@nrp will we be able to overclock the ram and use the Curve Optimizer Voltage Offsets to undervolt the CPU’s to make them more energy efficient?
I have the exactly same question for the 7840HS. Does it support undervolting CPU ?
@Platos the 7840hs does not support undervolting or any other kind of overclock. But that’s on AMD. See https://www.amd.com/en/product/13041
Ah ok, thanks.
I dont like that behavior from amd to lock such features. Thats only to justify the 7940HS. Customer unfriendly.
But is it at least supportet to use different powerlimits? Or is this also limited to minimum 35W like described on this website you linked ? So you could not drive it with 25Watt for example over mainboard-settings (BIOS).
Yeah I’m still considering changing my order to a 7940HS to undervolt.
I am not 100% sure it is an upsell tactic alone. The 7840HS will use lower quality silicon, and the 7940 is not THAT much higher clocked. Which to me means there is a rather narrow range of silicon quality (which I assume has the bulk of the silicon) that is not quite qualified for a 7940HS. Ofc there’s also the U series that use the same chip (lower leakage current, lower peak performance).
All that is to say, there might not be that much you could even get out of the 7840HS, and rather than let customers be hopeful and then dissappointed, it isn’t too surprising to have that exclusive to a product where you might reasonably expect to get some real gains due to more variation in the silicon quality.
In other words: With 7940HS you know you’re playing the silicon lottery, and you know you can be dissappointed. With the 7840HS, you’re almost certainly going to be dissappointed, since the “silicon lottery” is rigged by design.
That is all assumption ofc, I would much prefer them to not lock it down anyway, since any small benefit is better than nothing.
I’m very certain it will be adjustable. I hope upwards as well - just that this is a brute-force approach compared to curve optimizer voltage offsets.
The only reason I am hesitating (besides price for essentially the same chip) is I do not know if the curves can be adjusted in the BIOS or not. On linux, there are no great tools to adjust that, recent developments (Universal x86 Tuning Utility for example) are windows only, and that left some software on linux out of date, not supporting these recent chips.
So I’d like confirmation these curve optimizer can be set in the BIOS.
Yeah this is absolutely an upsell tactic. Like the -k cpus from intel. Sure you can say it clocks a bit higher and so on, but the main reason is to make sure, they can sell an overpriced version. If it would be open, no one would buy this overpriced stuff.
So i would not buy it, otherwise you support their behavior.
I mean, it is a durty tactic of companys to block overclocking, but it’s even mote stupid to block undervolting.
I dont even know if intel is allowing it on their laptop cpus, which are not top-end. Like 15watt u cpu, p-series and so on. If i remember correctly, it was at least possible in earlier days.
And btw. i think you will not be able to benefit from higher wattages. Laptops are normally bottlenecked by the cooling. So you would probably not gain any perfomance with overclocking. But you would with undervolting.
But yeah, it’s a shitty behavior to block overclocking.
This depends on the workload. For me it’s mostly to reduce compilation times, which means it’d have time between to cool down. As long as the heatpipes can get it away fast enough, and the fans react fast enough, I’d hope to be able to have the CPU boost up with 65W for a minute max, and then another minute or so to cool down. Hopefully. The heatsink is quite small so maybe it can’t sink enough heat to matter even if it starts out cooler.
Anything I can do to make compilation faster tbh, since that directly correlates with productivity
Intel disabled undervolting in their Coffee Lake and newer laptop processors to mitigate the “Plundervolt” vulnerability. However, it turns out that was a mere recommendation so some OEMs enabled undervolting again, just for Intel to try to stop it again and a whole lot of other stuff happened. I won’t be able to explain it properly here, but you can read more in this Reddit thread.
Ah yeah exactly, it was plundervolt. Now i remember.
So they do not block it technically. I mean yes, it is bullshit what intel is doing there, but if their intention would be to block undervolt, they would just do it.
So Intels fault is their poor implementation, not “blocking” undervolt. They could do that more easily i think.
And yes ok, if you need overclocking for very short time, it can be helpfull. Not sure hoe amd does this on their laptop-cous, but on intel you can just set pl1 qnd 2 and TAU to let it overclock. But yeah, if there is a Limit for the wattage, it will not work.
As long as the 7840HS meets the specifications, I won’t be disappointed.
Each to their own. I didn’t spend £2k on a laptop as an undervolting project, it’s a productivity tool.
With the 7840HS I don’t play the lottery. I can expect it to match the spec. When buying the 7940HS for the few % performance improvement, I don’t play the lottery neither. Only when you buy the 200 extra to use undervolt/overclock you are participating in the lottery. You can get more out of the silicon, or you payed 200 more for a chip that barely reached the 7940HS threshold.
Sorry misleading wording - what I meant was pretty much exactly what @flovo said
So yes, with the 7840HS you get what you pay for, no dissappointment - only if you were expecting to get more out of it than the specs via undervolting, which you might have if it supported undervolting. Not an argument for disallowing it even if the gains are minimal, but here we are.
Also, I don’t think many are seriously buying the 7940HS for the few % extra performance alone (not judging if you do ofc). Even system integrators seem to agree, most thin and light gaming notebooks with dGPUs - which would benefit from better efficiency if anything - top out with the 7840HS (or GPU efficiency is just that more important than CPU efficiency for gaming).
No, the lottery and hope for efficiency gains from undervolting is, IMO, an intrinsic part of the whole package, including the price. Bit stupid and annoying, but given the need to pack silicon into classes to sell them, it does kinda make sense.
Undervolting nearly always brings a lot of efficiency gain. That’s not sth. like the 200MHz boost you get from the more expensiv version.
That can make easily 25% more perfomance/watt, especially on laptops, where you are cooling-limited.
Never had a case, where i was not able to gain alot of perfomance because of undervolting.
Undervolting is the must-have for everything, if you are not calculating scientific operations.
“I wan’t a stable pc, i’m a productive user” is just other wordings for “i have no clue about it”.
That’s totally fine, because if you are to worried about it, you do not have to do it. No one have to do it. But don’t claim undervolted systems are not suitable for productivity (or even unstable). Then you just have no clue about the topic. I run 100h of hours encoding with GPU and CPU undervolted. Games too. Even AI-things.
Undervolting gives you perfomance/watt like the next generation of Hardware or even better.
Yep, it can be pretty powerful on ryzen laptops, I want it for productivity reasons.
BUT, if you want the best undervolt possible, e.g. best undervolt for each core, it does take quite a bit of time to verify stability, since sometimes problems only start when the CPU core is NOT stressed, so to be sure a day or even more is needed to verify each change. And you need to verify each core individually, since lowering multiple at a time means you don’t know why it became unstable. Or you can just do a general undervolt and be limited to your worst core (missing out on most of the benefits).
So I can understand why it is seen as too daunting if you just want it to work. But even an all-core undervolt is very much worth it.
Yes, “only” allcore-undervolting is for sure VERY worth it. easily you can get 25% more efficiency. But don’t forget your graphics card. It is also worth it.
Sadly i can not do it on my current laptop, but at least i can do it on the graphics-card in my laptop.
But undervolting is not only about more perfomance, it is also a huge thing for battery-life and fan-noise.
You’re inferring an awful lot there from a simple sentence, then falling down the wrong rabbit hole there Platos. Stick to undervolting laptops, and avoid amateur psychology is my advice. 
But don’t claim undervolted systems are not suitable for productivity (or even unstable).
I never claimed anything of the sort.
I would like to know too. I didn’t realize on AMD’s spec sheet that there are other differences such as EXPO support and Curve Optimizer. The chip supports it, BUT will Framework expose these features in their firmware?
If not, then we’re back to the 7940HS just being a slightly faster 7840HS. The ability to support faster than DDR5-5600 and tighter timings may help further differentiate the R9 over the R7 to make that $200 up charge more palatable.
Again, up to the firmware now.
If we can undervolt using the curve optimizer in software, that would at least be safe and potentially bring even more performance. Exposing undervolting through the bios will allow the uncareful ones to brick their machines.
I undervolt my Steam Deck cpu curve using software and it brought a noticeable increase in the time cores can sustain max boost and reduced heat.
Yeah AMD (and Intel) have been… extremely solid in delivering silicone lately. I’ve gone through a couple dozen modern Ryzen CPUs and tested them and I’ve never seen one miss specs. RDNA is significantly newer of course and I have seen all of two in person so far and used them/benched them, but both hit their specs.
TSMC has been doing their job for quite a while now and has a handle on it haha. Chips are pretty maxed out of the box and consistently hit specs.
To be fair, they develop the silicon, test a bunch of samples, and then decide on how to seggregate their dies into classes. So the silicon never follows spec, but always hit it - if that makes sense. They test each chiplet (probably) and then each final CPU and bin it according to its silicon grade, including which cores have defects (-> lower core count parts, e.g. R3 or R5), how little leakage current they have (-> low power U class chips), and finally, how well they scale with higher power (R9 or R7).
Same for Intel ofc
Doubt there are many (or any) instances of CPUs not meeting spec due to the chips themselves
