Overclock.net banner
2881 - 2900 of 11667 Posts

·
Registered
Joined
·
2 Posts
Hello,

Just for information / update, for me BIOS 2206 don't have the reboot / shutdown problem. I had this problem since 2010, 2103 and 2204. My performance are the same as 1302 BIOS. Now i have even a better stability with my ram at 3733 Mhz, with lower SOC voltage (1.1 in 1302 to 1.05 in 2206).
 

·
Premium Member
Joined
·
2,827 Posts
Just a little heads-up on a new feature, that will find its way into the upcoming bios releases. The feature is called as "PBO Fmax Enhancer", and it will be available in the "Ai Tweaker/Precision Boost Override" -menu among with other of the “Precision Boost Override” related options, on compatible ASUS motherboards. This feature is exclusive to Zen 2 based Ryzen 3000-series CPUs ("Matisse" & "Starship") and it will work on both, the consumer (AM4) and HEDT (sTRX40) platforms.



So, what does it do then?

This is an opportunistic feature and because of that, exactly no improvements are granted, let alone being promised to or by anyone. What it means in practical terms, is that the results will depend mostly on the silicon quality and partly, on the CPU SKU as well. Because of that the achieved results are expected to vary rather wildly, even between the different specimens of the same CPU SKU.

As a real-world example, this is what was achieved on a random early-production R9 3900X CPU. Despite this most likely is far from the best-case-scenario, consider it as a such if you will: Cinebench R20 nT average effective frequency (APERF) >> 145.9MHz improvement (3946.8MHz to 4092.7MHz). The improvement during a single-threaded, scalar-vector path-tracing workload (SPT-AVX2): 61.6MHz (average effective frequency, APERF), 4548MHz >> 4609.6MHz. As said before, there is no question if the achieved results will vary, since they will and wildly so. Some of the combinations might illustrate even better results, some no change at all, and some even WORSE behaviour than the stock.

Cinebench R20 nT Default:



Cinebench R20 nT "PBO Fmax Enhancer Enabled":



SPT-AVX2 single-threaded hybrid path-tracer (SSE4.2/AVX2) workload, default:



SPT-AVX2 single-threaded hybrid path-tracer (SSE4.2/AVX2) workload, "PBO Fmax Enhancer Enabled":



At what cost is this done?

Reliability wise, while maintaining the same user-controller settings that directly affect the reliability (PBO Scalar), there will not be any meaningful or measurable difference in the reliability. In fact, in multithreaded workloads the reliability should typically be ever-so-slightly higher than stock, due to the slightly lower voltage. In single-threaded workloads there might be a tiny increase in peak voltage observed during the maximum boost, at least on some CPU SKUs.

That being said, all of the CPUs will still obey the absolute maximum (and in terms of an increase, immutable) voltage limit of 1.50000V, that is enforced by the power management of the CPU. So, in case a CPU already hits 1.50000V during the maximum boost, at stock, enabling the feature will not increase the voltage any further, without actions done by the user (e.g. a voltage offset).

Personally, I have observed ± 1 VID (i.e. 6.25mV) worth of a change in the maximum voltage, during single-threaded workloads. These observations have taken place on CPU SKUs with the default Fmax limit of 4.4 - 4.65GHz and therefore their peak voltage has been closer to, or even time to time at the 1.50000V limit, even at stock. Because of that, CPU SKUs with a more modest Fmax limit and hence a lower maximum voltage at stock (due to a greater "distance" until the hard 1.50000V limit), might illustrate larger absolute voltage increases.

So essentially, instead of a typical brute-force approach this is a more of a dance with the available margins. The margins that might, or might not be present in your piece of the silicon. In technical terms, what is effectively happening inside the CPU, is that the default V/F (voltage-frequency) curve is being optimized to allow slightly higher frequency headroom, at the expense of the potentially available margins.

A quick check of facts:

  • An ASUS exclusive feature for 3rd Gen. Ryzen CPUs (“Matisse” and “Starship”)
  • A supplemental feature (i.e. an enhancement) to the standard "Precision Boost" operation. Not available (nor required) in OC-Mode, regardless if “static” or "per-CCX".
  • Not compatible with the so-called "EDC tweak".
  • Will not directly increase the voltage (i.e. no voltage related rule attached). The user remains in control of all voltage related settings.
  • Will not either disable, or alter the parameters of FIT (i.e. silicon fitness / reliability).
  • The results WILL vary between the different CPUs and even between the different specimens of the same CPU SKU
So, what to do first, when I want to try this out?

The very first thing you do, is reading the post as a whole through couple more times, until you have even a vague idea of what was being said and meant. Flash the bios as usual, load the fail-safe defaults (yeah, no kidding, due to RM's hive) and start entering the settings of your choice. With the exception being anything related to the CPU voltage (VDDCR_CPU), discard all of your previous CPU voltage related settings that affect the output voltage level (i.e. offsets and load-line).

Also, the described behaviour has only been tested while the "Ai Overclock Tuner" option is set to "Manual" and the "Performance Enhancer" option to "Default". Due to the auto-rules (“do if's”) potentially affecting the other setting combinations, it's on the user’s responsibility to ensure that the parameters remain where they should be, if any other combinations are being used.

Troubleshooting, suggestions and hints:

  • Q: Any hints or suggestions, before I start testing this out?
  • A: Not a suggestion, but a hard requirement: Disable ALL of the voltage offsets (regardless if ±) AND load-line adjustments affecting the CPU voltage (VDDCR_CPU) prior using this feature, i.e. your previous CPU voltage related settings. Whatever settings might have worked for you before, forget about them since they won't apply anymore.

  • Q: Performance has deteriorated compared to stock, when all cores are utilized (e.g. CB20 NT)?
  • A: If this is a fact and not a margin-of-error kind of stuff (as usual), i.e. a 3 run reproducible average: The reason for the deteriorated performance is most likely clock stretching, which occurs when the core supply voltage falls below a certain threshold. Try increasing load-line level for CPU voltage (one step at a time), until the performance either matches, or exceeds the stock. If the performance cannot be restored through the use of load-line adjustment, restore load-line to "Auto" setting and add a positive voltage offset to the CPU voltage (again, one step at the time, i.e. 6.25mV). NOTE: Unless there is a valid reason, the user SHOULD NEVER touch the load-line setting during the "Precision Boost" operation, only while in OC-Mode (“static” or “per-CCX”). During "Precision Boost" operation, the CPU is in the control of the load-line, and a user set load-line override will prevent the CPU from controlling the load-line correctly.

  • Q: I'm having stability issues, when this feature is activated.
  • A: Proceed as described above, first try increasing the load-line and then try offset if the load-line adjustment didn't help. If neither helped, the margins in the silicon are insufficient and the feature should remain turned off ("Auto" or "Disabled").

  • Q: I'm seeing no improvement?
  • A: Tough luck, better luck with the next CPU.

  • Q: What voltage should I be monitoring?
  • A: "CPU Core Voltage (SVI2 TFN)" that is displayed under HWiNFO sensors, that’s only reading that matters for evaluating stable-state voltages without external equipment. Anything with a "VID" in it’s the name are basically just voltage requests, or commands. On a single main-plane design such as AM4 or even sTRX40 the VIDs are completely irrelevant, since despite each of the cores having their own VIDs, there is only a single voltage plane (and hence a single concurrent voltage level at the time).

  • A suggestion: ONLY enable "Max CPU Boost Clock Override" if you are close (within ~ 25MHz) of saturating the factory Fmax*, and upon saturation, only increase the Fmax by the minimum available amount at the time. Unless saturated, unnecessarily large increases will only deteriorate the result. * (3600 = 4.200GHz, 3600X = 4.400GHz, 3700X 4.400GHz, 3800X = 4.550GHz, 3900X = 4.650GHz, 3950X = 4.725GHz, for other SKUs check HWiNFO). The bios releases featuring the newly added "PBO Fmax Enhancer" will also increase the available Fmax margin from 200MHz to 400MHz, to ensure that SKUs with a low total Fmax ceiling (factory + offset), such as 3600 will have a sufficient headroom available. The initial test versions will still have the standard +200MHz limit however, the said change has already been implemented in the code.

  • A suggestion: In some cases, the results can be improved even further, by increasing the "Precision Boost Overdrive Scalar" value from the default 1x setting. This option directly affects the silicon reliability and because of that, it should be only increased from the stock if the performance improvements justify it (which they frankly, seldom do). Personally, I would never increase it to higher than 4x. The current level can be checked with HWiNFO (Central Processors >> CPU PBO Scalar (Reliability Reduction)).

  • A hint: All "Precision Boost Override" related parameters remain available, as usual. The user can increase PPT and TDC limits if they become saturated, to improve the performance even further. EDC is piped with TDC and there is no need for the user to adjust it.

A request: Out of a common courtesy, please don’t quote or link to this post, unless you’re certain that you understood what was being said and intended. That applies especially to you Paul.

Peter (Shamino) from ASUS will post some initial bios builds for you to test shortly.
 

·
Registered
Joined
·
54 Posts
Technically you can however, there is no official method to update the "wrong" binary.
You can strip the .CAP header (1000h) out of the bios binary and use e.g. Flashrom to update.
I think it would be nice if Shamino post the NON WIFI bios for the HERO too. I would like to test this new BIOS.

I have been suffering the random restart bug since the introduction of the 2000 series BIOSes... and always had to roll back to 1302 (which is perfectly stable)

I need the newer AGESA for several reasons but I wonder when is ASUS going to release a "problem free" BIOS... Please Shamino.
 

·
Registered
Joined
·
1,101 Posts
I think it would be nice if Shamino post the NON WIFI bios for the HERO too. I would like to test this new BIOS.

I have been suffering the random restart bug since the introduction of the 2000 series BIOSes... and always had to roll back to 1302 (which is perfectly stable)

I need the newer AGESA for several reasons but I wonder when is ASUS going to release a "problem free" BIOS... Please Shamino.
By the time they do, Zen 3 will be here and we'll be asking for Zen 3 BIOSes. Then the cycle repeats again, waiting (hopefully) for months before a stable BIOS arrives. Add to that we'd also have to wait for good chipset drivers.

Don't think x570 and Zen 2/3 will mature until late 2021.
 

·
Premium Member
Joined
·
2,827 Posts
I have been suffering the random restart bug since the introduction of the 2000 series BIOSes... and always had to roll back to 1302 (which is perfectly stable)
So your saying that the Crosshair VIII boards have been broken ever since the 1302 bios version, that released more than 6 months ago?
I admit I haven't used much of the C8x series board (C8F in my case) recently however, I cannot say I've noticed any issues.
 

·
Registered
Joined
·
193 Posts
just updated to 2206, I was using 1302. so far no issues. I would like to say that i have never had this motherboard crash since ive owned it. Ive had it since release. I really think most of the issues people are having is somehow related to the ram , they are using. If you have a stable set, I feel like this board has been very reliable.

one thing to note, i have pbo off, most settings are auto and far from tweaked. onboard audio has always been disabled and realtek lan disabled. other then that mostly stock, didnt see a huge point oc'ing 3800x.
 

·
Registered
Joined
·
54 Posts
So your saying that the Crosshair VIII boards have been broken ever since the 1302 bios version, that released more than 6 months ago?
I admit I haven't used much of the C8x series board (C8F in my case) recently however, I cannot say I've noticed any issues.
I am not the only one having issues, we have been talking about random restarts (even within the BIOS settings in the case of 2010-2204) since the 2xxx BIOS introduction. I thought that the 2206 BIOS had finally fix the problems but, yeah, random restarts when idle started happening after a week I believe, for some random reason (Sounds like BIOS corruption?)...

My ram modules have been working without issues from the beginning. I have two ryzen 3900X/Crosshair VIII setups, one with 2 modules of Samsung B-die chips and other one with 2 modules of Hynix DJR modules. Both systems working at CL16/3600 Mhz and tested with the "Karhu Ramtest" for at least ~+6400% coverage, so pretty much 99.98% stable... I have the same issues with both systems, so do other people at the forums. Only downgrading back to 1302 fixes the random restarts.

I include various links of the abandoned ASUS? forums and some "random" reddit threads, in case anyone is curious about the issues that plagues the Crosshair VIII 2xxx BIOSes...

BIOS-2204-Follow-up
BIOS-2103-High-Instability-Crashes
After quite some time it has begun again to randomly turn off (Reddit)
random restarts (Reddit)
 

·
Premium Member
Joined
·
2,827 Posts
I am not the only one having issues, we have been talking about random restarts (even within the BIOS settings in the case of 2010-2204) since the 2xxx BIOS introduction. I thought that the 2206 BIOS had finally fix the problems but, yeah, random restarts when idle started happening after a week I believe, for some random reason (Sounds like BIOS corruption?)...

My ram modules have been working without issues from the beginning. I have two ryzen 3900X/Crosshair VIII setups, one with 2 modules of Samsung B-die chips and other one with 2 modules of Hynix DJR modules. Both systems working at CL16/3600 Mhz and tested with the "Karhu Ramtest" for at least ~+6400% coverage, so pretty much 99.98% stable... I have the same issues with both systems, so do other people at the forums. Only downgrading back to 1302 fixes the random restarts.

I include various links of the abandoned ASUS? forums and some "random" reddit threads, in case anyone is curious about the issues that plagues the Crosshair VIII 2xxx BIOSes...

BIOS-2204-Follow-up
BIOS-2103-High-Instability-Crashes
After quite some time it has begun again to randomly turn off (Reddit)
random restarts (Reddit)
Bioses don't really corrupt.

If a system illustrates these issues when everything is running within the AMD spec and it has been verified to be an OS idenpendent issue, then some component within the system is defective. If it was a bios issue, everyone would have it.

Otherwise, it is most likely a matter of unstable OC settings.
 

·
Registered
Joined
·
54 Posts
Bioses don't really corrupt.

If a system illustrates these issues when everything is running within the AMD spec and it has been verified to be an OS independent issue, then some component within the system is defective. If it was a bios issue, everyone would have it.

Otherwise, it is most likely a matter of unstable OC settings.
But then, why do you think I dont have this issue with 1302 BIOS and older? I only suffer random restarts with 2010,2203-2204 and 2206. Something must have changed, don't you agree?
 

·
Premium Member
Joined
·
2,827 Posts
But then, why do you think I dont have this issue with 1302 BIOS and older? I only suffer random restarts with 2010,2203-2204 and 2206. Something must have changed, don't you agree?
So for example version 1201 works fine too?
If all versions after 1302 have issues on your system, it's probably somehow AGESA related.
1302 is the last version which was based on ComboPI AGESA. All of the newer ones are CombiPI V2, which include support for Renoir APUs for example.

Does the issue happen if you run Samsung B-die DIMMs at say, 2666MHz?
 

·
Registered
Joined
·
54 Posts
So for example version 1201 works fine too?
If all versions after 1302 have issues on your system, it's probably somehow AGESA related.
1302 is the last version which was based on ComboPI AGESA. All of the newer ones are CombiPI V2, which include support for Renoir APUs for example.

Does the issue happen if you run Samsung B-die DIMMs at say, 2666MHz?
1201 works fine, as fine as 1302 I would say. Only after 1302 I have problems.

I never tested running the DIMMs at 2666 Mhz but I also don't overclock. But could be... maybe new AGESA is screwing with the memory, maybe not (Since users of GIGABYTE, MSI and such are not reporting this issue). The thing is, with recent BIOS like the 2206, RAM pass all tests without an error. It is during idle (not working with the computer or even looking at it) when it just... restarts.

Since more people are affected and I have two separate systems with different DIMMS technologies... I would guess that there is an issue with newer BIOS when using certain setting or certain hardware.
 

·
Premium Member
Joined
·
2,827 Posts
1201 works fine, as fine as 1302 I would say. Only after 1302 I have problems.

I never tested running the DIMMs at 2666 Mhz but I also don't overclock. But could be... maybe new AGESA is screwing with the memory, maybe not (Since users of GIGABYTE, MSI and such are not reporting this issue). The thing is, with recent BIOS like the 2206, RAM pass all tests without an error. It is during idle (not working with the computer or even looking at it) when it just... restarts.

Since more people are affected and I have two separate systems with different DIMMS technologies... I would guess that there is an issue with newer BIOS when using certain setting or certain hardware.
Low MEMCLK is worth trying, since most of the stability issue on AMD platforms are DRAM related.
 
2881 - 2900 of 11667 Posts
Top