What's new
Carbonite

South Africa's Top Online Tech Classifieds!
Register a free account today to become a member! (No Under 18's)
Home of C.U.D.

Some Ampere and Ryzen anecdotes that may help someone

Status
Not open for further replies

dainluke

King Eggbert III
VIP Supporter
Rating - 100%
344   0   1
Joined
Dec 30, 2016
Messages
4,317
Reaction score
2,779
Points
9,335
Location
Stellenbosch
I'm just making this because I noticed a number of support threads based on Ryzen and Ampere, and I just wanted to collate all the fixes I've come across/tested. Some of these fixes are very intuitive and are just a result of my own error, but I listed them regardless because they're easy to make if you're not thinking about it.


AMPERE
Issue 1 (Ampere not performing properly):
I saw this with the first 3080 and first 3070 I owned. The crux of this issue seems to stem from the fact I was on a 1080p screen, but may likely be linked to the crappy PCIe 3.0 riser included with the Core P3.
Fix:
1440p mainly, but essentially by finding a way to get the game to leverage more of the core. It seems that below a threshold of demand, the core will actually become significantly under-utilized, and only begins to perform as it should when the CPU bottleneck is alleviated by either reducing the frame buffer load or by reducing CPU demand.
Onto the second statement, I found that the 3080 Gaming Z Trio, 3070 Strix, and my current 3090 all performed somewhat normally when directly in my motherboard (Z390 PCIe 3.0 anyway). Obviously, it could have been down to the initial idea of what was causing the underwhelming performance, but I still think the riser cable had a hand in it.


Issue 2 (power failure):
The first 3070 I acquired crashed hard while in CSGO and seemed to be having weird frame fluctuations.
Fix:
Turns out that one of the PCIe cables wasn't actually clipped in all the way and the issue seemed to disappear once I re-seated them.


Issue 3 (daisy chain cables):
The first time I installed my 3090, I tried to power it with the two daisy-chained PCIe power connectors because the pins on the Palit card are flipped and were forcing me to bend my cables weirdly.
Fix (me being stupid tbh):
It turns out that you absolutely need to plug in the main power connector and not the daisy-chained end.


Ryzen
Issue 1 (PC not turning on at all):
Fix:
My buddy had to use an older Zen part to update his X470 BIOS for his 5600X. It turned out that the 2400G wasn't seated properly. Thankfully he didn't bend any of the pins (not sure how but thank goodness).


Issue 2 (his PC still not outputting VGA even though the PC was now booting with the 2400G):
Fix:
He needed to use a single DIMM because the setup didn't like the other DIMM. Please keep in mind that each board has a preferred slot for single DIMM operation. He managed to update the BIOS and everything went smoothly thereafter.


Issue 3 (soul-crushing boot loops :/, XMP not working, manual memory tuning):
XMP issues are apparently not uncommon depending on the RAM kit you own and if the BIOS you're on has this bug. XMP didn't work on my X570 Tomahawk, and in Windows, my PC was BSOD'ing at 2133 which made 0 sense. This led me to have to configure VSoC, VDDG IOD, VDDG CCD, and memory voltage manually, alongside having to make an OC profile with XMP disabled (not something I was looking forward to so soon into owning Ryzen). I own a dual-rank kit, so inherently the supporting voltages are going to have to be slightly higher than standard, but there are many guides online that explain what kind of a range those voltages should be in. What your IMC/kit requires is going to be case-dependent.

Let's backtrack a little. While I was testing out different fabric clocks, memory clocks, and memory timings, I entered a value that my setup didn't like and I ended up in a boot loop. I allowed the board time to train but it just wouldn't train or output VGA. Obviously, I just drain power and clear CMOS right? Wrong. Cleared CMOS twice and even left the battery out for like 30-60 seconds. I couldn't get VGA. I ended up having to re-seat all my DIMMs and actually rearrange the order of my DIMMs. Eventually, I managed to POST.

I ended up with VSoC: 1.15V; VDDG IOD: 1.1V; VDDG CCD: 1.1V; VDDR: 1.45V alongside the profile in my sig, and some slightly tightened secondaries/tertiaries. To note: SoC voltage needs to be at least 0.04V higher than VDDG IOD and CCD. You can try 1.1 SoC and 1.05 IOD/CCD if you want to be more conservative and safe, I honestly couldn't care to test with them lowered though. Once my config passed 300% on HCI I was done for the morning, as it was 6 AM by that point 🙃.


Issue 4 (no VGA output in BIOS :D):
I decided to update to the new X570 Tomahawk BIOS because according to reports online initially (along with my own testing later), it allowed for more stable boost clocks, and seemed to alleviate even more of the USB issues. I also think it's supposed to address the XMP thing, but I couldn't be bothered to try XMP again. Well, unfortunately, after my initial boot and initial visit to BIOS to re-configure my settings (please note that on MSI, OC profiles can't be transferred between BIOS revisions = more fun 🙃), I seemed to not be able to output VGA upon trying to enter BIOS later on.
Fix:
No joke I had to use an HDMI to get into BIOS. Someone with an Aorus Master suggested this as the fix and it translated to my Tomahawk as well. Not entirely sure what the crux of this issue is, but hopefully this works for you as well. To note: I could still boot directly to Windows, and after finishing up in my BIOS with the HDMI, I swapped back to DP 1.4 again and all was well.
Alt fix:
Apparently, this dude who made a support thread and YT video exhibiting it had to re-seat his CPU to resolve this BIOS VGA output issue. He was also on a Tomahawk.


Issue 5 (USB issues...):
ROG Delta has a short USB C cable that sometimes gets tangled. I decided to unplug it, untangle it, and re-plug it. All of a sudden I see a Windows warning saying that my USB port experienced a power surge, and my headset won't turn on. I didn't think it was broken but kinda assumed this was one of those Ryzen USB bugs people have been on about. As a side note: my Commander Pro that was being powered by internal USB 2.0 also lost connection by this point 🙃.
Fix:
Literally just said fuck it and rebooted my PC. It actually worked.

Modest OC advice (curve optimizer):
If you're seeing blue screens then just reduce your negative offset. It's not the end of the world - you will find an adequate value. You may need to determine which of your cores are holding your CPU back and manually adjust the offset on each core. Also: do not go crazy on your EDC limit as it will actually reduce clocks as seen by myself, a random smaller YouTuber I came across, and according to a slew of posts online that talk about which PPT, TDC and EDC settings are optimal (not sure why and no it isn't my board's VRM).


Bonus:
Apparently HWinfo64's Corsair Link plug-in causes your Commander Pro to lose control of your fans. Guess how I found that one out.
 
Hectic!

Seems similar to the struggles here:

So 2 MSI boards, both giving endless grief. Since you have achieved all that, any ideas how to fix the issues in the thread above?
To be honest, the reason I don't often comment on support threads is that I'm never sure if my input is valid and I don't want to make up solutions that may have no impact.

If I had to take a crack at it:
From what I've seen and from what I've read online, when your Ryzen CPU is unstable it causes RAM errors and vice versa, and I would imagine that has something to do with the infinity fabric and its communication with RAM, the CCDs, and the entire IO interface on the CPU (massive guess). If he has BSODs at high CPU load but he doesn't have any manual control over his Vcore or clock speed (if the chip had degraded his board would compensate as the chip would ask for more voltage via VID. More than likely his IMC just does not like the IC on his RAM and as a result, he may need to increase VSoC and VDDG IOD to brute force stability.

In my honest opinion though, Evetech upgrade kits are problematic because the RAM they offer you is almost always a toss-up with regards to if it can be stabilized on Ryzen. I have Samsung B-Die and when XMP wouldn't load but stock 2133 caused page file errors and WHEA errors, my heart sank. I was extremely relieved once I managed to configure settings that were 24/7 stable. He's either going to have to up his SoC to 1.1-1.15V (1.2+ is not really advised), and increase his IOD, CCD, and DDR voltages; then perhaps disable XMP and try to configure his own set of timings.

The thing is, even though his CPU load is correlated to the BSODs, some of those BSODs correlate to RAM errors, and unless he did his own manual tweaking, CPU instability is highly unlikely.
 
To be honest, the reason I don't often comment on support threads is that I'm never sure if my input is valid and I don't want to make up solutions that may have no impact.

If I had to take a crack at it:
From what I've seen and from what I've read online, when your Ryzen CPU is unstable it causes RAM errors and vice versa, and I would imagine that has something to do with the infinity fabric and its communication with RAM, the CCDs, and the entire IO interface on the CPU (massive guess). If he has BSODs at high CPU load but he doesn't have any manual control over his Vcore or clock speed (if the chip had degraded his board would compensate as the chip would ask for more voltage via VID. More than likely his IMC just does not like the IC on his RAM and as a result, he may need to increase VSoC and VDDG IOD to brute force stability.

In my honest opinion though, Evetech upgrade kits are problematic because the RAM they offer you is almost always a toss-up with regards to if it can be stabilized on Ryzen. I have Samsung B-Die and when XMP wouldn't load but stock 2133 caused page file errors and WHEA errors, my heart sank. I was extremely relieved once I managed to configure settings that were 24/7 stable. He's either going to have to up his SoC to 1.1-1.15V (1.2+ is not really advised), and increase his IOD, CCD, and DDR voltages; then perhaps disable XMP and try to configure his own set of timings.

The thing is, even though his CPU load is correlated to the BSODs, some of those BSODs correlate to RAM errors, and unless he did his own manual tweaking, CPU instability is highly unlikely.

I have tried to set the core voltage of the cpu in the bios, but my mobo is on the MSi click light GSE version which does not give me the option of setting a value in the box, but rather brings up a list of presets to choose from ( Auto, Override Mode, Offset mode, Override + Offset mode and amd overclocking ) none of those options sets the voltage lower, and BSOD still occurs after i restart the pc . I have also used a different set of ram , Gskill 3200 CL14 which came highly recommended for ryzen , high price also , but got the same BSOD issues.

What I have doen in the meantime to resolve the issue of some sorts, was to download the Ryzen Master software and in that when PC booted into windows I set the core voltage to 1.25 v and then ran a 24 hour stress test on cpu and gpu with no more BSOD after that. This makes me believe that the cpu ( R5 3600 ) is not stable on the “ stock “ high volts that the mobo is giving to the cpu which is between 1.416 and 1.458 on idle .

My best solution on this is to get a 2000 model ryzen cpu ( 2700 / 2700x ) and then revert to an older standard bios version on the msi mobo that has all the normal settings available and sell the R5 3600, because at 1.25 v set on the ryzen master the pc works , but it is crippled at a stock speed of 3600mhz with no boost / turbo active.

Or I could get a newer B550 mobo that would be better compatible with my newer cpu.
 
I have tried to set the core voltage of the cpu in the bios, but my mobo is on the MSi click light GSE version which does not give me the option of setting a value in the box, but rather brings up a list of presets to choose from ( Auto, Override Mode, Offset mode, Override + Offset mode and amd overclocking ) none of those options sets the voltage lower, and BSOD still occurs after i restart the pc . I have also used a different set of ram , Gskill 3200 CL14 which came highly recommended for ryzen , high price also , but got the same BSOD issues.

What I have doen in the meantime to resolve the issue of some sorts, was to download the Ryzen Master software and in that when PC booted into windows I set the core voltage to 1.25 v and then ran a 24 hour stress test on cpu and gpu with no more BSOD after that. This makes me believe that the cpu ( R5 3600 ) is not stable on the “ stock “ high volts that the mobo is giving to the cpu which is between 1.416 and 1.458 on idle .

My best solution on this is to get a 2000 model ryzen cpu ( 2700 / 2700x ) and then revert to an older standard bios version on the msi mobo that has all the normal settings available and sell the R5 3600, because at 1.25 v set on the ryzen master the pc works , but it is crippled at a stock speed of 3600mhz with no boost / turbo active.

Or I could get a newer B550 mobo that would be better compatible with my newer cpu.
Your stock voltage is typical for Ryzen. As I'm typing this I took a peek at HWinfo64 while watching a series on my web browser and my CPU was sitting at 1.41-ish. Usually parts scale with voltage and it's very rare that too high a voltage can cause BSODs. What the issue could be is that your automatic voltage is behaving incorrectly at high loads. If possible, run a benchmark like R23 with HWinfo open and see how far down your load voltage drops. What's more likely is that when you took manual control of your Vcore, you circumvented your CPU/board dropping Vcore too far.

The concept of too high a Vcore (especially at stock) causing BSODs is highly unlikely. Generally speaking, unless you're worried about transients causing oxide breakdown or heat caused by high voltage (not something that you'd need to worry about at stock), you can run voltage as high as you want. 1.45V idle with a very conservative LLC setting would be perfectly safe on most CPUs, but most people tend to not take that approach as it's rather daunting.

Ryzen at 1.4+ volts at stock is absolutely normal, and with little to no current load, the equation would be something like maybe:
1.42V X 20A = 28W of heat at a modest load in Windows. You're definitely correct in that it's probably voltage related because you tested expensive B-Die. More than likely it's the board and the current BIOS having buggy voltage regulation. I will say though - override mode should allow you to alter Vcore in BIOS. It's odd that it isn't. Afaik override mode has had that function on every board I've owned so I would assume that BIOS revision is just borked.
 
Your stock voltage is typical for Ryzen. As I'm typing this I took a peek at HWinfo64 while watching a series on my web browser and my CPU was sitting at 1.41-ish. Usually parts scale with voltage and it's very rare that too high a voltage can cause BSODs. What the issue could be is that your automatic voltage is behaving incorrectly at high loads. If possible, run a benchmark like R23 with HWinfo open and see how far down your load voltage drops. What's more likely is that when you took manual control of your Vcore, you circumvented your CPU/board dropping Vcore too far.

The concept of too high a Vcore (especially at stock) causing BSODs is highly unlikely. Generally speaking, unless you're worried about transients causing oxide breakdown or heat caused by high voltage (not something that you'd need to worry about at stock), you can run voltage as high as you want. 1.45V idle with a very conservative LLC setting would be perfectly safe on most CPUs, but most people tend to not take that approach as it's rather daunting.

Ryzen at 1.4+ volts at stock is absolutely normal, and with little to no current load, the equation would be something like maybe:
1.42V X 20A = 28W of heat at a modest load in Windows. You're definitely correct in that it's probably voltage related because you tested expensive B-Die. More than likely it's the board and the current BIOS having buggy voltage regulation. I will say though - override mode should allow you to alter Vcore in BIOS. It's odd that it isn't. Afaik override mode has had that function on every board I've owned so I would assume that BIOS revision is just borked.
Yes , this Bios GSE Lite version of MSi is not good at all. I have a Msi Z370 Krait Mobo and it is clear and easy to use. I have downloaded 7 different updated versions of the current Mobo’s bios with the latest one that does not even get to the windows home screen with out BSOD.

I have done the cinebench r23 bench on its own, but then the PC does not BSOD, if I run multiple bench marks, Cinebench r23, Heaven GPU bench, and CPUZ cpu stress and download some games, then the pc BSOD. I got to push the PC to the limit to get the BSOD.

Do you think the voltage drops to low when running benches / gaming and then that causes BSOD ?
 
Yes , this Bios GSE Lite version of MSi is not good at all. I have a Msi Z370 Krait Mobo and it is clear and easy to use. I have downloaded 7 different updated versions of the current Mobo’s bios with the latest one that does not even get to the windows home screen with out BSOD.

I have done the cinebench r23 bench on its own, but then the PC does not BSOD, if I run multiple bench marks, Cinebench r23, Heaven GPU bench, and CPUZ cpu stress and download some games, then the pc BSOD. I got to push the PC to the limit to get the BSOD.

Do you think the voltage drops to low when running benches / gaming and then that causes BSOD ?
That's almost certainly why. I think run it anyway but not to prove it's going to BSOD, but rather to see how low it drops. Then perhaps do a run with a manual 1.25V Vcore. That way you can compare the load high-current load voltages.
 
That's almost certainly why. I think run it anyway but not to prove it's going to BSOD, but rather to see how low it drops. Then perhaps do a run with a manual 1.25V Vcore. That way you can compare the load high-current load voltages.
Will download HWinfo tonight after work and see what happens. Thanks in advance
 
Will download HWinfo tonight after work and see what happens. Thanks in advance
I know my advice isn't the most straightforward but I can assure you that it will be worth it to do a direct comparison at load to see where your stock vs your adjusted voltage settings dip down to. I'm going to predict that stock is going to Vdroop significantly lower or perhaps have strange transient fluctuations. I once owned a Z390 Aorus Master that made me think my 9900K had trash bin silicon. When I acquired a Maximus Formula my CPU could run 5GHz @ 1.3V with a modest LLC and 5.1 @ 1.375.

When the Master received a Beta Resize-BAR BIOS I tried to dial in my Maximus settings and see if perhaps the older Master BIOS was the issue as 'Fixed Vcore behaviour' was listed in the new beta BIOS. Low and behold the Aorus Master was simply observing incorrect Vcore behavior for the period I used it for and the new 5GHz settings worked just as they did on the Maximus Formula.

Sometimes AGESA BIOSs just don't agree with certain boards. It sucks but Ryzen tends to have a lot of issues that I think would be alleviated if vendors were allowed to adapt the BIOSs themselves instead of having to wait for AMD to release them.
 
I know my advice isn't the most straightforward but I can assure you that it will be worth it to do a direct comparison at load to see where your stock vs your adjusted voltage settings dip down to. I'm going to predict that stock is going to Vdroop significantly lower or perhaps have strange transient fluctuations. I once owned a Z390 Aorus Master that made me think my 9900K had trash bin silicon. When I acquired a Maximus Formula my CPU could run 5GHz @ 1.3V with a modest LLC and 5.1 @ 1.375.

When the Master received a Beta Resize-BAR BIOS I tried to dial in my Maximus settings and see if perhaps the older Master BIOS was the issue as 'Fixed Vcore behaviour' was listed in the new beta BIOS. Low and behold the Aorus Master was simply observing incorrect Vcore behavior for the period I used it for and the new 5GHz settings worked just as they did on the Maximus Formula.

Sometimes AGESA BIOSs just don't agree with certain boards. It sucks but Ryzen tends to have a lot of issues that I think would be alleviated if vendors were allowed to adapt the BIOSs themselves instead of having to wait for AMD to release them.
I bought a MSi Tomahwk B350 and a 1700x when ryzen launched and that 1700x could never go over 3.8 ghz without BSOD’s, had to dial it down to 3.7 ghz to be stable with the XMP set on 3200 with my Gskill CL14’s. Now this 3600 is giving me more reasons to never ever go ryzen again.
 
I bought a MSi Tomahwk B350 and a 1700x when ryzen launched and that 1700x could never go over 3.8 ghz without BSOD’s, had to dial it down to 3.7 ghz to be stable with the XMP set on 3200 with my Gskill CL14’s. Now this 3600 is giving me more reasons to never ever go ryzen again.
Was that stock behaviour or did you try to manually set it to 3.8? I didn't jump on the Ryzen train until this current platform I bought. From what I remember, even 4.0 was difficult?
 
Was that stock behaviour or did you try to manually set it to 3.8? I didn't jump on the Ryzen train until this current platform I bought. From what I remember, even 4.0 was difficult?

The mac boost was 3.8 ghz on the 1700x, but if I wanted to run my CL14's @ 3200mhz I had to downclock the cpu to 3.7 to run stable :( .
 
Status
Not open for further replies

Users who are viewing this thread

Back
Top Bottom