Hacker News Reader

Hacker News

Comment by uweihfuio | Hacker News Reader

Parent Full thread

uweihfuio

No, but you can see that AMD beats Apple in Power-Performance:

https://www.cpubenchmark.net/compare/AMD-Ryzen-7-PRO-5850U-v...

AMD 5850U 1253 scores vs Apple M1 1004 scores

Waiting for apple fanboys telling me something like "but TDP is not a good measure"...

View on HN

Not an «apple fanboy». You should justify that TDP is the measure we are looking for, instead of just the available measure. TDP is a maximum, and an average one for that matter.

Those measures should use something like `powerstat` results standardizing other conditions - in fact, they should use benchmarking (just like the numerator of the numbers you report does, but not the denominator). This system here has a TDP of 6W, but uses 3.3W during light use, and peaks around 7.5W.

Sure, there could be a literal interpretation of "Performance/Power" which somehow could link speed and TDP (benchmarking at the same stated TDP), but (°) it would be imprecise as TDP is an estimate (so consumption should be benchmarked) and (°) it would not represent some more useful scenarios like "which system allows squeezing more working time out of a battery" (which is not represented by considering maximums - especially given technologies like big.LITTLE).

Similarly, Intel often states TDPs of 25W, 45W, 65W, etc... while peaking actual power usage at nearly double those numbers during "power boost."

So unfortunately TDP has basically become a marketing term and we're back to literally measuring power usage instead. Which of course gets complicated when you consider that different machines have wildly different heat management systems.

"TDP" as such is going away for Alder Lake. Intel is now using a much clearer base power and turbo power number. So for example, the new i5-12600K has a Base Speed of 3700MHz @ 125W and Turbo Speed of 4900MHz @ 150W (if and when thermals allow it) [0]. The old marketing would've just called this a TDP of 125W.

[0] https://www.anandtech.com/show/16959/intel-innovation-alder-...

M1 Max operates with a peak of 50W. With a top performance difference of let’s say 20% (unattainable by Intel), I care more about having a computer that runs at 50W tops than a water boiler at 150W.

It’s fascinating to me that society was willing to accept laptops that could not safely be used on a human’s lap for so long. My ‘old’ 16” MacBook Pro could probably cook an egg.

Isn't that a part of the reason why many times they were marketed as "notebook PCs" instead of "laptops"?

if your electric kettle is only 150W then it must take you a long time to make tea.

Why are you comparing with the lowest spec M1 chip?

I think this might be the comparison you were after? https://www.cpubenchmark.net/compare/AMD-Ryzen-7-PRO-5850U-v...

I compared 15W CPUs and you putted 30W CPU (M1 MAX) in the comparison... Nice try but such tricks will not work with me.

To be strictly fair, the AMD Ryzen 7 PRO 5850U is a 10-25W TDP depending on manufacturer configuration, not nailed to 15W. Do cpubenchmark verify that it's running at a 15W TDP when coalescing the numbers?

There is no guarantee for that, but usually 5850U is used in ultra-thin laptops and they don't have enough cooling for more than 15W... For example, CPU in Thinkpad X1 Extreme GEN 4 is limited 27 WATT (at least models without vapor chamber) and it's a laptop with quite a good cooling and larger size.

the m1 max is a 90w cpu

Source?

That Ryzen chip has 20% slower single thread performance, which is what really matters for desktop workloads

That's significantly less true than it was in the past. I think it's pretty easy to argue that the browser is the most important desktop app now and all modern browsers are multiprocess / multithreaded.

Beyond that, almost everything that does signal processing of some shape or another (audio, video, whatever) is multithreaded these days, so that one's out too. Compiling code is multi-process too.

I actually struggle to think of important work loads that are single-threaded these days. Obviously most simple apps are, but mainly because they don't use enough CPU to bother optimizing. We're more than a decade into multi-core systems being the norm and most CPU-heavy apps have taken advantage of that.

Performance of UI is mostly single threaded

Sure, but there are very few UIs that require 100% of a modern CPU core. Exceptions exist, but they're just that, exceptions.

I'd go one step further and say that if modern apps are doing that that they should move the CPU heavy tasks out of the message thread. (And most have, again, since we're like 10-15 years into multi-core systems being the default.)

I struggled to understand that you compute score/TDP

> Waiting for apple fanboys telling me something like "but TDP is not a good measure"...

Not an Apple fan, but TDP is not even a measurement, it's a specification of the heat dissipation required for that CPU. So in theory it corresponds to the maximum power dissipated by the CPU under load, but in practice two machines with the same TDP may be equipped with differently dimensioned heatsinks which will influence your metric. You should consider using a real average power consumption measurement to establish your metric.

> So in theory it corresponds to the maximum power dissipated by the CPU under load

I think this isn't quite right - TDP is the amount of heat the system needs to be able to dissipate in order to prevent thermal throttling at a given base speed. If the system can dissipate more (and provide more power), chips can clock higher to a turbo or boost speed.

yes, that's why I put "in theory". But you're right, I should have written something like "with a simplistic perspective" instead.

Preemptively calling anyone with some opinion an “Apple fanboy” isn’t nice, or probably accurate.

Not in single core. For multi core it’s probably the SMT, but that’s a security risk.

Reply on news.ycombinator.com