Apple Reportedly Building M2 Ultra and M4-Powered AI Servers

[neilpmas]

I’ve always hoped that Mac would move to a cloud based system you log into. This feels like a much better way to have a Mac Studio in most geographic regions. Then you could have a very thin laptop with enormous computing power or build it into a studio display.

 

[Bonte]

Apple needs dedicated AI server chips, the M2 Ultra is way underpowered for that task and the desktop M4 is probably the same. They missed the boat but the M-chip architecture should allow for a rapid design to get these out by next year, but I'm no expert.

 

[bobjonesco]

Just waiting for Tim's keynote.

'I'm thrilled to announce our latest breakthrough product. While our competitors all built 'AI' things that are hard to use, we invented an even better Siri. We're calling it 'Apple iA' and we think you're gonna love it!

 

[ngextreme]

Is that joke alluding to the issue that Microsoft uses Linux on their servers?

Microsoft has been investing in Linux for 20 years. MS is a major funding source to the linux foundation. There is no issue there.

 

[NEPOBABY]

Microsoft has been investing in Linux for 20 years. MS is a major funding source to the linux foundation. There is no issue there.

Personally think MS will move Windows to Linux within a decade and have a compatibility layer for Win64 apps. I’m not saying that as one of those Linux fanboys. It’s just that Windows is a shambles under the UI. Even the UI is a shambles. They keep trying new lipstick on a pig.

macOS changes significantly under the surface every few years. That does break compatibility with older apps but it keeps the OS secure, fresh and reliable.

 

[Bonte]

I’ve always hoped that Mac would move to a cloud based system you log into. This feels like a much better way to have a Mac Studio in most geographic regions. Then you could have a very thin laptop with enormous computing power or build it into a studio display.

Yeah, that was the promise in the nineties. It works somewhat with games and render farms these days, but constantly uploading gigabytes of Photoshop files isn't that handy and lag is always a thing.

 

[jakey rolling]

I think you are forgetting about the Apple Vision Pro buddy!

Pretty much everyone has already forgotten about the Apple Vision Pro. Including most consumers.

 

[aid]

What's the point of AI servers if apple users have on device AI?

Apple still need to create their own LLM models in their data centres - it is only once these models are created (which requires a crazy level of compute power) that they'll be pushed to the devices for inference (asking questions).

 

[neilpmas]

Yeah, that was the promise in the nineties. It works somewhat with games and render farms these days, but constantly uploading gigabytes of Photoshop files isn't that handy and lag is always a thing.

I’ve used it lots at work where it works great. If you have to upload raw video I get it’s bad. Otherwise amazing

 

[gank41]

And yet, MS is way bigger than Apple. What's up with that? DO you not know that MS is developing IVAS based on HoloLens 2? Beside your article is a year old which is outdated lol.

Go to most any business in the world and they're running Windows. It's rare to see Macs in a work environment, although they do exist (I use two Macs for work). Go to some homes across the country and they might have a Mac, but most homes have at least an iPhone. If you removed the Business world from the whole "Windows is on most computers" scenario, I bet people would be surprised at how many Macs are in use.

 

[gank41]

Just had a thought- what if Apple wasn't going to somehow give new iPads the opportunity to run macOS natively, but what if Apple was gearing up some back end infrastructure to run macOS in some kind of "Citrix OneDesktop" type of "App"? Maybe instead of turning the iPad on and you get macOS (lite or whatever they'd call it), maybe "macOS" is just an app on your Home Screen that connects to a Cloud Based (running on these servers) Desktop with your iCloud files and stuff setup?

I run a Citrix/Windows OneDesktop session on my 2018 iPad Pro daily for work. Seems like a 1st party version of that would work better than a 3rd party version, no? Hopefully?

 

[Tito360]

Makes sense for them to leverage their own hardware, especially if their AI implementation is designed around their neural engine and GPU designs as opposed to "commodity / off-the-shelf" hardware.

And I do not expect Apple to offer these servers at retail, so this is not the "return of the Xserve".

IBM 2.0

 

[asiga]

I don’t know what you are talking about. The LLMs are not big data unless you call lots of parameters data..

From the wikipedia page on big data: "the best interpretation is that it is a large body of information that cannot be comprehended when used in small amounts only". Now, if you consider that the datasets used by what is called "AI" nowadays can be very well over 40GB trained by searching for statistical patterns in lots of data.... if that's not big data, the iPads announced today will run Windows.

If I ask AI to remove the background from an image, it’s not big data but a smart algorithm.

Nope. Current "AI" (I use double quotes because I don't consider it true AI) doesn't use graphics algorithms. Everything is neural networks from statistical analysis of big data. "AI"-based background removal has nothing to do with traditional graphics algorithms. "AI" doesn't follow a logic. Graphics algorithms do (BTW, true AI does follow logic, but the current "AI" doesn't).

But using an LLM to jumble together a smart sounding answer without “understanding” what it’s saying.

Using an over 40GB file constructed by looking for statistical patterns in a huge number of texts sounds a lot like big data. Specially when you even admit it doesn't understand what it's saying.

But it will not download and run an entire ChatGTP model on device.

Yes. It won't be on-device. Just a wish for marketing purposes.

 

[JitteryJimmy]

A lot of SoCs are made for Apple, including the A and M series. Hundreds of millions are delivered a year.

They probably produced another 100 million SoCs that are generally good but which aren't suitable for a consumer device: not enough functional RAM, not enough functional cores, or that can only run reliably at 98% of the specified clock speed.

Intel would sell these as low-end components for a $150 each, but right now Apple does nothing with them if they can't meet the grade for their low end products.

Apple could dump these into a landfill... or they could take the best of the rest and install them in a high density carrier, and with a custom OS they could use them as a part of very high density multi-cpu super computer.

 

[wbeasley]

As the EU push to open Apple up, this would be the perfect fight back.
Apple dev chips that leverage software on Apple devices to give them a competitive edge which fall outside EU regulation of customer code.

They can’t/wont offer the same to Android devices so a better service at OS level could still attract and keep Apple customers happy. Regardless of interfering government regulations 😉

 

[PBG4 Dude]

IBM 2.0

Funny thing; decrement each letter of IBM by 1, and you get HAL.

 

[DanVM]

One word! The iPhone! Both Meta and Microsoft stupendously failed at a smart phone. Apple breaks thought failed markets and does what no other company can do! Don't hate the player! Hate the game!

Microsoft layoffs effectively kill HoloLens & mixed reality projects | AppleInsider

Microsoft's layoff of 10,000 employees may pause the company's HoloLens and headset-based projects, with the entire teams working on mixed reality and virtual reality products decimated by the HR axe.

appleinsider.com

The iPhone and iPad were big hits for Apple. However, they also have products that were not as successful, like HomePod and Apple TV. We'll see what happens in the next few years to see in which group AVP will be.

 

[sunny5]

Apple never breaks out device sales. Your argument is logically flawed.

Also, Meta and Google are only good at one thing: convincing advertisers that we actually give a crap about the ads they serve based on vacuuming our personal data.

Go play with your Google and Meta spyware. Have fun. While you are at it, enjoy Microsoft’s dumpster fire of an OS

Windows 11 Update Causing Serious Problems For Millions Of PC Users

Hot on the heels of another chaos-causing update, there's another one that's seeing dozens report problems ranging from crashes, blue screens, stuttering and performance

www.forbes.com

There is no such thing as NEVER. Never heard of Big Sur gate?

 

[ChrisA]

This is very incorrect. An LLM is not a “snapshot” of the internet. It’s not a database you query to get “the truth”. You’re confusing LLM’s and AI powered Search.

The above is correct. No argument.

But there is a way we can THINK of an LLM as a "snapshot" of the training material. It is in effect is VERY compressed "lossy" copy. Lossy in the way a very low bit-rate MP# file is a snapshot of the music, it throughs a lot of information away

In fact great amounts of computational resource are used to find the best way to compess many some petabytes of material into a roughly 100 gigabyte space. They call the search method "gradient decent".

It is a fact that compression removes redundancy and very good compression preserves much information while removing it. After removing 99% of the training material, what you have left over is just the rules and patterns and connections between the data points. The AI people would say that "'regularization" enables the model to generalize. I would say that compression removes redundancy and leaves only the rules and patterns.

In any case, it is not a lossless snapshot. What would be the point to that when Google and others already have it?

 

[huge_apple_fangirl]

I’ve always hoped that Mac would move to a cloud based system you log into. This feels like a much better way to have a Mac Studio in most geographic regions. Then you could have a very thin laptop with enormous computing power or build it into a studio display.

Apple will never do this because they would lose out on hardware sales, which is the bulk of their income. There’s a reason they’ve been fighting tooth and nail against web apps on iOS, they certainly aren’t going to willingly turn macOS into an “app” that you can launch from your ten year old MBA or gasp… PC.

 

[neilpmas]

Apple will never do this because they would lose out on hardware sales, which is the bulk of their income. There’s a reason they’ve been fighting tooth and nail against web apps on iOS, they certainly aren’t going to willingly turn macOS into an “app” that you can launch from your ten year old MBA or gasp… PC.

Yeah, but everyone says this until the competition do it. macOS is only so much of a sales pitch.

 

[MacBH928]

Would be amazing if Apple started offering a newer version of their Xserve lineup again.

whats amazing about Xserve? you can buy any computer and make it your server.

Training models like LLMs requires huge resources. On-device there will only be the finished model.

so they train the software on big servers and the end result gets downloaded to your device? How many GBs is that?

 

[neilpmas]

so they train the software on big servers and the end result gets downloaded to your device? How many GBs is that?

Ollama is between 5 and 40GB depending on how many parameters you want. That’s just a guide for you it could be very different.

 

[flybass]

From the wikipedia page on big data: "the best interpretation is that it is a large body of information that cannot be comprehended when used in small amounts only". Now, if you consider that the datasets used by what is called "AI" nowadays can be very well over 40GB trained by searching for statistical patterns in lots of data.... if that's not big data, the iPads announced today will run Windows.


Nope. Current "AI" (I use double quotes because I don't consider it true AI) doesn't use graphics algorithms. Everything is neural networks from statistical analysis of big data. "AI"-based background removal has nothing to do with traditional graphics algorithms. "AI" doesn't follow a logic. Graphics algorithms do (BTW, true AI does follow logic, but the current "AI" doesn't).


Using an over 40GB file constructed by looking for statistical patterns in a huge number of texts sounds a lot like big data. Specially when you even admit it doesn't understand what it's saying.


Yes. It won't be on-device. Just a wish for marketing purposes.

Points other than the meaning of “big data” are valid.

 

[jakey rolling]

Would be amazing if Apple started offering a newer version of their Xserve lineup again.

Xserve was a wonderful machine to people who had never seen a server before. It was an awful machine to people who had.