Ryzen Owners Zen Garden

[tabascosauz]

Still getting the random WHEA. What a shame. Right now I'm on AGESA 1.2.0.0 and I reset the BIOS to only run XMP at 3600/1800 and nothing else.

View attachment 187378

At least that means it probably wasn't an overclocking instability issue. Not like WHEAs are a huge deal for everyday tasks, but it'd just be so nice to set up important online interviews without having to worry about random crashes. Lately I've been using my phone for Zoom correspondence lol

Tried disabling C-states to rule out insufficient idle voltage? I think on Asus BIOS there's only DF Cstates for the Infinity Fabric, not sure if the cores have their own standalone setting. Global C-state Control to kill two birds with one stone, can find them all easily using the search function.

 

[8BitZ80]

Tried disabling C-states to rule out insufficient idle voltage? I think on Asus BIOS there's only DF Cstates for the Infinity Fabric, not sure if the cores have their own standalone setting. Global C-state Control to kill two birds with one stone, can find them all easily using the search function.

Yeah it turns out that under Advanced and AMD CBS options there's a Global C-State Control that I disabled just now. I'll clear the log again and see if that resolves it. Sounds like it might, because often the WHEA errors are recorded when the system is powered up from sleep mode.

 

[tabascosauz]

Turns out there are some gains to be had for Renoir on AGESA 1200. ~0.5ns in AIDA, better Copy, ~2sec in membench, ~100pts memory in Geekbench. And it's doing the bullshit L3 read speed less often now. No gains on Matisse so far.

 

[8BitZ80]

View attachment 187371

Just checked my 5600x in CTR, and it looks to be unique

Check yours, guys.

As far as I've read, these are just cut down 5900X/5950X (depending on what CPU you have) with one CCD disabled, and in no way sharing cores between CCD's

What does the CCX layout show up as? Makes me also wonder if the amount of cache is any different.

These were my screenshots:

 

[R00kie]

What does the CCX layout show up as? Makes me also wonder if the amount of cache is any different.

These were my screenshots:

View attachment 187467

View attachment 187466

Same as yours.
Since the second CCD just sits there, it wont come up on anything other than that piece of text at the bottom.
The cache amount is still the same

 

[Mussels]

I wonder if that makes yours easier to cool, even if the cores arent active they are increasing your contact area to the heatsink

 

[biffzinker]

I wonder if that makes yours easier to cool, even if the cores arent active they are increasing your contact area to the heatsink

I don’t imagine there would be any difference unless there was 4/3 cores active on the other CCD die. The chiplets are spaced apart.

 

[Mussels]

I don’t imagine there would be any difference unless there was 4/3 cores active on the other CCD die. The chiplets are spaced apart.

heats gotta travel back into the PCB and up via the inactive die, its not being produced there bure surely somes gotta pass that way

about equal to slapping a fan on the back of your motherboard, SOME heat goes that way

 

[8BitZ80]

It looks like any Zen 3 processor can max out an RTX 3090 at 1080p. We probably won't start seeing major CPU bottlenecks on the 5600X until the next generation of graphics cards.

 

[HD64G]

Some findings from my system's performance testing with SAM feature on and off: https://www.techpowerup.com/forums/threads/amd-radeon-smart-access-memory.274803/post-4456466

 

[8BitZ80]

Tried disabling C-states to rule out insufficient idle voltage? I think on Asus BIOS there's only DF Cstates for the Infinity Fabric, not sure if the cores have their own standalone setting. Global C-state Control to kill two birds with one stone, can find them all easily using the search function.

The WHEA errors seem to be finally gone. I think it's been a week and I haven't seen any in the logs. So yeah, disabling global C-states did the trick! Thanks for the tip.

 

[R00kie]

The WHEA errors seem to be finally gone. I think it's been a week and I haven't seen any in the logs. So yeah, disabling global C-states did the trick! Thanks for the tip.

might also be worth updating your BIOS, Asus have released their non beta BIOS'es just a couple of days ago, the beta's weren't very stable for me, same WHEA errors, but so far so good.
They do mention improved system stability.

 

[tabascosauz]

The WHEA errors seem to be finally gone. I think it's been a week and I haven't seen any in the logs. So yeah, disabling global C-states did the trick! Thanks for the tip.

might also be worth updating your BIOS, Asus have released their non beta BIOS'es just a couple of days ago, the beta's weren't very stable for me, same WHEA errors, but so far so good.
They do mention improved system stability.

Good to hear. If previously the WHEA errors were rampant and on a daily basis, I'd say you probably fixed it for now.

There's *some* merit to fixing stability through AGESA, but I'd go as far as to say that the lower quality or degraded chips are unlikely to be swayed, and more likely to require permanently disabled C-states to function properly, as it's unlikely that a vendor will increase default idle voltage across the board to accommodate that lowest common denominator of silicon.

I didn't used to need to touch C-states, but my 3700X from launch is at the point where it needs DF C-states permanently disabled or it'll throw WHEA errors, reboot, or occasionally refuse to POST. I've never gone beyond running 3733 daily, I've never all-core OC'd beyond 4.0@1.195V. It's the same deal on my B550M TUF and B550I AX.

Doesn't instill much confidence in the longevity of current Ryzen. Not even so much the 7nm chiplets, as the I/O die and substrate linkages.

 

[8BitZ80]

might also be worth updating your BIOS, Asus have released their non beta BIOS'es just a couple of days ago, the beta's weren't very stable for me, same WHEA errors, but so far so good.
They do mention improved system stability.

I didn't have any stability issues on the 1801 beta bios (which Asus have subsequently deleted....I wish they wouldn't do that in case I wanted to flash back to it later). But I've been waiting for a non-beta bios for two months now, feels like forever lol I updated to the 1804 non-beta and it seems to be running great.

Good to hear. If previously the WHEA errors were rampant and on a daily basis, I'd say you probably fixed it for now.

There's *some* merit to fixing stability through AGESA, but I'd go as far as to say that the lower quality or degraded chips are unlikely to be swayed, and more likely to require permanently disabled C-states to function properly, as it's unlikely that a vendor will increase default idle voltage across the board to accommodate that lowest common denominator of silicon.

I didn't used to need to touch C-states, but my 3700X from launch is at the point where it needs DF C-states permanently disabled or it'll throw WHEA errors, reboot, or occasionally refuse to POST. I've never gone beyond running 3733 daily, I've never all-core OC'd beyond 4.0@1.195V. It's the same deal on my B550M TUF and B550I AX.

Doesn't instill much confidence in the longevity of current Ryzen. Not even so much the 7nm chiplets, as the I/O die and substrate linkages.

Man that would suck if they degrade over time. It makes a good case for undervolting though, especially since overclocking seems to be hardly worth it anymore. The stock voltage on my 5600X goes up past 1.4V and I feel like that's way too much.

 

[biffzinker]

The stock voltage on my 5600X goes up past 1.4V and I feel like that's way too much.

It only does that though during a single/dual core boosting.

 

[freeagent]

Just curious.. at stock with PBO enabled, 3 of my 6 cores boost up to 4625MHz. But when I try to manually set 4600 in the bios, and go to load up my trusty testing app and hit go, the pc will reboot after a few seconds, its like I crossed a magical current domain and it clips and reboots. Voltage doesn't help, LLC doesn't help. I do only have 1 CPU 8 pin installed, I cant use the other 8 pin because it wont separate to four pin, and I am thinking the CPU is not getting enough juice.. Does it sound like I am on the right track? The CPU will load windows at 4700 all core manual.. so its not like a frequency wall.

 

[tabascosauz]

Just curious.. at stock with PBO enabled, 3 of my 6 cores boost up to 4625MHz. But when I try to manually set 4600 in the bios, and go to load up my trusty testing app and hit go, the pc will reboot after a few seconds, its like I crossed a magical current domain and it clips and reboots. Voltage doesn't help, LLC doesn't help. I do only have 1 CPU 8 pin installed, I cant use the other 8 pin because it wont separate to four pin, and I am thinking the CPU is not getting enough juice.. Does it sound like I am on the right track? The CPU will load windows at 4700 all core manual.. so its not like a frequency wall.

No, it's most definitely a frequency/voltage wall...lol. That's exactly what a wall looks like.

I can "run" 4400MHz on my 4650G all day long if booting into Windows is a benchmark for stability......it literally isn't possible to run any actual CPU stress test (no, Cinebench is not a stress test) at those all-core speeds. If degradation wasn't a thing, I could overvolt it to the moon if I wanted to. It will not be stable even for 1 second.

Booting into Windows really doesn't take much more stability than booting. A bit more so for RAM overclocks, but CPU overclocks not so much.

Where are you seeing 4625MHz? If it's HWInfo, go to settings (right click taskbar icon) and enable Snapshot CPU Polling, then see what kind of speeds it's achieving after you've used your comp for a few hours. 4625MHz literally means nothing unless you're running a single-threaded benchmark and the loaded core shows 4625MHz continuously under Effective Clock. "Core Clock" may as well be smoke and mirrors due to the granularity of Ryzen clock control.

 

[freeagent]

Ok that makes sense. Yeah the 4625 is just single core stuff I guess. Under the hardest loads at stock it will hover between 4275-4300.

That snip is regular use.

 

[tabascosauz]

Ok that makes sense. Yeah the 4625 is just single core stuff I guess. Under the hardest loads at stock it will hover between 4275-4300.

That snip is regular use.

View attachment 188416

If you're aware of the CPPC core quality order already, you can pretty much hide all the Core Clock and Core Ratio fields and you won't be missing out on any useful information (unless you check Snapshot Polling like I said, makes it a bit more accurate).

If you're not hitting close to 4.6GHz in Effective Clock in any core during your gaming or benchmarking, then 4.6GHz is meaningless for actual performance. If you regularly hit about 4.55GHz or higher effective, then the 4.625GHz might actually not be too far off.

 

[freeagent]

Aida shows 4475 in most of its boxes, its pretty close.. I get a faster pi time at stock lol.. I think..

Ill try it out for sure, but these are the stock numbers I have seen through the past 4 bios updates, vcore is under control now and doesn't spike up to 1.5+v on individual cores anymore.

 

[tabascosauz]

Aida shows 4475 in most of its boxes, its pretty close.. I get a faster pi time at stock lol.. I think..

Ill try it out for sure, but these are the stock numbers I have seen through the past 4 bios updates, vcore is under control now and doesn't spike up to 1.5+v on individual cores anymore.

Funny you mention that, that's basically been my experience with PBO since 2019. Even the profiles I make that are verifiably faster in benchmarks have questionable performance in actual games. COD:MW gets noticeably more stuttery with any PBO settings. Before you ask, I have a MP custom game preset made just for testing this over the past year.

And so I've been running a stock 3700X for quite some time. Smooth as a baby's bottom, even though the clocks might not look as good.

 

[freeagent]

Oh yeah man. I am kicking myself a little bit for not stretching to a 3700x but I honestly thought I would be rockin zen 3 now.. I only got the XT because it was the same price as the vanilla, and I had just watched the XT vs 10600K video from GN haha. No biggie.. not as much horsepower as the bigger guys but it’s still pretty sweet. I am running the newest bios for my board and I am really happy with things. It’s nice to see a platform mature a little. Now I’m just waiting to buy a GPU as the CPU isn’t really a high priority anymore..

 

[tabascosauz]

Oh yeah man. I am kicking myself a little bit for not stretching to a 3700x but I honestly thought I would be rockin zen 3 now.. I only got the XT because it was the same price as the vanilla, and I had just watched the XT vs 10600K video from GN haha. No biggie.. not as much horsepower as the bigger guys but it’s still pretty sweet. I am running the newest bios for my board and I am really happy with things. It’s nice to see a platform mature a little. Now I’m just waiting to buy a GPU as the CPU isn’t really a high priority anymore..

Didn't know you were in Manitoba.

I think you made the right choice haha. I happened across a Memory Express flyer about 2 weeks ago, $469 3700X. I paid $459 in August 2019......for what was a hot new product at the time...amazing "pricing" in 2021

And then there's my vaporware 5900X on order since November...from how ME and CC are going I don't expect to see it within the first three quarters of this year. Dunno whether to buy a "reasonably" priced scalper 5600X, tough it out and keep waiting on 5900X, or just stick with the ailing 3700X and buy a M27Q monitor instead.

I wasn't feeling so great about my GPU purchase 16 months ago (thought it was a waste of money, and immediately had to RMA it once for coil whine), but with each passing month the absolute comedy show that is GPU availability in Canada is making me feel better and better about my 2060S FE.

 

[8BitZ80]

It only does that though during a single/dual core boosting.

True, but I just dial it down a bit anyway. I'm testing out a negative 30 curve optimizer on all cores and stability seems good. Now it's around 1.32V single core and 1.125V multi core.

 

[Mussels]

True, but I just dial it down a bit anyway. I'm testing out a negative 30 curve optimizer on all cores and stability seems good. Now it's around 1.32V single core and 1.125V multi core.

jelly, some of mine only do -5

i got the first 3 done and then we had 35C ambients so i gave up on anything that made more heat for a while