I think you could probably make a convincing argument that what software you choose to make is protected speech, and that this law is unconstitutional.

Plus, with klipper, itâ(TM)d be even worse to implement. The pi attached to do the bulk of the processing is optional - the actual printer is the low power bit that does nothing but move the print head. Effectively this would ban klipper entirely, because the printer side MCU stands no chance of verifying a part.

Can you give an example of where it gives bullshit results? In general it seems to align pretty well with general overall chip performance. Obviously itâ(TM)s not going to give you performance for a specific task, but it does seem to correlate pretty well with overall system performance.

I think itâ(TM)s easily explained. Most people on slashdot are early tech adopters. When ChatGPT 3.5 burst onto the scene, they tried it. They tried using it to generate code, and they got laughable results. Theyâ(TM)re now convinced that AIs generate terrible code because theyâ(TM)ve not since gone back and given any reasonably recent Claude a go.

Geekbench runsâ¦

- clang
- gz
- WebKit
- pdfium (chromeâ(TM)s pdf renderer)
- mobile net (a commonly used open image classifier)
- grep
- sqlite
- astc
- bc7
- dxtc
- DeepLab
- content aware image resizing

Almost every single test it runs is a commonly used open source bit of software, and those that arenâ(TM)t are implementations of commonly used algorithms. I honestly have no idea where the idea that âoegeekbench doesnâ(TM)t have realistic testsâ came from, but itâ(TM)s bullshit. Is it going to accurately model *your* mixture of tasks - no. But are the tasks it chooses a reasonably representative sample of things people do with computers? Absolutely.

To be fair, Appleâ(TM)s RAM prices are atm lower than PC RAM prices, but I acknowledge that we live in interesting times.

The really neat trick is to run Linux in Appleâ(TM)s virtualisation framework, and then install the libraries that let you call into Rosetta whenever a process needs x86.

No, the usual things are where the company making the software has (usually for PHB related reasons) decided not to press the compile button for another platform. The fact that even Microsoft is in this camp shows how ludicrously bad the windows on arm situation is. All MS needed to do was take the Apple approach of âoebuild it for the new hardware or youâ(TM)re not getting to release it on our storeâ, and then âoeweâ(TM)re removing support for legacy x86 applicationsâ, and then âoeweâ(TM)re removing support for â¦â. Unless you push companies to do the right thing, theyâ(TM)ll just sit there with their thumb up their bum going âoeour software still works on a 486, and thatâ(TM)s the way we like it.

Itâ(TM)s not even like MSSQL is an outlier. The Minecraft bedrock server doesnâ(TM)t have an arm build. It has a Linux build, but not an arm one.

So what youâ(TM)re saying is âoeitâ(TM)s difficult to compare different designs because theyâ(TM)re different designsâ?

Yes the memory is on the package, and that helps make it fast⦠but thatâ(TM)s literally the point, designing a system to have memory integrated into the package makes it (among other things) ludicrously fast.

Even at enormous parallelisation these days the M5 is up there.

The only two non-overclocked CPUs that beat the M5 Max at parallel work on geekbench are the threadripper 9985WX, and the 9975WX. Those are 32/64 core CPUs and the 18 core M5 Max is basically as fast.

For graphics, the M5 Max is also basically competing with the best. It appears to be around the same performance as a 5070Ti, while somehow consuming only 25W. I have no doubt that the M5 ultra will be competing with the 5090.

I mean, yes, they expect laptops to not cook and crush their laps⦠itâ(TM)s literally the first design requirement. That said, even if chonk *was* permitted, it appears that the MacBookPro would still win. The M5 Max MBP appears to be quicker even than giant 19â inch thick workstation laptops with high end CPUs and discrete GPUs. Even more so when you unplug them, and run off battery.

https://youtu.be/ieog1kj1AQE

The reason to âoepreferâ them is that you can get directly comparable results for many machines without having to go and test every single machine. The trade off is that the results you get are only an approximation of the tasks you might want to do.

If you want to run blender, and do movie quality frame renders, then the various benchmarks that test that are a great way to compare. If you run solidworks and need to know how quickly itâ(TM)ll compute a complex union⦠well⦠good luck, youâ(TM)ll need to figure it out yourself somehow.

Thatâ(TM)s not adobe knowing where their bread is buttered, thatâ(TM)s Apple having designed a ridiculously fast CPU.

I mean, the lower level managers wonâ(TM)t have a job to do once they have no one to manage. Itâ(TM)ll be reduced to just the managers who are needed to figure out what the product is.

Given the small scale plans that AI is already able to make, I really donâ(TM)t thing itâ(TM)ll be long before it can choose an architecture that works, make a document describing that architecture, then follow step by step instructions to build it. At the moment it breaks down small problems into immediately actionable steps, and does them. It wonâ(TM)t be long before itâ(TM)s able to do that recursively and then iterate what the best design is. It also wonâ(TM)t be long before itâ(TM)s better at it than software engineers. We typically focus on one area, thinking about the general effects on the rest of the system only. An AI will be able to make detailed plans that consider all the interactions with the rest of the system.