The M3 Pro - bringing NVIDIA style de-contenting to the Mac!

[bcortens]

Lets get some praise out of the way first, the M3 generation of chips is really fantastic from a performance per core and probably from a performance per watt perspective (we'll have to wait and see).

I am making this thread to coalesce all of the negative changes to the M3 series in one place.


However, there are a number of changes that really highlight the penny pinching and attempt to squeeze every last dollar out of people.

My issues from least to most important are:

1. M3 Pro GPU Core Count is Down
   1. The M3 Pro is still a decent almost 2x increase in GPU core count over M3 and while I would have liked to see this go up to about 22 cores for this generation this is the most minor of the disappointments
2. M3 GPU Core count is flat.
   1. The M3 really ought to have 12 GPU cores, I noticed on their graphs in the keynote that once again they are pushing total GPU power consumption upwards. This could have likely been mitigated somewhat by having more cores running at a lower clock (this is speculation but the lack of increase is a disappointment).
   2. Apple really should be pushing the floor higher still - constantly pushing the base GPU performance of the M series upwards helps make every single Mac a more attractive GPU platform
3. M3 Max Memory bandwidth is now segmented (down or flat)
   1. I had hoped they would increase rather than decrease memory bandwidth, I know the maxed out configuration can be had with the same bandwidth as last generation but if your workload was memory bandwidth dependent you now have to buy the absolute maximum configuration to get the most memory bandwidth.
   2. This brings memory bandwidth into the realm of something consumers need to care about and in order to make an informed choice they now need to look for tests of their software and determine how much a difference this will make. Perhaps most software won't notice at all but this is increased complexity so that Apple can save a few dollars or pennies on each chip is really disappointing.
4. M3 Pro CPU core count is down
   1. This a big one - while we have to wait for benchmarks I will not be surprised to see the M2 Pro outperform the M3 Pro on many multithreaded workloads.
   2. This places the M3 Pro closer to the M3 in CPU than it is to the M3 Max wish will make Apple's accountants happy but in terms of price it is now a $900 dollar upgrade to go from 6 to 10 high performance cores.
   3. I completely agree that the M3 Max should be different from the M3 Pro in terms of high performance core count but double is disappointing as the M3 Pro no longer exists as a mid-point between the standard M3 and the M3 Max
   4. This is likely to encourage people to spend that extra $900 dollars if they need CPU performance - a rather unethical change compared to the way they treated the M2 Pro vs the M2 Max.
5. 8 GB Base Memory and 256 GB Storage
   1. The iMac has had 8 GB of base storage since at least 2012 - that is as far back as I went looking - that is over a decade without an increase
   2. The base MacBook Pro has had 8 GB since 2013 an actual decade without an increase
   3. The Air at least was upped from 4 GB to 8 GB in 2017 so that is only 6 years and counting (but that is still quite sad)
   4. People can try and justify this with the fact that memory hasn't fallen that much in price but if you look at long term trends we should at least have 12 GB of memory as base and based on trends in flash storage prices we should be looking at at least 512 GB of storage. While there are a few limited use cases where 8GB-256GB can work they are limited and require added caveats like the notion that the Mac will never be sold to someone with more demanding needs.
   5. I don't care what the competitors are doing, Apple sells a premium product for premium prices and should offer a premium experience even to those who buy the base model.
   6. I personally never recommend anyone buy the base model and I think cynically that is Apple's intent.

Complaint list over....

Note: I may come back and edit this later.

 

[Silvestru Hosszu]

Hard to find any fault in this write-up.

 

[magbarn]

Lets get some praise out of the way first, the M3 generation of chips is really fantastic from a performance per core and probably from a performance per watt perspective (we'll have to wait and see).

I am making this thread to coalesce all of the negative changes to the M3 series in one place.


However, there are a number of changes that really highlight the penny pinching and attempt to squeeze every last dollar out of people.

My issues from least to most important are:

1. M3 Pro GPU Core Count is Down
   1. The M3 Pro is still a decent almost 2x increase in GPU core count over M3 and while I would have liked to see this go up to about 22 cores for this generation this is the most minor of the disappointments
2. M3 GPU Core count is flat.
   1. The M3 really ought to have 12 GPU cores, I noticed on their graphs in the keynote that once again they are pushing total GPU power consumption upwards. This could have likely been mitigated somewhat by having more cores running at a lower clock (this is speculation but the lack of increase is a disappointment).
   2. Apple really should be pushing the floor higher still - constantly pushing the base GPU performance of the M series upwards helps make every single Mac a more attractive GPU platform
3. M3 Max Memory bandwidth is now segmented (down or flat)
   1. I had hoped they would increase rather than decrease memory bandwidth, I know the maxed out configuration can be had with the same bandwidth as last generation but if your workload was memory bandwidth dependent you now have to buy the absolute maximum configuration to get the most memory bandwidth.
   2. This brings memory bandwidth into the realm of something consumers need to care about and in order to make an informed choice they now need to look for tests of their software and determine how much a difference this will make. Perhaps most software won't notice at all but this is increased complexity so that Apple can save a few dollars or pennies on each chip is really disappointing.
4. M3 Pro CPU core count is down
   1. This a big one - while we have to wait for benchmarks I will not be surprised to see the M2 Pro outperform the M3 Pro on many multithreaded workloads.
   2. This places the M3 Pro closer to the M3 in CPU than it is to the M3 Max wish will make Apple's accountants happy but in terms of price it is now a $900 dollar upgrade to go from 6 to 10 high performance cores.
   3. I completely agree that the M3 Max should be different from the M3 Pro in terms of high performance core count but double is disappointing as the M3 Pro no longer exists as a mid-point between the standard M3 and the M3 Max
   4. This is likely to encourage people to spend that extra $900 dollars if they need CPU performance - a rather unethical change compared to the way they treated the M2 Pro vs the M2 Max.
5. 8 GB Base Memory and 256 GB Storage
   1. The iMac has had 8 GB of base storage since at least 2012 - that is as far back as I went looking - that is over a decade without an increase
   2. The base MacBook Pro has had 8 GB since 2013 an actual decade without an increase
   3. The Air at least was upped from 4 GB to 8 GB in 2017 so that is only 6 years and counting (but that is still quite sad)
   4. People can try and justify this with the fact that memory hasn't fallen that much in price but if you look at long term trends we should at least have 12 GB of memory as base and based on trends in flash storage prices we should be looking at at least 512 GB of storage. While there are a few limited use cases where 8GB-256GB can work they are limited and require added caveats like the notion that the Mac will never be sold to someone with more demanding needs.
   5. I don't care what the competitors are doing, Apple sells a premium product for premium prices and should offer a premium experience even to those who buy the base model.
   6. I personally never recommend anyone buy the base model and I think cynically that is Apple's intent.

Complaint list over....

Note: I may come back and edit this later.

Definitely Nvidiaesque in execution in order to cover the huge costs of 3nm. Decontenting to shrink the die size with drop in memory bandwidth is what Nvidia did with Ada for the midrange parts. Tomshardware ran an article where analysts believe Apple blew $1 billion for the development and rights to TSMC's latest node. They said Apple is one of the few shot callers that can dump this much money into R&D.

 

[dmr727]

Huh. All this time and I've never heard the term 'decontenting' before.

 

[magbarn]

Huh. All this time and I've never heard the term 'decontenting' before.

Like the OP said, while Apple didn't severely nerf performance as likely the better GPU Caches in the M3Pro make up for the loss of bandwidth, the gulf between the Pro and Max series became bigger this gen. Nvidia did the same in which the 3080 was a good value as it came close to the 3090 as they were based on the same die, just cut down a bit. With Ada, they used a smaller chip with less memory bandwidth to make the 4080. But we'll see next week when the embargoes lift.

 

[dmr727]

Nvidia did the same in which the 3080 was a good value as it came close to the 3090 as they were based on the same die, just cut down a bit. With Ada, they used a smaller chip with less memory bandwidth to make the 4080. But we'll see next week when the embargoes lift.

I remember when Nvidia did that, just never realized there was a term for it. But yeah, next week will be really interesting, that's for sure!

 

[kiranmk2]

I was looking to write something ery similar to this but didn't know what forum to post it in so I'll add my thoughts here.

Apple is trying to pursue two goals as far as I can see.

1. Improve the headline performance of their products and stay ahead of the competition. To this end, I'm sure the full-fat M3 Max will give a large uplift in performance over the the M2 Max (which seems to be the case given the leaked benchmarks showing similar multicore scores to the M2 Ultra).
2. Combat the stagnation (or decreases) in unit sales across their device range by increasing the revenue per device. Around 2015 iPad and iPhone sales numbers stopped rising and Apple stopped reporting them, instead reporting revenue which kept rising.

As @bcortens has outlined above, Apple seems to have thrown everything at the M3 Max chip, but potentially kneecapped the M3 Pro (less CPU/GPU cores than M2 Pro) and stagnated the M3 devices (retained 8/256 configs). All an attempt to get people to either pay for an upgrade that they wouldn't have with M1/M2 gen devices or remove certain lower price tier devices (RIP 13" MBP). I think Apple is again going to send out the base level chips for review, so the reviews at launch will only be of the upgraded chips...

If you look across their other devices ranges you can see this in action:

• iPhones now very much held back next to the iPhone Pro. The iPhone Pro is held back next to the iPhone Max (telephoto lens). Even the iPhone Max now has a higher entry price (yes, you get the same price as the 256 GB model from last year, but if you didn't need 256 GB you have to pay more)
• iPad Air sitting in an awful spot where the storage is so low on the base, but if you update it you are just about into Pro money. Again, the 11" Pro is held back from the 12.9" iPad Pro (miniLED)
• Apple Watch has stagnated with Apple focussing on the Ultra model

tldr: Apple is going to focus on pushing people to buying the upper spec'd products by holding back the low and mid-range products

 

[Beau10]

given the leaked benchmarks showing similar multicore scores to the M2 Ultra

FWIW, there's a plethora of M3 Max GB6 benchmarks publicly available. Just the unbinned variant.