I've seen in Linus episodes that even gamers are serious thinking going to others fr GPU because of drawing to much power lately! If the GPUs could figure out using ARM to power the GPUs they might get a leg up!
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple should convince Dr. Lisa Su (the CEO of AMD) to lead their chip division
- Thread starter Retskrad
- Start date
- Sort by reaction score
Power is an issue yes. I play games on my Mac whenever possible even though I have a 3080 Ti gaming computer. I also had to “ban” that gaming computer a few times over the summer when it was too hot out and my AC PLUS portable AC couldn’t keep up when the PC was OFF! I didn’t want to add on to the heat issues.
I sure hope you were being sarcastic! this seems like they shooting at Apple instead of Intel like they did in the past!
Didn't watch this video, but I did watch Su's presentation—she was comparing a 15-month-old Apple chip (M1 Pro) to an AMD chip that won't be released until Spring 2023! Plus she didn't compare the 7940HS to the M2 for SC speed (initial reports are they're about the same). And she conveniently didn't have a bar graph for relative GPU performance; surely the 7940S's integrated GPU is less powerful than the M1 Pro's (I assume MaxTech mentioned that). Finally, she used Cinebench, on which Apple Silicon has about a 10% penalty because it's not equivalently optimized for AS and x86.
A more reasonable comparison would be the 7940HS vs. whatever Apple M2 chip has a TDP comparable to the 7940S (not sure if that would be the M2 Pro or M2 Max), using a broader, more platform-independent benchmark like Geekbench. And also compare the battery life of the M2 Max to whatever combination of 7940HS + AMD discrete GPU is needed to equal the M2 Max's GPU performance.
Last edited:
Okay, I could not, and still cannot, make much sense of your comment. No problem though because I do not care much.
I posed a vidoe by Max where he compared a AMD 5900HS vs. a 14" MBP and in that video the AMD kept pace with the MBP but at nearly 1/2 the price tag.
Most consumers don't start picking laptops and comparing them based on TDP
Here's another comparison video, these folks are fairly well known in YT PC testing community.
You may want to ignore the cinebench portion of the video since that's a pain point for some folks. The other benchmarks show. the AMD and Intel beating out the M2. For instance at 6:52, we see the Handbrake benchmark. Matlab, is a rather poor showing for the M2 as well. There are benchmarks where the M2 does outshine particularly in the video production. I think the point I'm making is that the M1 and M2, when you sweep away the hype are not alwas the fastest, but tend to be the most expensive. AMD and Intel not only keep up with the M1/M2 but in many cases exceed it in various metrics, but at a much lower price point. Power efficiency is where Apple certainly has the crown, but that's just one category.
Still x86-64 not native. Certainly not optimized in any way for Apple silicon. I’d say it is a rather poor showing by Mathworks that they can’t even release a compatible version after 2 years (beta is available).
He says in the video during the test "Video encoding in Handbrake is a heavy multi-threaded task that utilizes AVX instructions on x86 processors which do appear to give them an advantage over Apple's ARM products."
Apple Silicon doesn't support AVX. Apple says "If your code includes instructions for the SSE, AVX, AVX2, or AVX512 units of Intel processors, update that code to support Apple silicon. The best alternative to processor-specific vector code is to use the Accelerate framework, which provides a vast library of vector operations optimized for all Mac computers."
R's been using Iain Sandoe's port of GCC (gfortran) to Apple Silicon for ~2 years for native releases, Scipy's been using it since late 2021, and GNU Octave is fully native on Apple Silicon as well. I'm pretty sure the Clang Fortran compiler (flang) has been heavily revamped/improved within the past year too, though I'm not sure if it's 100% ready for prime-time use. I don't think MathWorks has a good excuse here.
Selfishly I'd be happy if MATLAB never sees an Apple Silicon release, because the more people are nudged away from writing new scripts/libraries in that awful aging language/ecosystem the better. For legacy inertia reasons there's a lot of neuroimaging and biomedical research libraries/toolboxes written in MATLAB so it's still used quite actively in my field, but my goodness does it break almost every nice modern language convention. No proper package manager/dependency resolution (you just download a big .zip of .m scripts and hope you've installed the prerequisites), plays poorly with running scripts from the terminal or other IDEs, giant upfront costs if you don't have an institutional license, encourages all manner of terrible coding practices and breaks people's brains for learning other less-oddball languages... Engineers can keep MATLAB if they want, but the sooner the research sciences are pushed over to Python and R the better.
If I had a choice, I would prefer Julia to become the de facto language in science and engineering.
Julia hasn't worked its way into my corner of research just yet (cognitive science & physiology), but I've heard good things about it from my friend in epidemiology!
If it lets you define/call functions with regular named arguments (e.g. sample_rate=100), isn't deeply integrated with a proprietary IDE with a UI straight out of 2010, and uses something other than '%' for comments, it's already leagues ahead of MATLAB (no, I'm still not over the 2 months I had to spend learning MATLAB so I could port an EEG preprocessing toolbox over to Python).
Julia has other problems. First of all, it takes longer than Python/Matlab to run code for the first time, although it has improved a lot in recent versions. And secondly, package maintenance in Julia is not as good as in Python/Matlab because the Julia community is less well funded.
You may find interesting this set of packages.
Julia Neuro by JuliaNeuro
julianeuro.github.io