M5 vs. M6 Chip Buyer's Guide: How Much Better Really Is M6?

[Abydos]

Probably just companies that are going to flop over due to all the accumulated tech debt.
Or companies that somehow managed to survive despite still using archaic solutions, and are a living museum section of IT industry.

I would guess companies for which excel meets their needs, think of how many items in inventory and concurrent contracts they have - their needs may not exceed a mid-level system with 'stock' software not every firm needs state of the art hardware/software

 

[Cervisia]

The M6 halved the number of "normal" cpu cores, but added a bunch of "cinebench cores" to help with benchmarks. Given how much more the M6 Mini costs compared to the M4, it's absolutely atrocious.

 

[The Mercurian]

Per my experience, a LOT of industries still do that... Most people don't want the rigidity of "data science" tools, they want to maker a quick pivot table, a chart that they can copy to Powerpoint, etc.

Yeah ok I can understand your point here. But I would say once you learn some of the data science tools its way quicker to make a quick table and produce a plot than pointing and clicking in Excel. Might take a bit of effort initially but its well worth learning how.

Probably just companies that are going to flop over due to all the accumulated tech debt.
Or companies that somehow managed to survive despite still using archaic solutions, and are a living museum section of IT industry.

I mean yes this is what I suspect. But my worry for the folks doing serious work semi-manually in Excel is that this is the first work the AI / vibe coding will replace. Kind of by the nature of having some point and click steps in the workflow its vulnerable to being replaced by AI.

I would guess companies for which excel meets their needs, think of how many items in inventory and concurrent contracts they have - their needs may not exceed a mid-level system with 'stock' software not every firm needs state of the art hardware/software

Again.... this sounds like stuff ripe for replacement by AI. All you need is a middle manager who is an ai booster and hey presto some of the juniors not wanted anymore.


I think it you are excited by improving Excel benchmarks it might be time to take an intro to python /R/whatever course.

 

[Cervisia]

From my side where my work is mostly in computational biology rather than AI (hence more CPU than GPU/NPU based), I'm more interested in the decision between M5 Pro and M6. Here its less clear, and I think the slightly more powerful M5 Pro option for the mini probably wins out over the M6.

The M5 Pro has 5 or 6 super cores, the M6 has 2 (down from 3-4 of the previous gen...), and they don't seem to be boasting about single-thread gains, so expect a marginal one. Given this and the memory bandwidth (most scientific computation benefits hugely from that), the M5 Pro will leave the M6 is the dust.

Benchmarks are less and less relevant, because the 'cinebench cores' are there to boost the scores, but don't have the same benefits with every practical use case.

 

[Cervisia]

While the Mac Studio is supposed to be Apple’s TOL computer, it is far from their top seller. Therein lies the problem.

And why that might be? Maybe because most people won't pay that much money for an "obsolete" hardware?

 

[Biro]

And why that might be? Maybe because most people won't pay that much money for an "obsolete" hardware?

I think most people want laptops.

 

[meboy]

"10% higher unified memory bandwidth"

Not even close to good enough.
They need to get the base M chips up to around 400gb/s as fast as possible.
Pro 600+, Max 800+
This 200 to 240 per generation sort of improvement is way too little.
They're not even up to the level of a 6 year old RTX entry model GPU.

Hoping for a really big jump with M7 and a redesigned Air in 2027/2028.

 

[Siliconguy]

"10% higher unified memory bandwidth"

Not even close to good enough.
They need to get the base M chips up to around 400gb/s as fast as possible.
Pro 600+, Max 800+
This 200 to 240 per generation sort of improvement is way too little.
They're not even up to the level of a 6 year old RTX entry model GPU.

Hoping for a really big jump with M7 and a redesigned Air in 2027/2028.

How much power does your 6 year old RTX entry model GPU use?

Power consumption scales with the square of clock speed. In a Studio I grant it doesn't matter. On a laptop it does.

The base chips like the M6 should prioritize low power consumption. The Max and Ultra, sure, go all out. Liquid cooled Rx 4090 and 5090s are all the rage after all, it should be more successful than the water-cooled G5 was.

 

[Siliconguy]

I really don't understand Apple's M processor roadmap. The Mac Studio is supposed to be their top of the line computer yet it's always out of date.

Wait for the benchmarks to see. I suspect the M5 Max will stomp the M6 into the dirt, not the mention what the Ultra would do. The cost for that power is heat. Everything is a tradeoff.

 

[conmee]

Per my experience, a LOT of industries still do that... Most people don't want the rigidity of "data science" tools, they want to maker a quick pivot table, a chart that they can copy to Powerpoint, etc.

Most of my career has been in and around automotive. Many of the OEMs and Tier 1 suppliers I’ve worked with still utilize massive Excel sheets. Especially at the departmental level. Despite all the better data science tools, Excel still somehow lives on. General Motors, Ford, Hyundai, Kia, Bosch, Honeywell, JCI, Reynolds, Valeo to name a few. Also, Williams F1 Team Principal James Vowles lamented for the first few years of his tenure that Williams was still using Excel instead of an ERP system and was a major reason for their lack of competitiveness in F1. Supposedly they finished migrating all their Excel sheets just last year.

 

[The Mercurian]

Most of my career has been in and around automotive. Many of the OEMs and Tier 1 suppliers I’ve worked with still utilize massive Excel sheets. Especially at the departmental level. Despite all the better data science tools, Excel still somehow lives on. General Motors, Ford, Hyundai, Kia, Bosch, Honeywell, JCI, Reynolds, Valeo to name a few. Also, Williams F1 Team Principal James Vowles lamented for the first few years of his tenure that Williams was still using Excel instead of an ERP system and was a major reason for their lack of competitiveness in F1. Supposedly they finished migrating all their Excel sheets just last year.

What kind of calculations are they doing with these massive Excel sheets if you can say?

 

[mistymac]

Most of my career has been in and around automotive. Many of the OEMs and Tier 1 suppliers I’ve worked with still utilize massive Excel sheets. Especially at the departmental level. Despite all the better data science tools, Excel still somehow lives on. General Motors, Ford, Hyundai, Kia, Bosch, Honeywell, JCI, Reynolds, Valeo to name a few. Also, Williams F1 Team Principal James Vowles lamented for the first few years of his tenure that Williams was still using Excel instead of an ERP system and was a major reason for their lack of competitiveness in F1. Supposedly they finished migrating all their Excel sheets just last year.

Its because of older generations being comfortable with using Excel IMO. Plus like another user mentioned there is less systematic setup involved when doing data analysis quick & dirty with excel only. Why do all that extra work when I can just slap in a pivot table etc.? Well, there are good reasons not to do that but try reasoning with those who say otherwise lol. Drives me crazy where I work to the point that I've siloed my own version of data formatted correctly in a semantic model via PowerBI. Still even after showing off all the rad dashboards and reports my boss is like totally uninterested in using it himself.. I'm like okay dude suit yourself and the rest of the owners if you don't want to understand your company data more clearly, easily, and efficiently. 🤷🏻

 

[Nickerbocker]

It really comes down to the type of work you do as to whether these gains are meaningful. A lot of people commenting that their M1 is still fine likely don't do any heavy workloads like processing video and probably don't care about AI/LLMs. Which I think is perfectly fine - any modern computer should feel snappy when performing an ordinary computer task. I wish the PC/Windows market would get that memo.

 

[name99]

What kind of calculations are they doing with these massive Excel sheets if you can say?

People (usually) aren't using Excel for "massive calculations".
They are using Excel as a DATABASE that
- has a UI they understand
- is easily browsable or modifiable
- can easily have "minor" calculations attached to it

As a simple example of this kind of thing, imagine the BOM/budget for building a new house...

 

[conmee]

Its because of older generations being comfortable with using Excel IMO. Plus like another user mentioned there is less systematic setup involved when doing data analysis quick & dirty with excel only. Why do all that extra work when I can just slap in a pivot table etc.? Well, there are good reasons not to do that but try reasoning with those who say otherwise lol. Drives me crazy where I work to the point that I've siloed my own version of data formatted correctly in a semantic model via PowerBI. Still even after showing off all the rad dashboards and reports my boss is like totally uninterested in using it himself.. I'm like okay dude suit yourself and the rest of the owners if you don't want to understand your company data more clearly, easily, and efficiently. 🤷🏻

That’s part of it. But I’ve seen a lot of it due to middle and upper managers building little kingdoms for themselves and the last thing they want to do is be more efficient resulting in headcount reductions. Fewer heads smaller kingdom. And if AI (or PowerBI) can do all the underlings work or make them super efficient, for example, then why have a manager. A director level can just use advanced tools to eliminate (hypothetically) large swathes of headcount and reduce span and control. Same with using any better tech, it’s not just down to older people not embracing new tech. They will if there is a direct incentive and job security to go with it. If they do the mental calculus and determine AI or PowerBI or whatever is going to reduce the workload they start to worry. There’s also the issue of gatekeeping institutional knowledge. By having data and info in less accessible disparate locations, managers often think they make themselves irreplaceable if they are the only ones that know how to get to the data or extract content (no one is irreplaceable no matter how difficult some people make it - hiding/changing passwords, deleting files, at best it’s a nuisance or challenge to replace someone). So there is a lot of inertia to adopting new tech due to FUD. But also don’t put it past a company to find a more efficient way of doing something and then bank that savings with some layoffs.

GM has a huge spare parts and warranty division that was called Service & Parts Operations (SPO) but they renamed it Customer Care and Aftersales (CCA). One manager was simply tracking every component vendor and that component vendor’s suppliers. Hundreds of thousands of individual skus. I recall years ago when I was with IBM, Buick asked for help because they were seeing an uptick in O2 sensor failures but they appeared completely random. And even some of the replacement parts from the same vendor would fail. Buick couldn’t figure it out. So we took a few of their immense spreadsheets and built a database with DB2 (IBM’s database software) specifically so our data scientists could more easily slice/dice info. Turns out, that one supplier (due to pricing pressures from GM) sourced some of their copper from a cheaper Indonesian supplier instead of Australian supplier halfway through a model year. It was the O2 sensors using the Indonesian copper that were more liable to fail. But the vendor still had some of the good O2 sensors in stock and would sometimes send a good sensor. So GM went back to the O2 sensor vendor and negotiated a slightly higher purchase price to guarantee they got the reliable O2 sensors every time. But this kind of upstream quality control miss happens all the time in automotive because the OEMs twist the arms of their vendors without mercy and the vendors do everything they can to cut costs to meet OEM pricing demands. I have little love for the OEMs.

One other example from which is more generally about how we on the outside aren’t always aware of the motivations of managers and executives. It isn’t always readily apparent why they make some of the seemingly dumb decisions they make until we have a better idea of what incentives are at play. I was part of a consulting team at Chrysler many years ago and they engaged us to help them forecast vehicle demand. At the time, they would simply reach out to each of their plant directors and ask what volume they could produce and they would tally these numbers from the plant directors and they would make that many vehicles. They said they wanted to better utilize company resources and not rely on stuffing the dealer channel and driving up days on lot if they could better determine how many vehicles to build. We teamed with a company called i2 that made this proprietary demand forecasting software in the early 2000s (eventually bought by Oracle if I recall after some lawsuit). After about 3-6 months of work and developing a proof of concept system that allowed execs and managers to input variables, the system would suggest the optimal amount of vehicles to manufacture based on all the input from Chrysler plus all the third party analysis and market info. The execs were giddy! As if an oracle told them exactly which models and trim levels and how many of each to make. A few weeks later they called us in to cancel the project. We were stunned at the turnaround in mood. It was explained to us that they couldn’t use the system to determine production. They said all the plant directors pushed back vehemently against the project. Turns out the plant directors were highly compensated to keep their plants at 95% or higher capacity year round. If they dipped lower then bonuses and other benefits for management and plant workers would suffer. We asked if compensation plans could be reworked. Yes, for the plant directors, but no for all the union employees. And if a plant was not at maximum capacity they couldn’t just lay off people to fluctuate with demand because of the union. If a plant dipped to 80% for example, any unneeded workers would simply get their full pay just showing up and sitting in the break room all day. This isn’t meant to be an anti-union example, but rather to show there are a lot of potentially competing interests in any corporate initiative or tech adoption and so sometimes what seems a no-brainer to us on the outside is a Gordian Knot of near impossible challenges internally. That’s one reason new CEOs like to bring in “their” people when a change in leadership occurs. Sometimes that’s what it takes, a wholesale housecleaning, to make significant changes or adopt cutting edge tech. But for every Steve Jobs, there are thousands of Mary Barras (GM) or Jim Farleys (Ford) just trying to keep the ship from sinking.

 

[ensjesse]

Well for now everything runs fine. That is the power of ARM and the current ecosystem. I was so pissed when I bought my 68K Performa right before Power PC was the next wave. I have always learned to try and stay close to the latest as possible. When the M1’s
came out I bought everything that had them. Air,
Mini,
Ipad Pro with the M2
and they are still going so much stronger than any of the computers I have owned of years past but I will pick up the M6 mini just to be “safe” lol