Hacker News Reader

Comment by xxpor | Hacker News Reader

>We are aware of a limited number of reports online claiming that excess voltage while overclocking may have damaged the motherboard socket and pin pads.

Well, if the overclocking part is true, shrug. Not that it's great, but it's also not completely unexpected for bad things to happen when you OC.

View on HN

This specific case (EXPO) is a grey area.

I personally don't like overclocking, yet I've always enabled EXPO, for many reasons.

First, because the producers of my RAM modules advertise it very clearly as feature; AMD advertises it quite loudly (https://www.amd.com/en/technologies/expo) as well. I think (not 100% sure) that it's even often set in RAM benchmarks .

Modern systems have very aggressive frequencies/dynamic frequency scaling, which blurs the limit of what constitutes overclocking. Intel sells unlocked models of CPUs that are supposedly can be "overclocked" (https://www.intel.co.uk/content/www/uk/en/gaming/overclockin...).

Many RAM producers sell modules based on their tolerance, so the fact that they're pushed to the limit - within the spec - is part of the sale.

Having said that, I definitely draw the line to using voltages (or anything else) that are outside the specification, and I consider crossing it overclocking.

Ultimately however, EXPO is a specification, so I don't consider it overclocking.

On AM4 many boards set too high IOD voltages when enabling XMP. Afaik no detrimental impact except about 10 W superfluous power draw.

It doesn’t seem particularly far-fetched that EXPO might suffer from the same issue except it accidentally sets batshit voltages instead of just too high. Afaik the SoC and IMC supplies are unsophisticated, there is no DVS going on, just BIOS setting a particular voltage in the VRM.

I agree. If the RAM module is on the QVL, the RAM module has an EXPO profile, the EXPO profile should be considered as in spec IMO.

>Ultimately however, EXPO is a specification, so I don't consider it overclocking.

The O in EXPO stands for overclocking, and if that wasn't clear enough AMD themselves state using the whole word that EXPO is overclocking numerous times repeatedly.[1]

[1]: https://www.amd.com/en/technologies/expo

Indeed. Overclocking in 2023 is different than it was 20 years ago. You probably weren't overclocking back then without some inkling of the risks/dangers of eg. increasing the voltage to your CPU. And because you knew the risks you were probably stepping up very, very slowly and stress testing at each increment. Now you can kinda just grunt "me want faster computer" and push a button and it works. Unless it goes horribly wrong and catches your CPU on fire.

Chipmakers meanwhile can have their marketing cake by selling chips with certain properties, but pushing software that makes them "better than advertised" which seems like a great value. And then eat the legal cake by avoiding liability when chips fail because "running this hardware outside of the default specs is not supported."

"the pre-programmed voltages used to support EXPO memory overclocking profiles"

Bad things shouldn't be expected when using pre-programmed voltages.

And chips shorting out beyond thermal limits shouldn't be expected ever.

AMD recommends that customers use EXPO then they claim it's warranty-voiding unsafe overclocking.

And doctors recommend safe-injection sites, while also claiming that shooting heroin is unsafe. Shocking!

It’s called “harm reduction” — i.e. “if you’re going to do stupid thing X anyway, you may as well use our weak guardrails for it, to hopefully minimize the damage you’re causing compared to doing it through some random third-party solution.”

Oh please. They advertise their performance based on using things like EXPO. It's insanely anti-consumer for them to turn around and say "if you want to reach the performance we sold you this CPU on, you need to turn in this feature that voids your warranty and could potentially blow up your CPU". I don't understand why you're defending them here.

AMD has generally presented benchmarks with market-typical XMP/A-XMP/EXPO memory; their CPUs benefit more than the competition from that. Now some people have the urge to turn around and go “well why you’d do that, what are you, stupid?” for some reason.

> Well, if the overclocking part is true, shrug. Not that it's great, but it's also not completely unexpected for bad things to happen when you OC.

IIRC the core issue is that some motherboards loaded EXPO overclocking profile (basically, overclocking parameters for memory that are written on memory stick for "guranteed", "tested" OC performance) and caused overvoltage of CPU.

I don't remember if they found whether that's mobo reading profile wrong and burning up CPU or just some RAM sticks coming with too optimistic profile and mobo doing "well, I trust you, here are extra volts".

The wild part is that it is AMD tech (Untel has similiar equivalent in XMP), yet it also voids warranty as any other overclock

In the old days that would be true but nowadays OC is so mainstream that temperature sensors are expected to keep the SoC within safe limits, shutting down if needed but preventing damage.

Something escaped design models and test scenarios for this to happen with end users.

This doesn't sound like a thermal issue with the chip itself though, more in the physical socket.

In the article it said that apparently in some cases excess voltage damaged the temperature sensor which caused chip to not register overheating and just ask VRM for more power

Sounds like the thermal sensors don’t fail safe. If the thermal sensor isn’t returning readings or those readings are not within a reasonable range the CPU should shut down and then fail subsequent POST tests with a descriptive error.

Overvolting can do funny stuff to electronics. It could "just" go out of spec and start lying by some degrees.

I wouldn't be surprised if next gen had that duplicated, it can't cost that many extra transistors

> In the old days that would be true but nowadays OC is so mainstream

I do not agree with this statement. If anything, overclocking is less prevalent than ever before, despite what all of the LTT's and friends make videos about.

The reality is modern CPU's are crazy fast, and auto-overclock (boost/burst) all on their own. There's little reason to push the CPU's further - and more significant gains can be had much safer by upgrading CPU generations than squeezing out a few extra percent of performance in benchmarks.

There will always be enthusiasts that want to overclock - but I don't think we can consider overclocking mainstream.

LTT has repeatedly said that overclocking under normal circumstances has been worth less and less as GPUs and CPUs have been increasingly built to ride thermal and power limits at stock. And that overclocking can even hurt performance by disabling that behavior.

For CPU clock, I agree the gains are marginal, yet BIOS settings still allow the end user to trivially further bump up the clock. There are young people with a lot of time on their hands who will do it for that 3% uplift.

Further, with memory we have things like XMP which again are blessed by the vendors and are 100% within the definition of overclocking. Just because a memory module can run at XMP rates doesn't mean the motherboard or the CPU have the signal integrity or power integrity margins to support the timings. And this is something component vendors can't optimize from the factory (only system vendors can), because it depends on multiple parts.

In the old days one couldn't enter BIOS setup and find even a single mention of overclocking, and the term itself was taboo because it meant operating the system outside the design spec. Nowadays a lot of vendors use OC in marketing. I think this alone makes my point about being mainstream.

In a way, OC is absolutely mainstream now. It's now integrated into the product itself, which dynamically clocks itself depending on thermal headroom.

There's thus minimal need for users to do any manual overclocking any more.

I believe we've seen issues like this before from motherboard manufacturers, who are pushing more voltage to the CPU to get more performance out them. Recently is was on the Intel side of things.

This sort of thing was even a problem 13 years ago. (https://www.overclockers.com/socket-1156-burning-issuesis-fi...)

It's not new- component OEMs just screw up sometimes. I suspect that PGA sockets have less problem with this, though AMD has had just as long to get LGA right given their server processors have used it for nearly as long as Intel has so while it does happen, there really isn't any excuse for it.

Reply on news.ycombinator.com