Marvin

melgross said:

tht said:

melgross said:

Well, now it’s the 23rd to the 31st. Ok, I hope it’s better.

It'll come in no time. 

My main interest with the nanotexture is airplane usage. It's a crazy lighting environment which can make using a computer display frustrating. Since I'm not on airplanes too much these days, and it likely reduces contrast and sharpness, probably not worth it to me. Waiting on a live look at it myself though.

I’m 50/50 about having ordered the nano display. I’ll have to compare it to my wife’s, which will be here tomorrow. I wanted to try it. I like it in the Studio Display. This has to be different in some way or we wouldn’t get able to use the Pencil with it. These technologies get improved over time so this should be more rugged. As far as visual differences, I can see some in the reviews as vied in my M2 iPad Pro. The nano does have less black and likely slightly obscured shadow detail, but as some have mentioned, it’s mostly at somewhat oblique angles which isn’t a real problem.

I’m generalynot a fan of matte displays, because of these reasons and have said that here and elsewhere. In a perfect world we would have glossy displays used in dark rooms. That’s how we did our work back when, in my company. Hopefully, I won’t regret the choice.

It looks worse off-angle but not so bad straight on and normal lighting probably won't be as bright as the reviews:





The nano texture test is at 9:30 in the video. The glossy one looks best to get the deeper black levels.

AppleInsider said:

And, rumors developed a few days ago that it was coming to iPad in early 2025.

It would be nice if Mac apps would just run on iPadOS and didn't need modification. Games have their UI in-game so are different from other apps. The following video shows it is playable without modification when run remotely and it has controller support:



It should just need publishing to make it available. It can use 8GB of memory though so would probably crash on iPads with 8GB memory or less. It would only be usable on the 16GB models so not worthwhile publishing it.

tmay said:

blastdoor said:

radarthekat said:

It might not be crazy for Apple to buy an EV maker. 

Rivian has a focused product line participating in two very popular segments over the last few decades; trucks and SUVs. 

Apple could take on Rivian as a wholly-owned subsidiary, shielding it from Liability in any accident lawsuits that might come along.

A large company buying a small company brings the small company’s products to a very large customer base, while also providing funding to scale up manufacturing.  Apple’s Beats acquisition likely paid back Apple’s $3b investment in a few years just from selling Beats through Apple’s brick & mortar and online stores, as an example. 

Apple could contribute significant technology to Rivian along with tight integration to Apple’s ecosystem.

in any case this news suggests to us that Apple is not quite done thinking about a future of some sort in the EV/transportation market.  And that’s a good thing. 

I agree. The price is down a lot since the IPO, it isn't led by a narcissistic white supremacist drug addict, and the product is pretty good. 

If it made sense to buy Beats, I could see it making sense to buy Rivian. 

Market value of Rivian at this moment is a mere $10.2B, so a buyout would be feasible. At that, I would also buyout Canoo Trucks at the same time, which is valued at under $170 million, and would complement Rivian nicely.

Apple will never have a better deal on a functioning EV company, but that would require Apple fund Rivian for quite some time, and ultimately, it would still be a low margin Auto manufacturer, just as Tesla has become. But there is a large and steady market for RV's, lifestyle, and, utility vehicles, and autonomous taxis, so all would likely bolster Apple's vertical market appeal. Add in the existing R&D, and that "federation" looks pretty good.

Canoo;

https://www.youtube.com/watch?v=wAiJsB5CQUw

I say go for it.

They have huge yearly losses though so a bit more than $10b:

https://www.sec.gov/ixviewer/ix.html?doc=/Archives/edgar/data/1874178/000187417824000014/rivn-20231231.htm
https://www.macrotrends.net/stocks/charts/RIVN/rivian-automotive/net-income

$4-7b loss every year.

It's from Cost of Revenue so it's difficult for them to cut this loss, they lose money on every vehicle they make:

https://electrek.co/2023/10/03/rivian-ceo-very-clear-steps-profitability/

Average price per vehicle is $80k and they lose $32k on each. The loss is dropping, they say they can manage to get break-even by the end of this year but it's a massive risk and a money pit for investors until it can turn around.

The early days of Tesla were the same, they were near bankruptcy around 2018:

https://www.macrotrends.net/stocks/charts/TSLA/tesla/net-income

It took 2 years to turn profitable. If the same turnaround happened, it could be a worthwhile investment. I suspect Amazon won't sell though, they have plans to get 100,000 vehicles from them for deliveries.

nubus said:

discountopinion said:

Woah.. NPU in M4 seems to be 2x of TOPS than M3... Memory bandwidth is also up 20%. GenAI here we go.

Can't wait to see an Ultra or Extreme version of this baby.

M4 is delivering A17 Pro (2023) level NPU performance. Is it enough for 2024/2025 outside iPad Pro? Intel Lunar Lake is aiming for 45 TOPS. Will M4 Pro and Max deliver more?

The TOPs numbers are not quoted the same way by different manufacturers. If they use INT8, they can double the number over FP16, if they use INT4, they can quadruple the number.

Nvidia claims their 4090 is 1321 TOPs but in practise is only 2-3x faster than M3 Max on AI tasks.

Some manufacturers are counting the whole device CPU + GPU + NPU, others just the NPU.

Apple's Neural Engine page lists iPhone numbers as FP16:

https://machinelearning.apple.com/research/neural-engine-transformers

M3 Max has a 14TFLOPs FP32 GPU, which is theoretically 28TFLOPs FP16 and 56TOPs INT8.

If the M4 Neural Engine is 38TOPs FP16, then it's 76TOPs INT8 and an M4 Max chip would be 17TFLOPs FP32 = 68TOPs INT8 so M4 Max total = 144TOPs INT8.

Nvidia's claim suggests the 4090 is nearly 10x faster but clearly isn't and the following page shows numbers with INT4:

https://wccftech.com/roundup/nvidia-geforce-rtx-4090/

Further confusing the issue is there are numbers that include the use of sparsity, which is a technique for compressing matrices with unused values so they can fit more in a process at the same time and they double the numbers again:

https://blogs.nvidia.com/blog/sparsity-ai-inference/

For the Nvidia 3090, it's described as 35TFLOPs FP32, 71TFLOPs FP16, 284TOPs INT8, 568TOPs INT8 with sparsity, 568TOPs INT4, 1136TOPs INT4 with sparsity.

If Apple did the same, they'd go 144TOPs INT8 -> 288TOPs INT4 -> 576TOPs INT4 w/sparsity. They can put it on a slide saying 576 AI TOPs but it's not a meaningful number.

For marketing, companies like Nvidia promote the highest numbers because they are trying to convince big data center contractors to go with their product over AMD/Intel and the latter do the same.

Intel could easily be quoting INT4 with sparsity numbers, in which case divide them by 8 when comparing numbers.

It's not entirely wrong for them to use lower precision numbers because an AI model might allow using them but you can't compare different precision values to make assessments about new hardware, you have to compare the performance at the same precision.

blastdoor said:

narwhal said:

Qualcomm and Microsoft will debut Snapdragon X Elite Windows laptops in May that they claim match the specs of a base model M3. I suspect Apple wants to release a faster iPad before then to take a bit of wind from their sails.

We’ll see… the M1 Pro beats the m3 on multithreaded benchmarks. And if Apple had been willing to overclock a single core on the m1, it might have been able to match the m3 on bursty single core benchmarks too. But an M1 Pro uses a lot more power and requires more cooling. 

So if all they’ve done is create something equivalent to an M1 Pro, then they are still behind Apple (and really no better than Intel).

According to recent reports, they were lying:

https://www.semiaccurate.com/2024/04/24/qualcomm-is-cheating-on-their-snapdragon-x-elite-pro-benchmarks/
https://www.tomshardware.com/tech-industry/qualcomm-faces-benchmark-cheating-allegations-snapdragon-x-eliteplus-benchmarks-claimed-to-be-fraudulent

OEMs got less than half the performance they claimed. They are also being sued by ARM over their license.

jellybelly said:

nubus said:

It doesn't make sense to ship any AI product before WWDC. There is no new iPadOS/AI to make use of that new AI power.
Even worse for Air as it will be obsolete on arrival.

Launching M4 now will make MBA M3 "the last student computer without AI". It is, but Apple shouldn't put back-to-school sales at risk.
Same for iPhone. We're 4-5 months from next gen. No way that Apple will go full Osborne and kill all shipping products now without new products.

WWDC should be XCode AI + frameworks using very limited memory + the message that everything is possible with 18 TOPS and 6-8 GB as that is A16/M3.
A18/A18 Pro/M4 can deliver a new level of performance but iPhone has to be first.

Re WWDC and need for a new iOS with new AI— wouldn’t the new AI need new hardware to have that ML/AI capability?  And developers need that hardware in hand to try the new code at WWDC.  (I’m finding myself starting to see a logic in this rumor being true)

New AI software would work on existing hardware and OS, M3 + current OS runs everything just fine. It uses too much memory for mobile though, easily 10GB RAM for basic language models and 20GB+ for image generators and there's no swap on mobile. The higher iPad Pros go up to 16GB RAM but the 8GB models won't handle them without a new memory setup e.g running them from SSD.

The Neural Engine in M3 is fast enough for local AI and I don't see M4 being able to provide a significant boost for mobile hardware. In laptops, they can ship a 2x Neural Engine because it would only use 20W but there's no cooling in the mobile devices.

I expect the upcoming iPad Pro to use M3 and the focus being the Pencil. If Apple has AI announcements later, such as local AI model shipped with next iOS, the iPad Pro will run them, just as the existing M3 lineup will.

I'd expect a small hardware upgrade with M4 like go from 35TOPs to 45TOPs Neural Engine and GPUs get a small boost. The responses running local AI LLMs on M1 is instant. It slows down as the conversation goes on as it has to use the content of the whole conversation but they can just cap it or have some more efficient way to handle it.