Qualcomm xisters, how do we cope?
https://www.tomshardware.com/pc-components/cpus/apple-debuts-m4-processor-in-new-ipad-pros-with-38-trillion-operations-per-second-on-neural-engine
 |
 Ape Out Shirt $21.68 |
|
Apple debuts M4 processor in new iPad Pros with 38 TOPS on neural engine
my ipad from 6 years ago works fine for watching porn and anime
Imagine running an LLM locally on your tablet.
It works shockingly well for stuff like SD (which isn't technically LLM but still).
I run llama3 8b locally on my m1 pro macbook and it's fricking great
what the frick are TOPS, how do I compare it to GPUs that use FLOPS as measurement?
>how do I compare it to GPUs that use FLOPS as measurement?
You don't.
you're not supposed to
>make up unit
>only attribute that unit to your own products
>claim old product = fewer units and new product = more units
>no unit means no competition
>applegays rave
https://www.qualcomm.com/news/onq/2024/04/a-guide-to-ai-tops-and-npu-performance-metrics#:~:text=Today%2C%20one%20of%20the%20leading,Neural%20Processing%20Unit%20(NPU).
moron.
>other company made up number, so it's not made up
moron.
POO
38 tops is quite good for a 15W chip. Strix Halo will hit 45 tops. Snapdragon X Elite hits 45 tops. They are the requirements for Windows 12/AI Ready PCs. I wonder how it will scale with the Pro/Max/Ultra chips but it is an excellent showing.
Is there software that actually takes advantage of AMD's and Snapdragon's AI cores?
It's been years and you still can't run Stable Diffusion on them, but you can on Intel, Nvidia and Apple Silicon.
I'm not sure, but it's a valid point. I run Nvidia and Apple Silicon only for my local AI usage. I am just pleased Apple isn't being caught off guard and is keeping in line with competition when many thought it wouldn't. All of these tops comparisons in the 30-50 range are funny in the context of my ~1320 tops 4090.
It's more of a 10W chip.
Where did you find the data? I'm curious how much my 4060 has.
Nviida's website has it listed in the specs section for each respective card family.
what page is this?
Sweet, thanks. Though guess it's still not enough to compare potential SD performance 1-1 given how Core ML is inferior.
yeah, and TOPs are one thing but a 4060ti will get shit on by a 3090ti all day thanks to 3x the vram and a lot more silicon grunt to throw at any compute need even with less TOPS.
Each M chip has used very slightly more energy than the last, so I was getting ahead of the efficiency/perf per watt pedants. But agreed, and really it is at 5-7W most of the time to hit that 10hr battery mark, I imagine.
https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/
https://www.notebookcheck.net/Apple-MacBook-Air-13-M3-review-A-lot-faster-and-with-Wi-Fi-6E.811129.0.html
M3 seems a bit more efficient actually, specially in idle. I'm extra curious about the 3 performance core variant of M4, could potential lead to some serious efficiency gains without costing too much performance in workloads that depend on GPU/Neural Engine.
>'m extra curious about the 3 performance
those are aborted dies anon, they will be less efficient because of that
I expect exactly the same thermal performance and battery as the 4P-core variant
Man, that'd be sad, but not too unlikely.
I got the 512GB 9 core model, I am interested to see if the battery is any better with one less p core. I couldn't stomach nearly $2K for 1TB and 16GB of RAM I will most likely never use.
Historically, the binned M-Chips get better battery life than the full M chips, at least in MacBooks. eg: The binned 8 Core M1 Pro, and 11 Core M3 Pro. They also had large cuts to the GPU so that may play a bigger role. Not sure about the binned M2 Pros, I hated that gen.
>the binned M-Chips get better battery life than the full M chips
how much better? source?
I have seen the advantage demonstrated across various reviews over the years, can't bear to find a source right now sadly but I will say, as someone who hated the concept of the binned chip being so prominent, I did take that small win when I found out about the slightly better battery. It was a genuine benchmark showing in the 20-40m range. Anecdotally, I notice slightly less heat on the binned models as well, which matters a lot to me.
They always do, 16GB RAM models also get around 3 hours less battery life than the 8GB models.
the models being discussed all had 16GB and 18GB of RAM. My 36GB M3 Pro 14" Macbook lasts days, what are you on about?
In full load? Kinda crazy that some RAM would cost so much battery life.
> nobody cares: the thread
wow. maybe the fourth time around apple will sell more machines and get them back market share they've continually lost since m1? lmao. nah. i'm kidding. nothing changes at all.
> version for air
do you homosexuals forget how these versions of chips are literally crippled in their air products?
> could
people are still far more interested in performance than battery power so it makes no difference. only the dumbest of dumb fricks will be buying such overpriced cancer.