danox

Apple is hardly behind but they are on a different path and actually are able to execute on hardware and software Apple doesn't need Google the Tensor is 5 or 6 years behind Apple right now and Google is a long way from running AI locally on their feeble Tensor SOC'S and Samsung is (hopelessly) even further away and Microsoft isn't even in the mobile ball game right now, note Wall Street is clueless about this little detail in their AI Hype train.

AI on the edge is the pathway Apple is on, phoning home and having the end user wait is the path Google, Microsoft, Meta, Qualcomm and Samsung are on. Why? Their SOC'S/OS software are way behind Apple. 

What Apple does need to do/consider is leveraging the full benefit of Apple Silicon, the low power/wattage, speed, UMA memory combined with a OS optimized to the hardware just screams out, the M3 Studio Ultra with 256 gigs (coming at WWDC) or a Mac M3 Extreme (CarPlay wasn't all for nothing) Pro tower with 512 gigs just says welcome to the world of inference on the Mac. Apple is/has been working on software tools to support developers for long time.

https://arstechnica.com/gadgets/2020/08/apple-explains-how-it-uses-machine-learning-across-ios-and-soon-macos/.  From 2020 before the current AI Hype

https://venturebeat.com/ai/apple-researchers-achieve-breakthroughs-in-multimodal-ai-as-company-ramps-up-investments/

https://daily.dev/blog/mm1-the-advanced-30b-parameters-multimodal-llm-from-apple

Fidonet127 said:

What I'd really want is a AI or something that is easy to use to manage mail and messages. At least beef up ShortCuts a lot to be able to work with mail and messages. Please add the rest of the iWork apps with robust ShortCuts. ShortCuts has a lot of potential, but is stuck with the iOS mentality and is no replacement for AppleScript.

The Mac Pro wasn't a massive failure, except for two camps, those needing to add GPUs and those needing massive amounts of memory. Apple has yet to allow eGPUs for ASi Macs, and without a change in the chip(s), may not ever allow eGPUs or GPUs. Memory is probably doable, though I'm not sure there is enough of a market. This isn't the trash can Mac. It has lots of cooling, slots to add cards and you can add lots of internal storage. 

tmay said:

The end result of this is that the EU is becoming even less competitive; more regulation in lieu of increased investment in innovation.

Perhaps that $2B will be used to mitigate some of the issues plaguing farmers, now rioting against both EU regulations, cost of operation, and competition from other countries where EU farmer inefficiencies are pronounced.

While that is going on, the EU is a dangerous crossroads with its auto manufacturing, as Chinese imports are poised to expand into the EU. 

tht said:

danox said:

tht said:

Very nice test.

And to think, this see-through photon-to-photon latency needs to drop by half to 6 ms to expand headset use cases. Crazy how much compute and lower latency this product needs to reach maturity.

10 finger typing on a virtual keyboard is going to need some bonkers hand tracking too. It may need to use a different keyboard layout to do really do it.

The Apple Vision already the best needs a M3/R2 faster GPU and more software optimization. In short Apple needs one more generation which Apple probably already has in the testing lab now due to product lead times (the M3 is already here the only question is the R2 done?). No surprise bandwidth/SOC size/battery/wattage are the holdups. Who do you think in the tech world currently is closer to solving all those challenges to stay ahead of the competition? Hint a vertical computer company with in house software/hardware. Example the Apple Vision wouldn't be possible using Intel, AMD, Nvidia and Qualcomm who in time wants half your revenue for two chips.

From the link www.chioka.in/what-is-motion-to-photon-latency

“To improve motion-to-photon latency, every component in the motion-to-photon pipeline needs to be optimized. For a movement to be reflected on the screen, it involves the head mounted display, connecting cables, CPU, GPU, game engine, display technology (display refresh rate and pixel switching time).” 

I'm not saying Apple can't do it. I'm saying it's amazing how much compute power is needed. 8K per eye, 180 Hz framerate, 2x to 4x to the camera pixel counts. I'd like them to get to a point where you can simultaneously use a 220+ PPI external monitor along with the virtual displays. Read fine text on an iPhone. Everything has to double or quadruple to do that.

So, a doubling everywhere. That's 70m pixels at 180 Hz, multiple streams of 24MP cameras (?), doubling of hand tracking frame rates. That's a 10x increase in compute power needed. So, R4 level that has 10x the throughput. Also would need a variable focus lens.