helpful for me when I update the bios tomorrow and start my own testing.

 

[repoman0]

No errors overnight with both CCD and IOD at 1.025. My chip definitely seems to need both, where I see many getting away with just IOD. I also remembered that I ran for a day or two with them set at 0.95V and eventually rebooted with an uncorrectable bus error. Don’t know what ECC algorithm they’re using but most that I’m familiar with should have expected values of thousands of correctable errors per uncorrectable one. Since I never saw any correctable ones the earlier 3202 bios has to be broken. Wonder if that’s the case on all the asus boards ...

 

[Kenmitch]

MSI x570 ComboAM4PIV2 1.2.0.0 uEFI

Well it looks like this uEFI might just be the one that allows greater then 1800 FCLK without WHEA errors. I'm kind of puzzled why it won't boot up at 3800/1900 for some reason or another. I had the same issue with the last uEFI version, but in the past the MB didn't have any issues with 3800/1900 other then generating WHEA errors.

So far so good with 3866/1933 as far as generating WHEA errors. It's looking like I have to give up on the 3800/1900 goal and settle for 3866/1933 which is a good thing in the end.



I should be able to at least drop down to CL14 as I've tested and ran 3866 CL14 in the past.

My b-die is good for higher clocks which I haven't tried yet. Most likely WHEA errors will eliminate the possibility of higher clocks, but I'll try and see later just to see if so.

 

[Noid]

moved ...

 

[Makaveli]

Anyone trued the " New " CTR 2.0 RC3 ?

Just ran the diagnostics ---- with a hard reset after 5 mins
( never completed )
Passed it's own 10 min stability test at it's selected 4350Ghz
( boost mode )

And he says it's " safe " ? lol

Oddly ---- it's says I have a " bronze" sample.
( maybe a indicator of it's underclocking ability ? if so, Ok. I believe that )

The analysis is questionable.
Afterall, I have not see anyone yet to surpass my results.

Question - Zen 3 and CTR

I've just installed this and been playing with it a little bit. What are you thoughts on CTR and have you used it? The Diagnostic game me this so far. I've not done any of the tuning yet and curve optimizer is also turned off. The only thing on is PBO2 + 200Mhz...

forums.anandtech.com

 

[Kenmitch]

If only the FCLK was as strong as my b-die!

I get no WHEA errors, but the latency and bandwidth are suffering from being out of sync. 1933 FCLK is as high as I can go without WHEA errors.

 

[Noid]

It's your UCLK ---- it's at 1/2
Thats the weird divider thing I was talking about.

You are the only one that I know of that got this to almost " stick "
( but your not at DDR - you need 2200 for UCLK )

 

[Kenmitch]

It's your UCLK ---- it's at 1/2
Thats the weird divider thing I was talking about.

You are the only one that I know of that got this to almost " stick "
( but your not at DDR - you need 2200 for UCLK )

UCLK = MCLK is set in the uEFI, but it always defaults to 1/2 speed in the end. It was at least fun playing around with. A higher FCLK might offset some of the penalty, but the WHEA errors pop up beyond 1933 FCLK.

I'm just going to drop down to CL14 @3866MHz/1933 and call it a day most likely.

 

[HK-47]

Just finished updating my rig from a from an old Rampage IV 1680v2 (4.6GHz) to the Asus Dark HERO VIII with a 5800x...Wow what a difference lol! When I first booted up at stock setting I was getting temps in the low 80's using some crappy thermal paste I had laying around. When I would enable the AMD PBO CO with -15 and 100 boost I would hit 90c in CB R20 MT. So I ordered some Thermal Grizzly Kryonaut and reset my Corsair 150i cooler. I guess I had done a bad job with the other paste. After reapply the new paste my temps were in the low 70s at stock running CB R20 MT.

My current settings as of now:

ram 3800/1900flck 16-19-16-36-60 1T 1.38v DOCP is 3600
MB setting all auto besides amd advanced PBO

AMD PBO setting
-30 all cores
limits = motherboard
boost = 100

going beyond 100 at - 30 and my system becomes unstable in idle states and this also hits 80c where the 5800x stops increasing boost in MT loads. I may try to tune some per core to see if I can get idle states stable at 125 boost. But I'm not sure the time is worth the reward. What I have now seems to be working good and temps are at 80c in R20 MT and sustain 4.68-4.7ghz.

 

[Makaveli]

Just finished updating my rig from a from an old Rampage IV 1680v2 (4.6GHz) to the Asus Dark HERO VIII with a 5800x...Wow what a difference lol! When I first booted up at stock setting I was getting temps in the low 80's using some crappy thermal paste I had laying around. When I would enable the AMD PBO CO with -15 and 100 boost I would hit 90c in CB R20 MT. So I ordered some Thermal Grizzly Kryonaut and reset my Corsair 150i cooler. I guess I had done a bad job with the other paste. After reapply the new paste my temps were in the low 70s at stock running CB R20 MT.

My current settings as of now:

ram 3800/1900flck 16-19-16-36-60 1T 1.38v DOCP is 3600
MB setting all auto besides amd advanced PBO

AMD PBO setting
-30 all cores
limits = motherboard
boost = 100

going beyond 100 at - 30 and my system becomes unstable in idle states and this also hits 80c where the 5800x stops increasing boost in MT loads. I may try to tune some per core to see if I can get idle states stable at 125 boost. But I'm not sure the time is worth the reward. What I have now seems to be working good and temps are at 80c in R20 MT and sustain 4.68-4.7ghz.

What software are you using to check temps?

I still find ryzen master gives me temps 10c over other apps.

Using these settings for PB02

 

[HK-47]

Im using Ryzen Master and HWinfo. I haven't seen much difference in temp reading. My motherboard QLed shows CPU package temp which is about 10c under the ccx temp of Ryzen master. The apps you are using are probably looking at the package temp.

 

[Makaveli]

Im using Ryzen Master and HWinfo. I haven't seen much difference in temp reading. My motherboard QLed shows CPU package temp which is about 10c under the ccx temp of Ryzen master. The apps you are using are probably looking at the package temp.

I updated the Ryzen master app version and now both HWinfo and it show the same temps.

I hit 90c on the Cinibench also but the Aida64 stress test with AVX on will only hit 80c

At full load I usually set the H150i to the extreme profile for pump and fan settings.

 

[HK-47]

I was getting 90c on R20 MT prior to redoing my paste and being more careful applying it with my H150i. Using thermal grizzly now im at 80c in R20MT. In Aida 64 stress test im in the lower 70s.

 

[Makaveli]

I was getting 90c on R20 MT prior to redoing my paste and being more careful applying it with my H150i. Using thermal grizzly now im at 80c in R20MT. In Aida 64 stress test im in the lower 70s.

I did remount my cooler and reapplied some Artic Silver 5 but not seeing a reduction in temps in cinibench.

What are the highest single core boost speeds you are seeing with current settings?

And which version of the H150i i'm on the Gen 1 model.

 

[HK-47]

Its a H150i Pro. I'm not sure what revision it is. I got about 4 months ago so I would think its a newer one. DO you know how to tell?

Here are some screen shots of HWinfo showing my single and MT.

 

[Makaveli]

Its a H150i Pro. I'm not sure what revision it is. I got about 4 months ago so I would think its a newer one. DO you know how to tell?

Here are some screen shots of HWinfo showing my single and MT.View attachment 39338View attachment 39339

I'm on the Pro RGB model which uses a Asetek pump and only 1600 RPM fans all models after that are using a CoolIT pump and 2400 RPM fans. Which also makes them louder when you crank the fans speeds.

Pro RGB 10 - 25 dBA
Pro RGB XT 10 - 37 dBA



And since you said you are only using a 100 Mhz for PB02 your single core boost of 4.95Ghz is about right.

i'm using 200 so I boost to 5025 to 5050 Mhz on single core.

 

[HK-47]

I must have an older model then my original fan where 1600 rpm. I got it as a renewed item from corsair on amazon LOL.

I found there is a fine line for Curve vs boost.
To much boost and to much negative curve you are unstable at idle.
to much boost and not not enough curve and you are thermal throttled.
I can reduced my curve to say -15 and up my boost to 200 get better single core scores but then multithread performance drops from hitting 80C not allowing higher boost levels and clock stretching. Its kind of like robbing Peter to pay Paul lol. This is why Asus Dark Hero has the dynamic switching feature. You can tune your PBO curve for single thread then tune a manual OC for MT work loads.

This board and CPU are new to me, I just got it last week. Now that I'm starting to figure out how this PBO curve stuff works next is to tune PBO curve for single thread and setup Dynamic OC for MT workloads. As I play around with stuff I will post my finding.

 

[Makaveli]

I must have an older model then my original fan where 1600 rpm. I got it as a renewed item from corsair on amazon LOL.

I found there is a fine line for Curve vs boost.
To much boost and to much negative curve you are unstable at idle.
to much boost and not not enough curve and you are thermal throttled.
I can reduced my curve to say -15 and up my boost to 200 get better single core scores but then multithread performance drops from hitting 80C not allowing higher boost levels and clock stretching. Its kind of like robbing Peter to pay Paul lol. This is why Asus Dark Hero has the dynamic switching feature. You can tune your PBO curve for single thread then tune a manual OC for MT work loads.

This board and CPU are new to me, I just got it last week. Now that I'm starting to figure out how this PBO curve stuff works next is to tune PBO curve for single thread and setup Dynamic OC for MT workloads. As I play around with stuff I will post my finding.

Ryzen Master has Core 04 as my best core and Core 01 as second fastest.



I have CO set for -15 on the second core and -17 on the best core while rest of the cores are at -10. I'm still testing this to see what is optimal.



The Dark Hero board has some extra features that will certainly help, and I would hope so considering its price.

 

[HK-47]

I started with just going all cores to - 30 with 0 mhz boost overide. And started working up the boost until I got unstable then turned the boost down 1 level. So im at 100mhz -30, 125mhz -30 would sometimes crash at idle loads. -15 and 200mhz I can do which gets good single core scores but lowers MT scores. I will do some testing later tonight and see how high I can get my single thread scores. I need to do it anyways for my dynamic OC switcher profile I have save in bios.

 

[EXCellR8]

Got to finally fiddle with my setup a little bit; my 5600X is 'silver sample' according to CRT... feels so runner up

Always a bridesmaid, never a bride.

 

[Justinus]

Got to finally fiddle with my setup a little bit; my 5600X is 'silver sample' according to CRT... feels so runner up

Always a bridesmaid, never a bride.

How do you determine that with it?

Edit: Nevermind. Found the igors lab tutorial.

Interesting little test. Doesn't feel like I got a golden sample since 4.7 GHz is barely attainable within temp limits on my custom loop.

Then again I have extremely stringent stability standards, so it may be throwing a wrench into it.