AMD to Introduce Adaptive Undervolting to Precision Boost Overdrive for Ryzen 5000

[InVasMani]

This is amazing getting up to 2% single thread 10% multi-thread is big boost especially on the multi-thread and the crazy part is you'd want higher multi-thread in general anyway on higher core count chips at least I would. That enables so much additional headroom for stuff like multi-threaded compression I mean up to 10% additional performance from that is kind of mind boggling. This increases the relative value Ryzen 500 series a lot. I can't wait to see what happens with Epyc/ThreadRipper surely they'll introduce this on the next chips series in those platforms as well that has huge ramifications given how multi-thread capable those chips are it's absolutely enormous plus factor in that those chips tend to be a bit more single thread starved comparatively and that extra 2% is still rather nice to have. Think about it this way a 3.5GHz all core 32C chip would be up to a 3.85GHz all core chip. I wonder if this is also in RNDA2 as well or will come with RDNA3 this is great tech and chances are they'll improve it in time I can see them tweaking it down to 0.5ms eventually next generation it might be 0.75ms followed by 0.5ms.

 

[Mussels]

I saw all the leaks hinting at this, still totally excited for it

I just have but one question: How on earth this is not the default behavior?

It wasn't ready in time for launch, i guess?


at a guess, warranty wont cover if PBO is part of the problem
You cant RMA because your chip wont boost higher, or undervolt lower than some overclocker on youtube, etc

 

[PowerPC]

Thanks to AMD for showing us what Intel should have been doing.

But I bet Intel is totally gonna blow everything away with their GPUs... right?
What does Intel even have at this point?

 

[thesmokingman]

question is how they know if PBO was used in case of warranty claim?

is PBO when 1st used sending a report to headquarters with cpu serial #?

PBO is AMD's boost control, check pic below.

 

[biffzinker]

PBO is AMD's boost control, check pic below.

That's stock out of the box boost. PBO is equal to overclocking.

 

[teeradbacesi]

How will this work with motherboards that automatically overclock themselves? AFAIK, my Asus board defaults to running things as high as it can get away with and I've never been able to find out how to stop it from doing that. Will this Adaptive Undervolting turn that off?

 

[HD64G]

How will this work with motherboards that automatically overclock themselves? AFAIK, my Asus board defaults to running things as high as it can get away with and I've never been able to find out how to stop it from doing that. Will this Adaptive Undervolting turn that off?

I guess that feature is made to allow a user to lower voltage per stage of use even when the CPU tries to oc as high as possible. And the results will depend on the silicon.

 

[Nordic]

My 3900x is perfectly functionable and reasonable. It has more cores than I really need. I only use them for crunching, which I do like to do. Why do I want a 5000 series chip so bad?

 

[Mussels]

My 3900x is perfectly functionable and reasonable. It has more cores than I really need. I only use them for crunching, which I do like to do. Why do I want a 5000 series chip so bad?

cause they're finally a product that isn't a 5% improvement every year!

 

[Octopuss]

How does lower voltage result in higher performance?

 

[Jism]

How does lower voltage result in higher performance?

It's due to the heat of the chip and PBO able to boost higher or longer,

Simply said; i run my 2700x at stock with a 360mm radiator and 6 fans. In order to keep the CPU constant below 60 degrees i have to slight undervolt. This allows the heat to stay below the 60 degree mark and thus keep the boost constantly on 4.2Ghz. When it passes the 60 degrees, the boost clock will go lower because the way it's designed todo. When i woud'nt undervolt manually (there is some headroom per chip really) it would pass 60 degrees pretty fast and when i do undervolt it will stay below.

So yes; thats how you get more performance out of it. Chips that run cooler btw do require a bit less power in comparison chips that run into their 80 degree mark. I have a older RX580 watercooled on a 240MM rad including 2 fans, which stay constantly below 60 degree mark when gaming. The peak power consumption was 300W knowing it runs from stock 1366Mhz to 1533Mhz which is almost 200Mhz more.

Polaris is'nt a great overclocker perse; it's just memory bandwidth limited in alot of scenarios. But because of the good cooling it allows me to crank up the voltages and keep it below 60 degrees for higher clocks.

 

[Kapone33]

I meant default for 5000 series which has been on the market for months. Did they "just" come up with this? Geez meez...

You mean weeks

 

[thesmokingman]

That's stock out of the box boost. PBO is equal to overclocking.

No it is not. PBO is default to auto which is stock. You just lmao, ignored what AMD wrote. Read the damn picture it says not considered oc.

Now whomever started this thread down this warranty idiocy is also wrong. The undervolting will become part of the whole PBO thing, default so it will not be considered oc or warranty killing.

All that said, if you overclock you ALREADY FREAKING KNOW IT DOES WHAT IT DOES TO WARRANTY. Complaining and repeating this over and over doesn't change anything. It never stopped ppl from overclocking anyways.

 

[NoJuan999]

AMD Ryzen Precision Boost Overdrive Explained: Precision Boost 2 vs Precision Boost Overdrive vs AutoOC | Hardware Times

AMD’s CPU clock scaling algorithm, Precision Boost 2 was introduced with the 2nd Gen Ryzen processors and an additional feature, Precision Boost Overdrive was introduced with present Gen Ryzen 3000 chips. There’s been a lot of confusion regarding these two boosting algorithms and the addition of AutoOC (often found in the same BIOS section as Precision Boost) makes the whole affair even more complicated. In this post, we’ll explain the differences between these similarly worded technologies and how they function.

 

[Makaveli]

Undervolting invalidates the warranty with AMD.

PBO also invalidates the warranty with AMD.

In my 20+ years of building pc's I've never sent a cpu in for RMA.

 

[Mussels]

PBO also invalidates the warranty with AMD.

In my 20+ years of building pc's I've sent a cpu in for RMA.

I sent one that was dead out of the box, so sad. Thats the only dead CPU i've seen in the last 30 years, that i didn't deliberately kill myself. Hell two of my systems now are AM4 ryzens with missing pins, CPU's are tough. Motherboards OTOH...

 

[InVasMani]

It's crazy how similar adaptive undervolting is to what I was speaking about involving phase inverting clock skew two months ago and how a PAM4 type of cycle duties modulation approach could net more performance by introducing more peak valleys with shorter dips.I mentioned the I/O die could already pretty much do a lot of that potentially with the right software aka OS/firmware/bios support. The chart is basically PAM4 and cycle duties tech in a nutshell same in principle for the most part. My conclusion is you could get a roughly 50% gain if you were to phase invert between cores and retro actively switch to them at the appropriate time intervals. That said factoring in voltage with adaptive undervolting the gains would be up to about 12.5% given voltage is square, but then you have heat as well so a bit less in reality is possible from a real world standpoint plus the efficiency of how often the voltage itself is regulated.

They haven't actually added more cores though, but 10% multi-thread performance is close to the 12.5% and for a single core max theoretical for a PAM4 type approach is 3.125%, but that doesn't factor in heat and switching efficiencies. Still all things considered it shows AMD could improve this further and eek out a bit more additional performance. They could drop it down to 0.5ms latency eventually perhaps and get a nudge more performance, but as they scale up core count it should add up as well because they have more load distribution to manipulate voltage and heat output further. It's crazy they implemented very similar to what I spoke about, but slightly different round about manner. It's quite fascinating to see how quickly they incorporated this type of behavior and how effective it is. I can see it improving a bit further with more cores and reduced latency. You'll always be retrained by voltage being squared and heat output however. There is defiantly diminishing returns on it very much though. I reasonably see them going to 0.5ms latency in due time on the polling rate checks to dynamically adjust things however or at least 0.75ms latency though both seem plausible.

First Signs of AMD Zen 3 "Vermeer" CPUs Surface, Ryzen 7 5800X Tested

*RDNA 1 and Zen 3 naming scheme consumer confusion incoming* But by the time those cpus come out RDNA 1 is gonna be mostly irrelevant anyway. If that turns out to be true an all AMD truly high-end gaming rig might become a reality for the first time since like... decades this end of year.

www.techpowerup.com

 

[Kapone33]

I actually RMAd a 2920x not too long ago, During the RMA process there was no mention of OC but only physical damage of the CPU before shipping (hence requiring you to send a picture of the CPU installed in a board). Then to ensure that you package the CPU in such a way that it will not be electrostatically and physically damaged during shipping. Even though Ryzen Master and Radeon Software both warn you about using the software to change settings I seriously don't think that AMD (based on my experience) uses that as much more than a CYA. I do not see how undervolting can void a warranty anyway, Especially when it's something that company provides. This feature has been on Radeon Software for a while now.

 

[InVasMani]

Undervolting shouldn't damage a chip itself, but can cause unreliability and AMD wouldn't want to be on the hook because some fool undervolted too far and thought the hardware itself was the problem and broken when it was user error that was to blame.

 

[theGryphon]

Someone enlighten me, how does/can undervolting damage the CPU? I mean in any way...

 

[GrayAx]

How does lower voltage result in higher performance?

In short: Performance is often limited by temperature, high temps means the processor thermal throttles. And a lot of adaptive-boost systems have limits on how much power they're allowed to use as well.
A cpu still does the same amount of work at a lower voltage, (provided it isn't undervolted to the point instability.) which means it does the same amount of work, at lower power. And thus lower temps.
A made up, theoretical example is a cpu that can boost to max speed, using 150W, but if it stays at 150W too long it gets too hot and drops down to 140W, lowering performance.
Meanwhile, if the cpu was undervolted, it might be running at 140W at max speed, meaning it can stay at max perpetually.

Someone enlighten me, how does/can undervolting damage the CPU? I mean in any way...

I'm not sure if it can, but not supporting it in warranty means they don't have to deal with the people who might experience the system being too-undervolted and thus be unstable.
After all, if a pc suddenly stops and crashes all the way down to boot while working on something heavy, then that is a very valid cause for concern for many users. As InVasMani points out.
Better to just have it be a blanket ban just-in-case, to avoid having to deal with a ton of consumer support, and people claiming the cpus are bad (because they crashed during high load and too low-voltage.)

I'm very glad that AMD is rolling this out, and I hope intel follows suit.
I've undervolted my last 2 (intel) laptops, and have been very happy with the ~ -10-20% power usage on them. I very much hope dynamic undervolting just becomes one of those standard cpu features everyone has and uses.
Having cpus perform worse and hotter just to increase processor yields in manufacturing is pretty frustrating, and it's good to see that they're finally willing to start playing with it.

 

[InVasMani]

Someone enlighten me, how does/can undervolting damage the CPU? I mean in any way...

It shouldn't damage the hardware itself, but it can be pushed too far and cause instability by reducing voltage too greatly relative to the frequency paired with it.

 

[Mussels]

Someone enlighten me, how does/can undervolting damage the CPU? I mean in any way...

it can't except in weird circumstances where voltages have to be similar to each other (DDR ram has that) but in this case its more that you cant warranty for undervolting related instability