Hacker News Reader

Comment by userbinator | Hacker News Reader

The existence of ThrottleStop and the experience of many others shows that this is also a persistent problem on Windows --- CPU is barely getting warm, and sometimes the fan doesn't even turn on, but then gets throttled to some insanely low speed. Search "800MHz throttling" for plenty of complaints. Disabling all the DPTF and other extra power management bloat puts the CPU speed and performance back where it should be.

View on HN

IEC 62368-1 places an upper bound on the temperature of materials in electronic devices that are likely to come into contact with users' skin, and that's incorporated into EU regulations[1]. You have two ways to achieve that - add enough cooling such that it's impossible for the device to ever produce enough heat to get that hot, or throttle components that are generating heat if the device is getting close to that temperature. The former requires a cooling system that's going to be significant overkill for most users most of the time, resulting in more expensive hardware that's bigger and heavier.

So, instead, you see the latter. This means that, yes, even if using official Intel DPTF drivers, systems will still occasionally thermally throttle. They may do so even if the CPU itself is not at a dangerous temperature, because the rest of the system may still be too hot and the CPU is the easiest tunable knob in terms of heat generation.

So, yeah, you can still obtain better performance by disabling DPTF and overriding power limits. And in the process you'll end up with a system that will become hotter than permitted by international standards, and you'll risk various types of failure ranging from inconvenient (the adhesive sticking things like rubber feet to the bottom of the machine tending to melt) to expensive (the unexpected levels of thermal expansion cycles causing components to fail earlier). It's fine that you're not worried about these things, but it's just straight up wrong to recommend that people override those limits without being clear about potential outcomes.

[1] https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CEL...

(Edit: replaced IEC 60950-1 with 62368-1, which is the up to date version, and linked to EU incorporation of that spec)

I feel like you're being overly charitable to laptop manufacturers. If such manufacturers always considered the thermal limits of their products when advertising their performance, then I don't think many people would complain.

Since that isn't the case, however, many people get a laptop that isn't capable of performing the tasks it's advertised for. The excuse of "well, we didn't think you'd really need that much power" doesn't really cut it, when said power was strongly advertised on the box.

Oh yeah the fact that a given laptop has a given CPU certainly doesn't mean that it's going to run at the same speed as another laptop with the same CPU, and it's unfortunate that manufacturers don't tend to give users the tools needed to figure out whether the thermal limitations are going to affect them. But review benchmarks do generally show the performance distinctions fairly clearly?

I'm sure there are benchmarks out there that comprehensively cover thermal behavior of laptops, but many sadly don't. I've had many laptops with great benchmarks that completely fall apart under prolonged load (>1 hour) Usually because the benchmark was 10-30 mins, during which the other components didn't heat up enough for throttling to occur.

I don't think the average laptop user should need to concern themselves with acquiring the necessary knowledge to discern good laptop benchmarks.

> I don't think the average laptop user should need to concern themselves with acquiring the necessary knowledge to discern good laptop benchmarks.

They don't. The average laptop user doesn't have sustained workloads that keep the system running at 100% capacity for over an hour, so they don't need and shouldn't want benchmarks reflecting such a use case. They're genuinely better served by benchmarks reflecting workloads that come in shorter bursts.

Reply on news.ycombinator.com