[HN Gopher] Apple M5 could ditch unified memory architecture for...
       ___________________________________________________________________
        
       Apple M5 could ditch unified memory architecture for split CPU and
       GPU designs
        
       Author : akyuu
       Score  : 68 points
       Date   : 2024-12-30 19:18 UTC (3 hours ago)
        
 (HTM) web link (www.notebookcheck.net)
 (TXT) w3m dump (www.notebookcheck.net)
        
       | vimy wrote:
       | > That being said, such a split design may help with improved AI
       | inferencing.
       | 
       | Unified memory is the only reason Macs are so coveted right now
       | for local AI. A single 192 gb ram Mac costs less than the
       | equivalent in standalone GPUs.
        
         | Tostino wrote:
         | I'm sorry, I think you mispronounced market segmentation...
         | This is Apple we're talking about.
        
         | novok wrote:
         | The execution speed for LLM inference gets so slow once you
         | reach models that even fill out a 64GB machine. I was tempted,
         | too, but then realized it was unusable past 48GB-ish and
         | stacking used 3090s was the best price / perf / vram ratio.
         | 
         | What are the good use cases for very large memory amounts?
        
           | behnamoh wrote:
           | I ran all my research papers (about LLMs) on a Mac Studio.
        
           | reitzensteinm wrote:
           | Mixture of Expert models, where all parameters must be in
           | memory but only a subset are accessed per token, are a sweet
           | spot for Macs.
           | 
           | DeepSeek v3 for instance has 671B params, but should have the
           | memory bandwidth of a 37B dense model with a batch size of
           | one.
        
           | Terretta wrote:
           | Unusable in what way? Llama 3.3 70b q8 with 100k+ context
           | runs as well as GPT 3.5 did a couple years back, except local
           | on Mac and smarter.
        
             | lumost wrote:
             | Large context has different perf than large model. Op was
             | likely thinking of running 400b models and finding the
             | compute wasn't enough to make the memory useful.
        
           | shikon7 wrote:
           | It's slower than using just GPU RAM, but it's still faster
           | than using a regular PC that has a much more limited bandwith
           | between the main memory and the GPU. It's some sort of middle
           | ground on how fast and how cheap you can do inference with
           | LLMs that don't fit into a consumer GPU's RAM.
        
         | gazchop wrote:
         | I don't think the principal market actually cares about that.
        
         | mrtksn wrote:
         | When LLMs and diffusion models began spreading freely, it was
         | very funny to have support for some beefy high end GPUs and
         | MacBook Air M1.
         | 
         | I hope Apple sticks with the architecture. Even if its not very
         | practical, its great to have it as possible.
        
         | halJordan wrote:
         | This isn't splitting the unified memory, it's splitting the soc
         | into a sip. The ram is still on the same interposer/substrate.
         | The actual mm distance won't be regressing to say the distance
         | of a dimm socket to the cpu socket.
         | 
         | The software will still see a single memory pool
         | 
         | I know im disagreeing with the article
        
       | wtallis wrote:
       | > Another intriguing aspect is the separate CPU and GPU design.
       | If true, this essentially means M5 will not use a unified memory
       | architecture (UMA) shared between the CPU and the GPU.
       | 
       | This does not follow. Intel is shipping unified memory processors
       | with CPU cores and GPU cores on separate chiplets but still
       | sharing the same memory controller (on a third chiplet, for
       | Meteor Lake and Arrow Lake). AMD is about to launch Strix Halo, a
       | high-end mobile processor that is rumored to consist of one or
       | two CPU chiplets and an IO die with a big GPU and 256-bit memory
       | controller.
        
         | jsheard wrote:
         | Plus Apple themselves are already kind of doing it with the M
         | Ultras - those are two chiplets glued together, each of which
         | is only connected to half of the systems memory directly, but
         | it still behaves like unified memory even though half of the
         | memory traffic has to be routed via the other chiplet.
        
           | wtallis wrote:
           | And in the datacenter space, AMD has taken things even
           | further with the MI300A:
           | 
           | > Twenty-four x86-architecture 'Zen 4' cores in three
           | chiplets
           | 
           | > Six accelerated compute dies (XCDs) with 38 compute units
           | (CUs), each with 32 KB of L1 cache, 4 MB L2 cache shared
           | across CUs, and 256 MB AMD Infinity Cache(tm) shared between
           | XCDs and CPUs
           | 
           | > 128 GB of HBM3 memory shared coherently between CPUs and
           | GPUs with 5.3 TB/s on-package peak throughput
        
         | j45 wrote:
         | Maybe Apple has figured out something better than a unified
         | memory architecture.
         | 
         | It's hard to rule out their ability to create silicon that is a
         | step change.
        
           | wtallis wrote:
           | I wasn't really trying to comment on what Apple could or
           | could not pull off. Just pointing out that Notebookcheck
           | seems to be misunderstanding what they're reporting on from
           | Ming-chi Kuo, and the headline itself seems to be something
           | Notebookcheck made up rather than something from Kuo's
           | rumors. So this whole thread is even more baseless than it
           | appears at first glance, but it would still be interesting to
           | have an informed discussion about the pros and cons of
           | unified memory for consumer SoCs, and alternatives.
        
           | TiredOfLife wrote:
           | They didn't even figure out unified memory - even original
           | xbox (20+ years ago) had that.
        
             | Keyframe wrote:
             | And SGI O2 30 years ago.
        
               | j45 wrote:
               | Great reminder. I should have been clearer that I meant
               | in the current context the note was being made about
               | Intel and AMD working on things that apple's not.
        
             | cma wrote:
             | I think the Apple II frame buffer was in a unified system
             | memory.
        
               | DonHopkins wrote:
               | Ha ha, right! And with interlaced rows, 7 pixels per
               | byte, and a cockamamie color generation scheme, it had a
               | hardware "graphics decelerator".
        
             | o11c wrote:
             | UMA is trivial if you have so little RAM that bandwidth
             | doesn't matter.
             | 
             | The original XBox (2001) had 64MB. I think my PC from 1998
             | had that.
        
               | j45 wrote:
               | Good point - maybe this is a part of Apple's reasoning,
               | or there's something else architecturally.
        
             | j45 wrote:
             | For sure, they aren't always the first, but they do seem to
             | scale through some things in their own way. Got me my first
             | fastest cpu laptop with meaningful battery life.
        
         | ksec wrote:
         | Agree. Here is an article on SoIC from Anandtech [1].
         | 
         | Edit: [2] The tweet doesn't even mention about UMA. The
         | interpretation is entirely made up by Notebookcheck, I feel
         | like I am reading WCCFtech again making stuff up.
         | 
         | I am just thinking if this allow Apple to do something crazy
         | like 1024bit LPDDR5x or HBM3e memory solution.
         | 
         | [1] https://www.anandtech.com/show/21414/tsmcs-3d-stacked-
         | soic-p...
         | 
         | [2]
         | https://x.com/mingchikuo/status/1871185666362745227?ref_src=...
        
       | danjl wrote:
       | UMA hurts the GPU too much. Widely parallel processing wants to
       | access memory in bigger chunks than a CPU. If you try to mix
       | access and modification, you lose the benefit of widely parallel
       | processing. Other GPU designers have considered and eschewed
       | unified memory models, to the tune of hundreds of millions in
       | research dollars.
        
         | convolvatron wrote:
         | I agree that single cache-line fetches are pretty poor for
         | parallel vector units, but supporting the former in an
         | environment designed for the latter doesn't seem to off-putting
         | (the CM-5 did this).
        
           | DonHopkins wrote:
           | By the way: Does your user name convolvotron refer to the
           | hardware 3d audio audio processing system originally
           | developed at NASA Ames Research?
           | 
           | Such a cool name! And it says just what it does.
           | 
           | https://spinoff.nasa.gov/node/8965
           | 
           | https://spinoff.nasa.gov/sites/default/files/thumbnail0000_2.
           | ..
           | 
           | https://pubs.aip.org/asa/jasa/article/92/4_Supplement/2376/7.
           | ..
           | 
           | Body Electric supported the Convolvotron for visually
           | programming VR simulations with 3D sound:
           | 
           | https://news.ycombinator.com/item?id=24266722
           | 
           | Did you ever meet (or better yet get a tour of Ames from) the
           | late Ron Reisman, and see the virtual reality, flight
           | simulator, and air traffic control systems his research lab
           | developed?
           | 
           | Vertical Motion Simulator:
           | 
           | https://www.youtube.com/watch?v=5-lHcv_olkE
           | 
           | Marvin Minsky flies a simulator and wears VR goggles:
           | 
           | https://www.youtube.com/watch?v=mOKENF_-z8Y
        
             | convolvatron wrote:
             | no. but thank you so much for the references. that's
             | actually really great.
             | 
             | I needed a username in the early 90s, I had just finished a
             | paper where we microcoded a CM-2 to support high-throughput
             | convolutions with spatially varying kernels for Hubble
             | image correction (before the launched the eyeglasses
             | mission), and I decided I could be the hero or anti-hero of
             | convolution.
        
               | DonHopkins wrote:
               | Then you would probably appreciate one of the more
               | obscure and specialized Kai Power Tools for Photoshop:
               | KTP Convolver!
               | 
               | https://www.macintoshrepository.org/724-kpt-convolver-1-0
               | 
               | I can't find any demos of it on youtube, but it's the
               | kind of obscure retro thing that LGR loves to review.
               | He's really into the better known Kai Power Goo, which is
               | a bit more accessible to kids than KPT Convolver:
               | 
               | https://www.youtube.com/watch?v=xt06OSIQ0PE
        
             | jimmySixDOF wrote:
             | It would be very interesting to dust off some of those old
             | projects with modern affordances like not needing 1 SGI
             | machine per eye lol but to review the first principles
             | thinking in the problem spaces you guys looked at like air
             | traffic control which is still the same use case and XR
             | should have better chance today of making out of R&D labs
             | onto a shop floor. I have seen talks from Tom Furness about
             | the early applications being tested and seems like we are
             | just now getting to a place on the development curve where
             | some of them might just be practical. Thanks for all those
             | links they will keep me busy for a while !
        
           | danjl wrote:
           | I think all the mobile GPUs use UMA. I think the tradeoff
           | point is some complicated function of power envelopes and the
           | benefit of more, though slower, memory vs raw performance at
           | any power or $ cost. Though there are several dozen important
           | algorithms that run much better on GPUs, there are really
           | only two of them, 3D graphics and ML tensors, that have had a
           | big consumer and broad professional appeal.
        
         | canucker2016 wrote:
         | Could you give some concrete examples, including when
         | (approximate year/decade is ok) they were considering UMA for
         | CPU/GPU?
         | 
         | As a couple of others have mentioned,
         | smartphones/tablets/laptops seem to be the driving force in
         | UMA's spread.
        
       | jack-bodine wrote:
       | Not mentioned in the article, but another motivation behind this
       | could be that with split a CPU/GPU Apple could try to up-sale on
       | both when purchasing Macs.
       | 
       | The prices they charge just to go from 16GB to 32GB of RAM is
       | outrageous ($400 for Macbook pro).
        
         | berbec wrote:
         | I love the fact you can buy two 16GB/256GB Mac Minis and have
         | cash leftover compared to somone that bought a single
         | 32GB/512GB Mac Mini. Apple's upsells are insane
        
           | lotsofpulp wrote:
           | I'm seeing 16/256 for $600 and 32/512 for $1,200 on apple.com
        
             | andrewmunsell wrote:
             | On the Apple Edu store, it's $499 for the 16/256 and $1079
             | for the 32/512
        
               | oarsinsync wrote:
               | Hahaha wow I just checked Apple UK and the base
               | 16GB/256GB is PS600. 32GB upgrade is +PS400, 512GB
               | upgrade is +PS200.
               | 
               | It should not cost that much! 2x Mac mini M4 16GB/256GB
               | should not cost the same as 1x Mac mini M4 32GB/512GB!
               | 
               | Can someone help explain this in a way that isn't just
               | absolute price gauging of the higher end customer base?
               | Are the components genuinely that much more expensive?
        
               | cma wrote:
               | > Can someone help explain this in a way that isn't just
               | absolute price gauging of the higher end customer
               | 
               | No, it's the same reason Nvidia has a vastly higher
               | margin on datacenter cards:
               | 
               | https://en.wikipedia.org/wiki/Price_discrimination
        
               | least wrote:
               | The reason why the change to 16gb was such a big deal was
               | at least in part because it meant people didn't feel
               | forced into shelling out 200 dollars (or whatever it was)
               | for an extra 8gb of RAM.
               | 
               | It creates this weird dichotomy of having arguably the
               | best value computer on the market in the base mac mini
               | with 16gb of RAM and 256gb of storage and some of the
               | absolutely worse value upgrades (like spending $400 on
               | 16gb of RAM or $200 on 256gb of storage).
               | 
               | There's not much to explain here; they price gouge
               | upgrades because they can. People that want/need MacOS
               | for their work will pay for it, even if begrudgingly. I'm
               | not necessarily happy about paying that much for these
               | spec bumps but the benefits of using a Mac still outweigh
               | the cons for me.
        
               | kube-system wrote:
               | > Can someone help explain this in a way that isn't just
               | absolute price gauging of the higher end customer base?
               | 
               | It's a pretty normal pricing strategy. It's more common
               | than not. Most products or services you buy _anywhere_
               | will be sold at higher margins for more premium
               | offerings.
               | 
               | It might seem strange when compared to legacy PCs with
               | socketed components, but this isn't that, nor are most
               | products. Even among PCs this isn't strange anymore: go
               | take a look at MS's pricing on their first-party PCs.
               | 
               | Calling this "price gouging" is not really the right use
               | of the term -- usually it refers to price increases of
               | basic necessities in emergency situations.
        
               | goosedragons wrote:
               | Microsoft isn't a great example. They basically just crib
               | Apple's approach. And they do at least still have
               | socketed storage so that's very cheap to upgrade if you
               | do it that way.
        
               | kube-system wrote:
               | All of the big OEMs are soldering memory on at least some
               | (if not all) of their thin-and-lights, and I haven't seen
               | a single one priced at margins that weren't significantly
               | above the cost of materials.
               | 
               | Either way, my point is that flat margin pricing is
               | exceedingly rare. Everywhere from the grocery store to
               | the car dealer is charging higher margins on more premium
               | products.
               | 
               | Luxury cars have higher margins than economy cars.
               | Organic milk has higher profit margins than regular milk.
               | And Macs with 32GB of memory have higher profit margins
               | than Macs with 16GB of RAM. The fact that the desktops
               | PCs of our past priced RAM upgrades nearly at cost was an
               | outlier; a courtesy, not anything normal.
        
               | lotsofpulp wrote:
               | This is broadly called price discrimination.
               | 
               | https://en.wikipedia.org/wiki/Price_discrimination
               | 
               | It is basic microeconomics that a seller wants to be able
               | to get as high of a price as buyers are willing to pay,
               | but since different buyers have different abilities and
               | willingnesses to pay, a seller can maximize their revenue
               | by providing options at different price points.
               | 
               | Especially with societal wealth gaps, the people able and
               | willing to pay higher prices are going to be able to pay
               | higher price premiums, resulting in higher profit
               | margins.
        
               | kube-system wrote:
               | Right, and a sibling comment already pointed that out, I
               | just wanted to expand on the topic with examples.
        
               | lotsofpulp wrote:
               | >Can someone help explain this in a way that isn't just
               | absolute price gauging of the higher end customer base?
               | 
               | Price gouging, as a meaningful term, is restricted to:
               | 
               | https://en.wikipedia.org/wiki/Price_gouging
               | 
               | >Price gouging is a pejorative term used to refer to the
               | practice of increasing the prices of goods, services, or
               | commodities to a level much higher than is considered
               | reasonable or fair by some. This commonly applies to
               | price increases of basic necessities after natural
               | disasters. Usually, this event occurs after a demand or
               | supply shock.
               | 
               | Using the term "price gouging" anytime a potential buyer
               | thinks a seller is asking for too much money renders it
               | meaningless. I ask for as much money as the buyers for my
               | labor will pay, as I assume the people selling to me do
               | also.
               | 
               | It's just business, you try to earn as much as possible
               | (and that could involve not maximizing in a specific
               | transaction to incentivize repeat business in the
               | future). But in no way is anyone under any duress when
               | deciding to buy an Apple device, so if a buyer does not
               | feel like being price gouged, they should buy something
               | else.
        
       | grecy wrote:
       | I've always wondered if they'll do something for a true monster
       | Mac Pro.. 128 cores, gobs of memory, etc.
        
       | ngcc_hk wrote:
       | Given they control both hardware and software, can they have both
       | like efficiency core + power core, with some memory is unified
       | (up to 64 GB) and some dGPU has different memory which they sell
       | against NVIDIA ... Apple has to grow further is hard, but having
       | a T$ firm in front of you with a hardware piece you have is much
       | better than building an Apple Car ...
        
       ___________________________________________________________________
       (page generated 2024-12-30 23:01 UTC)