Indeed. Current testing that measures just the temperature is basically the same as 20 years ago - when we had recent innovations like CPU shutting down the system when overheated instead of just frying itself. For the longest time when thermal throttling was introduced it was also very rudimentary - basically a binary situation between 100% performance or going down to like 25% when overheated with nothing in-between.

With modern CPUs pushing tons of extra watts for sake of marginal performance gains as well as precisely surfing the line of extracting vast majority of possible performance in given conditions it’s just terribly inadequate.

Well, the point of such lists, as well as any testing, should in theory be about judging products on being fit for purpose. What specific metrics you choose, how you weight them and how they are presented are all critically important. Especially if what you search for is not strictly the best possible performance and instead want a product that strikes a good balance between its various functional aspects as well as price.

For coolers you typically see them arranged mostly by temperature they reach under power virus workloads. Which IMHO is just fucking useless in light of how modern CPUs behave. Such metric will struggle in differentiating between multiple coolers that technically cause some thermal throttling as it doesn’t account for how much they actually throttle.

On the other end you have the option to just ignore temperature and instead measure performance impacts on actual workloads. Which is far more relevant to actual use cases and it allows for good differentiation between non-overkill coolers. But the cooler makers might not like that when it shows no difference between a $50 and $300 cooling solution.

Well, if you use it mostly for GPIO, you don’t really need any of that. At the same time though you likely don’t need a Pi 5 (or likely even Pi 4) to begin with as any Pi will do GPIO without problems.

That said the cooling requirements are often quite overstated - outside of niche uses where you actually run at full throttle for sustained periods of time, you don’t really need to care about it. Sure, it will throttle if you run stress-ng for more than few minutes, but so what? If your workload hammers the Pi at 100% of all cores for long time, you are using wrong platform to begin with.

People on regular basis outright hilariously overstate the impact of thermal throttling during normal operation. Thechpowerup did a pretty good experiment with the 7950X.

The big conclusion is that the issue has large impact only on artificial benchmarks that behave more like power virus than actual workload. Heavily parallelised productivity workloads can suffer a bit, but nowhere near enough to justify high end cooling for vast majority of users, especially on CPUs that aren’t strictly top of the line. And gaming is almost hilariously unaffected.

The silly thing is that Pi5, at least currently, tends to idle at basically the same or higher power than your average Intel N100 NUC-like PC. This might change in the near future with firmware upgrades (just like it has changed for Pi 4). Those N100 PCs can be had for ~200€ with SSD, memory etc - ready to go in comparable form factor.

Pi does still make a ton of sense if you keep it rather basic, but equipping it fully with a nice case/cooling, m.2 SSD and proper PSU does bring it within spitting distance of price of basic NUCs. Against those it loses pitifully in performance department (as well as efficiency!) - which you likely cared about at least somewhat if you went out of your way to add performance-related stuff to the Pi to begin with.

That doesn’t at all mean there will not be a dotcom bubble burst equivalent in the space. Right now there is a huge torrent of money flowing into it which is in stark contrast to how relatively few revenue streams have materialised and their actual long term financial viability is fuzzy.

Well, it does look like it’s just a scheduler fix at the very surface level. On the other hand it does seem to need some firmware support and presumably there is some reason why it only supports 2 games. So maybe it is something more complicated?

Main actual compromise with soldered RAM is that you have to pay through the nose for sensible amount of it and do it all up-front. Depending on your use case for a decent laptop today you’ll want 16 or 32GB RAM - so that you aren’t hampered by it 5 years down the line or so. Some vendors ask plainly absurd amounts of money for it.

As far as laptops go though I have just a general apprehension. It obviously depends on your specific use case, but for a ton of situations a standard PC along with a cheaper laptop will be a better option. This is especially egregious with gaming/workstation - if you think about getting decent GPU/CPU, an external monitor/keyboard/mouse etc, account for upgrades over the years - you quickly run into situation where desktop PC and cheap laptop together cost much less money for the same performance.

On the other hand there is something to be said about convenience of a setup with single powerful laptop and a dock that plugs into your peripherals.