Discussion Apple Silicon SoC thread

It has begun.


Apple is coming for AMD and Nvidia.

FlameTail said:

It has begun.

Apple Reportedly Developing Its Own Custom Silicon for AI Servers

Apple is said to be developing its own AI server processor using TSMC's 3nm process, targeting mass production by the second half of 2025. ...

www.macrumors.com

Apple is coming for AMD and Nvidia.

Click to expand...

I am sure this are only for Apple servers. They probably don’t want to depend on Nvidia.

poke01 said:

I am sure this are only for Apple servers. They probably don’t want to depend on Nvidia.

Click to expand...

The AI custom silicon perhaps could also be used on the studio/pro in a multi-chiplet configuration. Depends if there is a unifying architecture behind this effort across the product stack.

FlameTail said:

It has begun.

Apple Reportedly Developing Its Own Custom Silicon for AI Servers

Apple is said to be developing its own AI server processor using TSMC's 3nm process, targeting mass production by the second half of 2025. ...

www.macrumors.com

Apple is coming for AMD and Nvidia.

Click to expand...

I'm sure it will be based on an upscaled version of their own iPhone NPU, right? Or rather, the NPU in the iPhone will be a scaled down version of the server chip.

Saylick said:

I'm sure it will be based on an upscaled version of their own iPhone NPU, right? Or rather, the NPU in the iPhone will be a scaled down version of the server chip.

Click to expand...

Yes, but I assume it won’t purely be a tensor processor. The NPU on the current A&M chips exist along with CPU and GPU. Some of that type of processing will be required as well. And I/O.

ashFTW said:

The AI custom silicon perhaps could also be used on the studio/pro in a multi-chiplet configuration. Depends if there is a unifying architecture behind this effort across the product stack.

Click to expand...

This is likely. Apple never designs for one config only but for it to scalable as possible.

poke01 said:

I am sure this are only for Apple servers. They probably don’t want to depend on Nvidia.

Click to expand...

I'll believe it when I see it. There are always rumors about Apple doing servers, it never happens.

What does Apple need with a huge AI cloud when their whole thing is on device AI? They certainly aren't going to be competing with Nvidia selling stuff into the open market.

This is just some random Weibo user who has one accurate call in his past (I don't count "iPhone 7 will be waterproof" everyone was expecting that, he might as well count predicting Apple adopting 5G to juice his total)

Doug S said:

What does Apple need with a huge AI cloud when their whole thing is on device AI? They certainly aren't going to be competing with Nvidia selling stuff into the open market.

Click to expand...

For training models, I guess?

SteinFG said:

For training models, I guess?

Click to expand...

I'm not sure it would be worth paying for the development and deployment when you can rent hyperscalers machines.

Could be nice as an inference server.

Testing / playing with models is really easy on apple silicon, but performance could be better.

Eug said:

Servers? What are you guys smoking?

Click to expand...

Hopium

Doesn't the M3 Max's CPU benefit from the larger SLC it has access to?

M3 Pro : 12 MB
M3 Max : 48 MB

Yet they both have effectively the same ST performance!

Cinebench 2024

Geekbench 6

This is true for previous M generations as well.

FlameTail said:

Doesn't the M3 Max's CPU benefit from the larger SLC it has access to?

M3 Pro : 12 MB
M3 Max : 48 MB

Yet they both have effectively the same ST performance!

Click to expand...

Why would you expect more SLC is going to lead to higher ST performance? Especially in something like Cinebench?

I'll bet the working set size of the executable (i.e. the inner loops where the work is done) is easy fits in 16 MB, in fact likely fits in L2i. The data it works on will be larger but it probably has an access pattern that's easy for prefetching. If so SLC size (and probably L2d size) is irrelevant. Performance would depend on the DRAM being able to stream faster than the CPU can crunch, which it obviously can with LPDDR5 - and clearly doesn't need that much as PCs aren't hindered by slower DDR5.

Qualcomm Snapgragon X Elite is coming for Apple M3 & M4 SoC performance!

Harry_Wild said:

Qualcomm Snapgragon X Elite is coming for Apple M3 & M4 SoC performance!

Click to expand...

More like coming for M1 & M2 performance from the numbers we're seeing.

I'm loving all this misinformation from media. Qualcomm has shills like no other.

Harry_Wild said:

Qualcomm Snapgragon X Elite is coming for Apple M3 & M4 SoC performance!

Click to expand...

That's great, but I don't think it matters, people aren't going to move to Windows just because it has equal performance.

Now, we'll see if mass consumers care about the AI, but that alone was a reason I personally began migrating away from Windows. Unfortunately now it seems Apple is jumping on that bandwagon (paired with Google to boot), but I at least have some belief they will at least not be forced to use it and that maybe Apple will still hold to privacy and let me opt out of letting them siphon my personal stuff to fuel it and constantly slap me in the face going "WHY AREN'T YOU USING OUR AI SEARCH BING AI BY BING MICROSOFT BING AI SEARCH?!?!? ARE YOU SURE YOU DON'T WANT TO MAKE BING AI SEARCH YOUR DEFAULT SEARCH AND BING YOUR DEFAULT BROWSER FOR BING AI SEARCH?" like Microsoft is heading towards.

FlameTail said:

Hopium

Click to expand...

Great you're gonna start hyping that stupid French car company too?

FlameTail said:

It has begun.

Apple Reportedly Developing Its Own Custom Silicon for AI Servers

Apple is said to be developing its own AI server processor using TSMC's 3nm process, targeting mass production by the second half of 2025. ...

www.macrumors.com

Apple is coming for AMD and Nvidia.

Click to expand...

Meh, despite their M cores being great, it hasn't led to MacOS suddenly having 50% marketshare. Simply put, there are other factors.

poke01 said:

I am sure this are only for Apple servers. They probably don’t want to depend on Nvidia.

Click to expand...

I do think not helping Nvidia is part of it (Apple seems to f'n hate Nvidia still), but I'd guess keeping proprietary to Apple systems is the bigger point. I think Apple recognized the risk of using all the content you could just access (whether you had license/right to access it be damned; not just for the issue of the potential legality/financial aspect but also because why would you want to build AI around internet that is like at best 40% trash trolling and is just going to poison your models with noise?), and but also knew the privacy laws and other (like, good luck dealing with China with AI models not censored by the Great Firewall), so you'll need to isolate the data sets its built on.

Which, Apple could do wonders if they worked deals to license content (be the first to do so - before they get too deep in models built on the content wherein the content companies can really hold them over a barrel), helping the lawsuits against all the current AI stuff that is based on ridiculous amount of stolen IP, which if successful would probably bankrupt all the current AI companies except for Google and Microsoft (and would probably hit them so hard that investors will demand they layoff basically everything that isn't AI). OpenAI would probably be bought by Microsoft in exchange for paying their licensing bills (or rather that would be the ultimate result, as OpenAI would probably go under and then Microsoft would just buy what made sense to buy of it if they don't simply just take it via hiring Altman and his sycohpants).

I am aware there are some licensing deals made already, but its a drop in the bucket compared to what these are built on, and even bigger is if the EU's privacy rules come into play, then these models are screwed. Heck in the US, current privacy laws regarding childrens' data alone could probably unravel these models.

Doug S said:

I'll believe it when I see it. There are always rumors about Apple doing servers, it never happens.

What does Apple need with a huge AI cloud when their whole thing is on device AI? They certainly aren't going to be competing with Nvidia selling stuff into the open market.

This is just some random Weibo user who has one accurate call in his past (I don't count "iPhone 7 will be waterproof" everyone was expecting that, he might as well count predicting Apple adopting 5G to juice his total)

Click to expand...

Yeah, I don't buy it either. But, they could sell access to it as a service business to business (which could be significant for App developers; possibly could be big in the creative markets as well if Apple shows willingness to recognize and pay for licensing rights). It could also be significant if rules about AI, especially privacy and licensing rights, come into affect. All the current AI models are built on basically everything they could access - rights be damned (they try to claim otherwise but they've admitted the reality - lots of them straight up say that having to adhere to legal licensing rights would kill their viability which tells you everything you need to know). If they have to try to remove stuff based on IP and privacy, they'll pretty much have to start from scratch. If Apple establishes licensing and other protections, they could have a huge leg up, just from stability over the turmoil that others will deal with.

Eug said:

Servers? What are you guys smoking?

Click to expand...

Whatever farts are left in the couch cushions at this point it seems. Sheesh this place has become a nightmare of constant "OMG!!!!!" constant silly over the top reactions to everything. There's one poster in particular that seems to be driving it (who is hyping like basically everything, but weirdly acts like every single one is going to destroy the entire rest of the industry, and just bounces between threads on each).

darkswordsman17 said:

There's one poster in particular that seems to be driving it (who is hyping like basically everything, but weirdly acts like every single one is going to destroy the entire rest of the industry, and just bounces between threads on each).

Click to expand...

But at least he choose an appropriate handle for that behavior

darkswordsman17 said:

There's one poster in particular that seems to be driving it (who is hyping like basically everything, but weirdly acts like every single one is going to destroy the entire rest of the industry, and just bounces between threads on each).

Click to expand...

Apparently they're done with tech for a while so you do have some reprieve.

Apple will likely never have 50% market share, but they don't care as long as they have the 10% of the market that makes them one of the largest companies in the world by revenue and the largest non-oil company by profit.

The only way Apple gets to a 50% share in computers is either they become a niche product category or the world as a whole becomes considerably more wealthy.

Or they sell $200-$500 dollar laptops. Which is very unlikely.

Mopetar said:

Apple will likely never have 50% market share, but they don't care as long as they have the 10% of the market that makes them one of the largest companies in the world by revenue and the largest non-oil company by profit.

The only way Apple gets to a 50% share in computers is either they become a niche product category or the world as a whole becomes considerably more wealthy.

Click to expand...

poke01 said:

Or they sell $200-$500 dollar laptops. Which is very unlikely.

Click to expand...

Apple selling $500 laptops would require some extreme deflation and I've never seen a government that didn't like to run the money press slow enough for the economy to outpace it.

Interesting if true: https://9to5mac.com/2024/04/28/gurman-new-ipad-pro-m4-chip/