1.5V is the hard cap for Zen 3 - The Stilt / 1.55V is the cap for Zen 3 - AMD

[Deleted member 345482]

Following on from this thread, I've continued to do research and spoken with the AMD Technical Marketing guy who made the Zen 3 voltage slide, as well as overclocker The Stilt. Both have opposing views of what is acceptable, with the latter basing it on AMD's baked-in limits...

A little bit of context partly copied from my thread at OCN. It's a simple question, but after more than 5 weeks of searching for an authoritative answer, I'm lost. In theory, AMD should be that authoritative answer but AMD's hardware contradicts its employees (more below).

➡ The Question
At stock power limits with PBO and any vendor-specific motherboard 'enhancements' disabled, as well as automatic voltage and LLC, and with DOCP/XMP enabled or disabled, what is the maximum amount of voltage Zen 3 can request and will be delivered?

➡ The Reason For Asking
The reason for asking this simple question is after noticing HWiNFO reporting up to 1.531V being delivered (VDDCR_CPU/SVI2 TFN) to the CPU, and 1.550V being requested (VID).

Now on to the answers. In order to make it easier to read, I will insert thumbnails instead of full-sized images. If people find this difficult to see, please let me know and I'll edit the post.

➡ Answer #1: AMD promotional slide - the following is taken from the thumbnail side

Factory Voltage Range with no overclock, no undervolt, 100% stock operation: 0.200V - 1.500V

Answer: 1.500V

Max Boost (Lightly-Threaded/Bursty Workloads): ~0.450V - 1.500V

Answer: 1.500V
Source:


➡ Answer #2: AMD Support - the following is taken from an AMD support email

... the voltage that you're detecting is totally fine, [especially] after [checking] the good performance that you're receiving in multicore tasks.
The peaks [of 1.531V] mentioned in "SVI2 TFN" are for less than a second, and probably during a boost on the performance...

Due to the technology Precision Boost 2, your processor will always require the maximum amount of power that it [considers] safe (and your motherboard can deliver), in order to keep the frequency of all cores at the maximum when in use. If this wasn't possible to handle, you would receive crashes and most likely, the system will automatically shut down as protection.

Answer: 1.531V - 1.550V
Source #1:


Source #2: Phone call where I mention 1.550V, which unfortunately I can't provide a copy of.

➡ Answer #3: Professional overclocker 'The Stilt' - the following is taken from a discussion on Reddit and OCN with overclocker The Stilt

[Assuming the SMU operates at it's default configuration and the VRM configuration matches AMD's specifications and there is no voltage offsets or voltage / current biasing of any sort], Zen 2 or Zen 3 CPUs never request over 1.500V, it is specifically limited to this value just like AMD specifies in the slides ([un]surprisingly).

The effective Vmax can be lowered and some AGESA-versions have allowed that to the end-users however, there is no real way to increase the limit. SVI2 TFN (i.e. the telemetry) voltage for the CPU cores (VDDCR_CPU) can be seen reaching 1.500V when the actual (not the SW reading) VDDCR_CPU plane current is zero, i.e. there is no voltage drop (Rll xmOhm * 0 = 0).

Answer #1: 1.50000V
Note: After speaking privately with The Stilt and probing my CPU (following his instructions), he confirmed the CPU is hard-limited to 1.50000V. Therefore, it's my Asus B550-E that is doing something, although after The Stilt spending a day working on different BIOSes and boards, he couldn't work out what.
Source:

1.55000V is the maximum voltage SVI2 interface standard. 1.50000V (up to) is what is actually used on Zen 2 / Zen 3 CPUs, when they are being operated as intended by AMD (i.e. no foul from the motherboard side), period.

Answer #2: 1.50000V (delivered voltage)
Source:


And finally, confirmation of what the CPU reports


➡ Answer #4: AMD Technical Marketing 'AMD Robert' - the man who created the slide in Answer #1 - the following is taken from a discussion on Reddit with verified AMD employee Robert, who works in Technical Marketing and answers technical on the AMD sub-reddit.

1.55V is the maximum factory VID for automatic/stock operation on all Ryzen parts from 2016 until the time of this post. It's not unusual, but values around 1.5V are just more typical. But if the processor is very chilly and being lightly used, 1.55V is possible.

Answer: 1.550V (VID and delivered, i.e. VDDCR_CPU)
Source:


➡ Conclusion
Assuming everything is stock (excl. XMP/DOCP);

1. based on AMD's slide the maximum VID is 1.5000V
2. based on the person who made the slide the maximum VID is 1.5500V
3. based on AMD Support the maximum 'voltage' is 1.5500V
4. based on The Stilt, a world-famous professional overclocker, the maximum VID is 1.5000V
5. Based on my specific CPU, the maximum VID is 1.50000V

What are your thoughts? AMD's hardware contradicts AMD's statements...

The Stilt confirmed the exact CPU limit for voltage requests on my CPU. That limit, according to the CPU, is 1.50000V. That is, according to The Stilt, the case with at least the 5800X, 5900X, & 5950X. I realise this only applies to VID, but a higher VID often leads to a higher Core SVI2 TFN.

 

[Skull_Angel]

Possible margin of error in reporting? Though 0.031 does seem high for deviance; if programs report higher than a physical probe, that's what I'd usually suspect.

May be the materials' thermal properties have a larger variance in resistance than expected and it's somehow effecting the way it's metered/reported. Impurities/imperfections have the potential to create large variances at such a small scale and we do know that the manufacturing QC at this size isn't perfect.

edit; just realized there's another thread on this subject, might want to have this one locked or merged, haha

 

[Deleted member 345482]

Possible margin of error in reporting? Though 0.031 does seem high for deviance; if programs report higher than a physical probe, that's what I'd usually suspect.

May be the materials' thermal properties have a larger variance in resistance than expected and it's somehow effecting the way it's metered/reported. Impurities/imperfections have the potential to create large variances at such a small scale and we do know that the manufacturing QC at this size isn't perfect.

edit; just realized there's another thread on this subject, might want to have this one locked or merged, haha

I've spoken to the developer of HWiNFO, as well as The Stilt, about the Core VID (Effective) and both have stated that is an accurate figure, or at least it's more accurate than the VID reported at the top of HWiNFO and The Stilt's view is very clear - it should never go above 1.50000V.

I'd rather have the old thread deleted, if both can't simultaneously exist, as I believe this thread puts forward the case in a clear way with sources. I think (hope) that's more useful to people.

 

[FlawleZ]

Well what's requested and what's applied are two different things. We've long seen differences in voltage supplied between motherboards and even specific bioses. You would need to compare the CPU back to back with multiple boards and/or BIOS revisions to have any kind of idea if what is responsible for supplying that amount of voltage.

 

[pendragon1]

I'd rather have the old thread deleted,

you can close it but this is the duplicate thread.

 

[Deleted member 345482]

Well what's requested and what's applied are two different things. We've long seen differences in voltage supplied between motherboards and even specific bioses. You would need to compare the CPU back to back with multiple boards and/or BIOS revisions to have any kind of idea if what is responsible for supplying that amount of voltage.

The factory limit (as confirmed by The Stilt) for my CPU is set to 1.50000V (confirmation by probing the CPU with specialised tools) which is its vmax. No motherboard at optimised/stock settings should ever push the CPU over this value.

 

[Deleted member 345482]

you can close it but this is the duplicate thread.

View attachment 420964

I don't have the permissions to do that.

 

[pendragon1]

The factory limit (as confirmed by The Stilt) for my CPU is set to 1.50000V (confirmation by probing the CPU with specialised tools) which is its vmax. No motherboard at optimised/stock settings should ever push the CPU over this value.

the stilt is not amd, he did not make the chip. if amd is saying its fine, give up, your on a pointless crusade for some reason.

I don't have the permissions to do that.

odd.

 

[Deleted member 345482]

the stilt is not amd, he did not make the chip. if amd is saying its fine, give up, your on a pointless crusade for some reason.


odd.

He's not AMD, but even if ignoring him entirely, the fact remains the CPU reports its factory-set vmax at 1.50000V.

 

[FlawleZ]

He's not AMD, but even if ignoring him entirely, the fact remains the CPU reports its factory-set vmax at 1.50000V.

It sounds like you would be best served hitting up the motherboard manufacturers and asking why they are oversupplying voltage. In your case I would just reach out to Asus.

 

[noko]

There is a voltage drop from the VRM through the power plane to the CPU, if the CPU is asking for a higher voltage then 1.5v, it is maybe because it is not seeing 1.5v at the socket or internally so it is OK. CPU can still be limited to 1.5v internally but can request higher voltage from the motherboard because of inherit voltage drops from the VRMs to the CPU.

Modern CPUs look at a number of factors, such as Temperature, Current, how many cores active etc. Now if one pushes 1.5v to all cores at full load, allows temperatures to be rather high -> I would expect degradation would occur pretty quick. I know I zapped a Zen 1700x with over 1.75v for about a 5 minute period due to my mistake in the bios on voltage offset -> result was it could no longer do any kind of boost without crashing. Any voltage over 1.4v would cause a crash afterwards, ended up getting another 1700x and using the degraded 1700x in another motherboard with PB disabled with a manually OC since it worked fine with voltages less than 1.4v. Used it for a couple of years that way.

 

[Deleted member 345482]

There is a voltage drop from the VRM through the power plane to the CPU, if the CPU is asking for a higher voltage then 1.5v, it is maybe because it is not seeing 1.5v at the socket or internally so it is OK. CPU can still be limited to 1.5v internally but can request higher voltage from the motherboard because of inherit voltage drops from the VRMs to the CPU.

Modern CPUs look at a number of factors, such as Temperature, Current, how many cores active etc. Now if one pushes 1.5v to all cores at full load, allows temperatures to be rather high -> I would expect degradation would occur pretty quick. I know I zapped a Zen 1700x with over 1.75v for about a 5 minute period due to my mistake in the bios on voltage offset -> result was it could no longer do any kind of boost without crashing. Any voltage over 1.4v would cause a crash afterwards, ended up getting another 1700x and using the degraded 1700x in another motherboard with PB disabled with a manually OC since it worked fine with voltages less than 1.4v. Used it for a couple of years that way.

The Stilt is firmly of the view that the 'VID (Effective)' reading should never go above 1.50000V at stock settings as the fused Vmax is 1.50000V. Anyone using a 5000 series CPU at stock (like me) should never see that value increase over the factory limit.

 

[SmokeRngs]

The Stilt is firmly of the view that the 'VID (Effective)' reading should never go above 1.50000V at stock settings as the fused Vmax is 1.50000V. Anyone using a 5000 series CPU at stock (like me) should never see that value increase over the factory limit.

Unless the motherboard itself is feeding the CPU more voltage. It sure as hell wouldn't be the first time a motherboard manufacturer has done something like this in order to eek out a small win in benchmarks. From what I've seen Asus is notorious for doing this and has been for years.

 

[alxlwson]

The Stilt is firmly of the view that the 'VID (Effective)' reading should never go above 1.50000V at stock settings as the fused Vmax is 1.50000V. Anyone using a 5000 series CPU at stock (like me) should never see that value increase over the factory limit.

Fuses don't control voltage. They control current.

 

[Deleted member 345482]

Update: I appear to have fixed the problem. It's now been a little over a month since I (think I) found the fix. I've been monitoring with HWiNFO constantly, and the effective VID has not increased above 1.50000V, whereas before it would instantly raise to 1.55000V with most games and applications - event at idle.

- Both Martin, the developer of HWiNFO, and The Stilt have stated that my fix shouldn't have worked, but I'm not sure what else to say.
- The problem was AMD's driver installer stating that a driver install had been successful, when it hadn't actually checked whether Windows was able to properly initialise the devices.
- Since every part of the PC was new, I installed Windows 11 and the AMD Chipset drivers (and the two/three updates to the Chipset drivers) in a normal way.
- However, whilst the installer stated the drivers were installed, and Device Manager showed 'This device is working properly', under the Events tab (in Device Manager), it was clear Windows had never been able to correctly initialise any of the devices
- This applies to devices such as SMBus, GPIO, PCI, I2C, and even PSP
- After completely removing all AMD drivers (but saving the extracted AMD directory the installer creates on C:\), and then manually installing the driver for each device one at a time, Windows was then able to initialise the device
- Since performing that (and rebooting) the effective VID has thus far remained at 1.50000V
- Performance increased immediately, but only slightly (not overclocked, only XMP enabled with 1:1, Windows Update disabled for testing). Cinebench R23 1C: 1,610 | nT: 21,600. Again, that's only with XMP enabled on my relatively slow RAM (3600 CL18, 18-22-22-42) and obviously 1:1 FCLK, etc.

Fuses don't control voltage. They control current.

Not sure if you're being deliberately obtuse, or not. A feature can be fused off in silicon... it's very common.

 

[alxlwson]

Not sure if you're being deliberately obtuse, or not. A feature can be fused off in silicon... it's very common.

You stated the FUSED vmax is 1.500000.
That's not a hard feature such as cache or cuda cores. That's a control, likely in software(uefi/bios), not a fuse. I wasn't being obtuse. I was correcting/informing you. So in this case, my original statement stands.

 

[Deleted member 345482]

You stated the FUSED vmax is 1.500000.
That's not a hard feature such as cache or cuda cores. That's a control, likely in software(uefi/bios), not a fuse. I wasn't being obtuse. I was correcting/informing you. So in this case, my original statement stands.

Setting a fixed setting on the silicon at a factory level (it's the chip that reads 1.50000V, not BIOS/UEFI or AGESA) is still regarded as fusing a product. My correction stands.

 

[alxlwson]

Setting a fixed setting on the silicon at a factory level (it's the chip that reads 1.50000V, not BIOS/UEFI or AGESA) is still regarded as fusing a product. My correction stands.

Show me that in some documentation. Never once in my entire life have I heard someone refer to hardcode limits as a fuse or being fused.
I think you're making things up.

 

[Deleted member 345482]

Show me that in some documentation. Never once in my entire life have I heard someone refer to hardcode limits as a fuse or being fused.
I think you're making things up.

No.

 

[Mr Evil]

Show me that in some documentation. Never once in my entire life have I heard someone refer to hardcode limits as a fuse or being fused.
I think you're making things up.

Here's some documentation for a microcontroller where some configuration bits are described as "fuses": https://onlinedocs.microchip.com/pr...tml?GUID-5AB72304-955B-47E1-A9C0-EA312CB9C814
Those are reconfigurable, but the term "fuse" comes from literal tiny fuses in devices that are blown at the factory to permanently configure it.

 

[alxlwson]

Here's some documentation for a microcontroller where some configuration bits are described as "fuses": https://onlinedocs.microchip.com/pr...tml?GUID-5AB72304-955B-47E1-A9C0-EA312CB9C814
Those are reconfigurable, but the term "fuse" comes from literal tiny fuses in devices that are blown at the factory to permanently configure it.

Yup. But in this thread's case, it's still not a fuse. Not in the current controlling/protection definition, and not in the feature configuration definition.
OP is talking about code on silicon. That meets neither of these definitions. In fact, the definition is firmware. The CPU has firmware/embedded software on it for things like this.

 

[Deleted member 345482]

Yup. But in this thread's case, it's still not a fuse. Not in the current controlling/protection definition, and not in the feature configuration definition.
OP is talking about code on silicon. That meets neither of these definitions. In fact, the definition is firmware. The CPU has firmware/embedded software on it for things like this.

Hardware features are fused off, and not generally through firmware.

 

[alxlwson]

Hardware features are fused off, and not generally through firmware.

Vmax is not a hardware feature. That is controlled by firmware or embedded software on the cpu itself.

Really strange sword to die on.

 

[Deleted member 345482]

Vmax is not a hardware feature. That is controlled by firmware or embedded software on the cpu itself.

Really strange sword to die on.

According to The Stilt (I give up if you're not familiar with him), Vmax is factory-configured and it's possible to read this value straight from the CPU with specialised low-level tools. For example, I know the hardware-limited true Vmax of my CPU is 1.50000V, irrespective of what's reported by standard tools.

Odd that you're on HF and don't understand what fusing features off means... it's one of the most basic concepts for product segmentation and binning (i.e. a faulty GA-102 might have multiple SMs fused off at a hardware level during the binning process). You can see the result of what happens when that goes wrong with the EVGA 2060 KO series where some functional features weren't fused off. Gamers Nexus has a great video on it.

 

[alxlwson]

According to The Stilt (I give up if you're not familiar with him), Vmax is factory-configured and it's possible to read this value straight from the CPU with specialised low-level tools. For example, I know the hardware-limited true Vmax of my CPU is 1.50000V, irrespective of what's reported by standard tools.

Odd that you're on HF and don't understand what fusing features off means... it's one of the most basic concepts for product segmentation and binning (i.e. a faulty GA-102 might have multiple SMs fused off at a hardware level during the binning process). You can see the result of what happens when that goes wrong with the EVGA 2060 KO series where some functional features weren't fused off. Gamers Nexus has a great video on it.

Lol...

You aren't understanding what other people are talking about. Go back to your god Stilt and ask him yourself.

And yes, I understand feature fusing. I know the differences between what we are talking about. I was unlocking gpu pipelines with driver and bios mods, and cpu multipliers with pencil lead when I was 13.

 

[Deleted member 345482]

Lol...

You aren't understanding what other people are talking about. Go back to your god Stilt and ask him yourself.

And yes, I understand feature fusing. I know the differences between what we are talking about. I was unlocking gpu pipelines with driver and bios mods, and cpu multipliers with pencil lead when I was 13.

Putting aside your obnoxious tone, you still made a false claim about what fusing off a feature means. And when presented with information explaining what it means on a hardware level (I assume you are unable to use a search engine), you decided to resort to juvenile behaviour. Grow up.

 

[alxlwson]

Putting aside your obnoxious tone, you still made a false claim about what fusing off a feature means. And when presented with information explaining what it means on a hardware level (I assume you are unable to use a search engine), you decided to resort to juvenile behaviour. Grow up.

Oooo, insults!!!

 

[ZodaEX]

Oooo, insults!!!

Grow up. You're not 13 anymore.

 

[alxlwson]

Grow up. You're not 13 anymore.

MYOB

 

[pendragon1]

first he said stilt and amd do not know what they are talking about but he does and now hes using stilt as proof and getting insulting over it. lol....
edited speeling

 

[alxlwson]

first he said stilt and amd do not know what they are talking about and but he does and now hes using stilt as proof and getting insulting over it. lol....

Exactly. And now this other guy is white knighting lol.

 

[Deleted member 345482]

first he said stilt and amd do not know what they are talking about but he does and now hes using stilt as proof and getting insulting over it. lol....
edited speeling

I first stated that they had contradicting opinions, which remains true. I did not "[say] stilt and amd do not know what they are talking about (sic)". I have since confirmed the correct figure and updated the thread with my findings. My CPU is now operating within its factory-set limit of 1.50000V.

This type of behaviour on this forum, of all places, is more juvenile than reddit.

 

[pendragon1]

This type of behaviour on this forum, of all places, is more juvenile than reddit.

thats part of why its the only 25 year old hardware forum around...
learn to love it or live with it, or dont, im not your dad.

 

[Deleted member 345482]

thats part of why its the only 25 year old hardware forum around...

So you're claiming the juvenile behaviour and intimidation of new members is the key to HF's success? Personally, I'd say it's down to experienced, polite, and helpful users.

learn to love it or live with it, or dont, im not your dad.

I've not claimed you are my father.

 

[pendragon1]

So you're claiming the juvenile behaviour and intimidation of new members is the key to HF's success? Personally, I'd say it's down to experienced, polite, and helpful users.


I've not claimed you are my father.

yes, lighten up, ffs. we speak the truth and usually dont pull punches.
youve claims that go against the manufacturer and other known and trusted hardware "pros". so expect push back and people calling bs.

 

[Deleted member 345482]

youve claims that go against the manufacturer and other known and trusted hardware "pros". so expect push back and people calling bs.

If you'd taken the time to read the content, you'd see the contradictory "claims" come from three sources at AMD - including a technical director I've spoken with, two sources at Asus, and a professional overclocker (who it would appear is disliked here for some reason) who receives samples, and whose tweaks (and name) are in almost every Asus from the last decade.

So, you can 'call bs' if you like, but you're then accusing AMD, Asus, and a person who works with Asus and holds world records of "bs".

Please grow up.

 

[noko]

If you'd taken the time to read the content, you'd see the contradictory "claims" come from three sources at AMD - including a technical director I've spoken with, two sources at Asus, and a professional overclocker (who it would appear is disliked here for some reason) who receives samples, and whose tweaks (and name) are in almost every Asus from the last decade.

So, you can 'call bs' if you like, but you're then accusing AMD, Asus, and a person who works with Asus and holds world records of "bs".

Please grow up.

I have zero understanding on what point you are trying to make on this thread, maybe just to unimportant for me. Anyways CPUs have stock limits from decades past and people have routinely bypassed and exceeded those limits at will since. For example, this 5800x at 1.832v setting a record:

https://hwbot.org/submission/4889243_espo_sun_cinebench___2003_ryzen_7_5800x_8806_cb

Of course on Liquid Nitrogen to prevent the 5800x from nuclear exploding (exaggerating here).

 

[Deleted member 345482]

I have zero understanding on what point you are trying to make on this thread, maybe just to unimportant for me. Anyways CPUs have stock limits from decades past and people have routinely bypassed and exceeded those limits at will since. For example, this 5800x at 1.832v setting a record:

https://hwbot.org/submission/4889243_espo_sun_cinebench___2003_ryzen_7_5800x_8806_cb

Of course on Liquid Nitrogen to prevent the 5800x from nuclear exploding (exaggerating here).

If you took the time to read it, you would see that your reply is not relevant.

 

[noko]

If you took the time to read it, you would see that your reply is not relevant.

So your point was 1.5v was max stock limit? While actually depending on motherboard you can push way beyond that voltage. This is no different then most CPUs for the past 3 decades having a stock voltage and ways to OV from bios, software or modifying power regulators. As for what significance this is is the question?

 

[alxlwson]

So your point was 1.5v was max stock limit? While actually depending on motherboard you can push way beyond that voltage. This is no different then most CPUs for the past 3 decades having a stock voltage and ways to OV from bios, software or modifying power regulators. As for what significance this is is the question?

He was saying that there was some "fusing" on the cpu itself that set this value rather than it being hardcode or firmware on the cpu itself, and this the basis for this whole thread. And because he thought it was fused(lol), that it couldn't possibly exceed the 1.5v.
Just a case of someone reading information from one place, and trying to regurgitate it elsewhere and not actually understanding the background and facts. Among other things.