Mac Hardware Trends (1984-2022)

[ahurst]

Mind sharing your data tables? I’d like to play with it in a visualizer!

PM sent! Anyone else who wants the data just ask and I'll send it your way, I'd share it publicly but don't want to run afoul of any EULA issues with source I got the data.

Note that the actual database is just a big JSON file with info for each Mac, so I've written a Python script to extract/parse values of interest and spit out a CSV. If you want to try your hand yourself, I recommend you have some basic Python experience.

 

[ahurst]

As for desktops, the observed data plateau from around the same time — late 2012 to 2017 — may be attributed to Apple adding education-only and heavily downclocked budget variants of the iMac configured with reduced processors (some examples: [1] [2] [3] [4]) along with one Mac mini outlier otherwise not found with Apple’s other desktop products — leaving consumer and prosumer models to pick up that dragging element.

Ding, ding, ding! Colour-coding the dots by Mac model, those low multi-core scores are indeed largely from the weak base-model 21.5" iMacs meant for education (note: below plot is for iMac and Mac mini models only):

That big jump from 2017 to 2019 is due to the jump from 4-core chips at the high end to 6 and 8-core Coffee Lake CPUs. Looking at single-core scores, though, excluding those low-end Macs makes the slowing of year-over-year gains even more prominent, with the rate of improvement from 2013 to 2019 being notably shallower than from 2009 to 2013:

 

[mectojic]

Steve Jobs always said Apple should not ship last year’s processor in their products- even for the cheapest consumer and education products.

That’s why the original iBook was the second fastest laptop in the world when released, only slower than the Apple PowerBook. That’s why the Apple eMac (education Mac) shipped with a powerful new G4 processor, instead of the G3.

There is something really disappointing about seeing those underspecced iMacs from recent years. Same with the new budget 2021 iPad, which uses A12 instead of A15.

 

[ahurst]

Okay, this one took a while but is definitely interesting: here's the inflation-adjusted prices of different Mac SKUs over time, split by product category:

I used the 'priceR' package in R to automatically convert the original USD prices for each model to adjusted 2020 USD (the package doesn't support 2021 or 2022 yet). Differentiating between "desktops" and "workstations" in the pre-iMac, pre-Jobs era was tricky, but for 68k workstations I settled on "any Mac that's in the II family (including the SE/30) or has a full 68040", and for old-world PowerPC workstations I settled on "is an 8100, 9500, 9600, G3 tower, or Workgroup Server" (improvements welcome).

Some quick observations:

• There was a big drop in Mac consumer desktop prices pre- and post-iMac, which has remained pretty steady ever since.
• Mac laptop prices declined steadily until around 2010, where they've remained pretty stable ever since (though newer high-end models have raised top end prices)
• The trend line for workstations is pretty wild, with prices steadily on the rise since the start of the Intel era. I guess a lot of that has to do with Xeon chip prices and the cost needed to differentiate top-tier hardware from increasingly capable chips in consumer iMacs and MacBook Pros.
• lol at the 20th Anniversary Mac costing the 2020 equivalent of ~12000$ USD

Steve Jobs always said Apple should not ship last year’s processor in their products- even for the cheapest consumer and education products.

That’s why the original iBook was the second fastest laptop in the world when released, only slower than the Apple PowerBook. That’s why the Apple eMac (education Mac) shipped with a powerful new G4 processor, instead of the G3.

There is something really disappointing about seeing those underspecced iMacs from recent years. Same with the new budget 2021 iPad, which uses A12 instead of A15.

For sure. That's something I'm hoping will change in the Apple Silicon era: with the non-Pro M1 powering a wide array of Macs previously covered by a range of weak-to-midrange Intel CPUs, we should hopefully be seeing performance consistency across models closer to the G3/G4 PowerPC era.

 

[jackoverfull]

Fascinating.

 

[quarx]

Thank you very much, ahurst, for these gripping visualizations!

As Apple itself is advocating the use of mobile devices also for computationally demanding tasks, what do you think about a similar visual analysis of iOS device specs?

 

[PsykX]

I'm a big viz fan here. Great work !

Those CPU graphs are a great explanation as to why Apple ditched Intel.

Those storage graphs explain the transition to SSDs and why Fusion drives ended up being a bad idea, even though I still think it was the best idea in a long time. Apple played cheap and put small HDDs and small SSDs. Just imagine a Fusion drive with a 256 GB SSD and a 6 TB HDD. This is what should have happened and would have kept the trend linear.

Those base RAM graphs explain that Apple has been cheap about giving their users fair amounts of memory, and maybe also that Apple Silicon doesn't need as much.

 

[ahurst]

Thank you very much, ahurst, for these gripping visualizations!

As Apple itself is advocating the use of mobile devices also for computationally demanding tasks, what do you think about a similar visual analysis of iOS device specs?

I've got that data too so I can certainly try! The main hurdle for performance data is trying to unify Geekbench scores from early 32-bit devices and later 64-bit devices: I'd need to calculate the offset and scaling factor for GeekBench 2 iOS to GeekBench 5 based on the devices that have overlap (64-bit devices released before iOS 11) to get both into a single metric. Would be very cool to look at though, especially on the same plot as the Mac performance data for the same timeframe!

 

[Analog Kid]

Steve Jobs always said Apple should not ship last year’s processor in their products- even for the cheapest consumer and education products.

That’s why the original iBook was the second fastest laptop in the world when released, only slower than the Apple PowerBook. That’s why the Apple eMac (education Mac) shipped with a powerful new G4 processor, instead of the G3.

There is something really disappointing about seeing those underspecced iMacs from recent years. Same with the new budget 2021 iPad, which uses A12 instead of A15.

Steve Jobs often said things and then later did different things. The 2-core i3 iMacs were introduced in August 2011, before Jobs retired.

 

[B S Magnet]

Steve Jobs often said things and then later did different things. The 2-core i3 iMacs were introduced in August 2011, before Jobs retired.

The dual-core i3s weren’t deliberately underclocked, but they were sold as entry-level CPUs, as their placement within the Core iX line would indicate.

The underclocking by Apple wouldn’t really get underway in earnest until the 2014-era Core i5s in the iMacs (and that one Mac mini), all of which came equipped with dual-core i5s but at profoundly underclocked base speeds and non-upgradeable internals.

 

[Analog Kid]

The dual-core i3s weren’t deliberately underclocked, but they were sold as entry-level CPUs, as their placement within the Core iX line would indicate.

The underclocking by Apple wouldn’t really get underway in earnest until the 2014-era Core i5s in the iMacs (and that one Mac mini), all of which came equipped with dual-core i5s but at profoundly underclocked base speeds and non-upgradeable internals.

Which one was underclocked? Just looking at some of the links you presented above, it seems they're clocked at their specified base frequencies...

For example, the 2014 model:

https://ark.intel.com/content/www/u...54260u-processor-3m-cache-up-to-2-70-ghz.html

Specified by Intel at 1.4GHz, clocked by Apple at 1.4GHz.

 

[B S Magnet]

Which one was underclocked? Just looking at some of the links you presented above, it seems they're clocked at their specified base frequencies...

For example, the 2014 model:

https://ark.intel.com/content/www/u...54260u-processor-3m-cache-up-to-2-70-ghz.html

Specified by Intel at 1.4GHz, clocked by Apple at 1.4GHz.

I should have been clearer: significantly underclocked, relative to all the other product offerings by Apple from the same round of releases within the same product line.

Which products? These four:

iMac "Core i5" 1.4 21.5" (Mid-2014) Specs (Mid-2014, MF883LL/A, iMac14,4, A1418, 2805): EveryMac.com

Technical specifications for the iMac "Core i5" 1.4 21.5" (Mid-2014). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

iMac "Core i5" 1.6 21.5" (Late 2015) Specs (Late 2015, MK142LL/A, iMac16,1, A1418, 2889): EveryMac.com

Technical specifications for the iMac "Core i5" 1.6 21.5" (Late 2015). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

iMac "Core i5" 2.3 21.5" (Mid-2017) Specs (Mid-2017, MMQA2LL/A*, iMac18,1, A1418, 3068): EveryMac.com

Technical specifications for the iMac "Core i5" 2.3 21.5" (Mid-2017). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

Mac mini "Core i5" 1.4 (Late 2014) Specs (Late 2014, MGEM2LL/A, Macmini7,1, A1347, 2840): EveryMac.com

Technical specifications for the Mac mini "Core i5" 1.4 (Late 2014). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

These were all sold during a 2014–2017 window, coincident with when Apple released relatively few product refreshes (for the iMac) and none for the Mac mini. All four, speed-tested, were a significant order slower than their faster counterparts within the same product release, the latter of which were much closer to one another in speed-tested performance results.

The most egregious case instance is in the mid 2014 21.5-inch iMac 1.4GHz dual-core i5 (first link), whose Geekbench 2 average in 64-bit mode was 7399; its direct antecedent/predecessor, the early 2013 21.5-inch 3.3GHz Core i3, clocked in at 8262. The retail price between the pair was identical. The other 21.5-inch 2014 offerings’ Geekbench 2 averages — all carryovers from late 2013, sold through late 2015 — were closer: 10123, 10906, and 14981 for the highest end (which, notably, used the i7 CPU, whereas the others used the i5).

 

[Analog Kid]

I should have been clearer: significantly underclocked, relative to all the other product offerings by Apple from the same round of releases within the same product line.

Which products? These four:

iMac "Core i5" 1.4 21.5" (Mid-2014) Specs (Mid-2014, MF883LL/A, iMac14,4, A1418, 2805): EveryMac.com

Technical specifications for the iMac "Core i5" 1.4 21.5" (Mid-2014). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

iMac "Core i5" 1.6 21.5" (Late 2015) Specs (Late 2015, MK142LL/A, iMac16,1, A1418, 2889): EveryMac.com

Technical specifications for the iMac "Core i5" 1.6 21.5" (Late 2015). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

iMac "Core i5" 2.3 21.5" (Mid-2017) Specs (Mid-2017, MMQA2LL/A*, iMac18,1, A1418, 3068): EveryMac.com

Technical specifications for the iMac "Core i5" 2.3 21.5" (Mid-2017). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

Mac mini "Core i5" 1.4 (Late 2014) Specs (Late 2014, MGEM2LL/A, Macmini7,1, A1347, 2840): EveryMac.com

Technical specifications for the Mac mini "Core i5" 1.4 (Late 2014). Dates sold, processor type, memory info, hard drive details, price and more.

everymac.com

These were all sold during a 2014–2017 window, coincident with when Apple release relatively few product refreshes (for the iMac) and none for the Mac mini. All four, speed-tested, were a significant order slower than their faster counterparts within the same product release, the latter of which were much closer to one another in speed-tested performance results.

The most egregious case instance is in the late 2014 21.5-inch iMac 1.4GHz dual-core i5 (first link), whose Geekbench 2 average in 64-bit mode was 7399; its direct antecedent/predecessor, the early 2013 21.5-inch 3.3GHz Core i3, clocked in at 8262. The retail price between the pair was identical.

They aren't underclocked. They are clocked at their specified speed. Just as you explained that the lower performance 2009 product that Jobs released was "entry level", so is a 1.4GHz i5-4260U.

 

[B S Magnet]

They aren't underclocked. They are clocked at their specified speed. Just as you explained that the lower performance 2009 product that Jobs released was "entry level", so is a 1.4GHz i5-4260U.

Pedantry aside, OK, Analog Kid.


EDIT to add: The 1.4GHz Core i5 was designed by Intel for mobile devices.

Even if clocked officially at 1.4GHz, Intel downclocked a nominally faster, by design, iteration of the fourth-gen, 22nm Core i5 (the Haswell) to offer that product for mobile hardware, where cooling was constrained and ultra-low power consumption was needed.

At least three faster, cheaper desktop CPUs from the i3 line released the same quarter as the 1.4Ghz i5 (i.e., faster than the Ivy Bridge 3.3GHz Core i3 in the base 2013 iMac and faster than the 1.4GHz i5-4260U in the base 2014 iMac) could have been offered by Apple for an entry iMac.

Perceptually and practically speaking, the mid-2014 iMac at 1.4GHz was downclocked, relative to its direct predecessor (3.3 to 1.4GHz); slower than its predecessor (in Geekbench testing); and by design, relied on a CPU underclocked by Intel for use in said low-power, limited cooling mobile hardware (which the 2014 base iMac and the 2014 Mac mini were not).

 

[Amethyst1]

EDIT to add: The 1.4GHz Core i5 was designed by Intel for mobile devices.

Even if clocked officially at 1.4GHz, Intel downclocked a nominally faster, by design, iteration of the fourth-gen, 22nm Core i5 (the Haswell) to offer that product for mobile hardware, where cooling was constrained and ultra-low power consumption was needed.

To be more precise, the Core i5-4260U is also used in the 2013/2014 MBA. The low base clock is partially mitigated by high turbo boost clocks: 2.7 GHz for one core, 2.4 GHz for both cores, and assuming the iMac's or Mac mini's cooling solution can deal with heat better than the MBA's, turbo speeds may be sustained for longer periods of time. It's still clocked lower than the Ivy Bridge configuration it replaced as you say though.

 

[Bug-Creator]

That’s why the original iBook was the second fastest laptop in the world when released, only slower than the Apple PowerBook. That’s why the Apple eMac (education Mac) shipped with a powerful new G4 processor, instead of the G3.

There is something really disappointing about seeing those underspecced iMacs from recent years. Same with the new budget 2021 iPad, which uses A12 instead of A15.

Going back from a G4 to a G3 meant going back a full generation instead of just going down 1 or 2 minor annual refreshes.

Running on 5 year old HW in 2000 was far more painful as running on 10 year old stuff today, so the question does a consumer iPad really need anything beyond an A12 and will that change in it's expected lifetime?
I certainly don't feel the need for more CPU on my XR and don't expect that to change anytime soon.

It might also be the case that is using all of the 5nm the have ordered for other products and as such shipping the iPad with 7nm has a key role in hitting that price point.

 

[mectojic]

Going back from a G4 to a G3 meant going back a full generation instead of just going down 1 or 2 minor annual refreshes.

Running on 5 year old HW in 2000 was far more painful as running on 10 year old stuff today, so the question does a consumer iPad really need anything beyond an A12 and will that change in it's expected lifetime?
I certainly don't feel the need for more CPU on my XR and don't expect that to change anytime soon.

It might also be the case that is using all of the 5nm the have ordered for other products and as such shipping the iPad with 7nm has a key role in hitting that price point.

Slight clarification needs to be made here- the G4 had just been released to the iMac G4 in 2002. The iBooks kept using high-end G3s for a few more years, and G3 support continued until 2007. So the 2002 eMac could've easily been a G3, but Apple still wanted the education market to have the best.

By today's standards, the analogy would be having 3-4 year old G3s for everything except for the high-end Power Mac. Old Apple was putting M1 Pro into all their products, even their cheapest, and saved just the M1 Max for their highest end machine.

 

[Bug-Creator]

Old Apple was putting M1 Pro into all their products, even their cheapest, and saved just the M1 Max for their highest end machine.

In that case we should be happy that "old Apple" is dead, because putting an M1Pro into something like a MBA or iPP would not yield any performance boost as the M1 (which is the same generation) is already thermal throttling in those systems.

 

[MajorFubar]

Great observations but this is really typical of what happens every time some new technology is invented: every new release is exponentially better than the last, then after a while a plateau is reached, and the improvements become more evolutionary than revolutionary. My home studio is based around an 11 year old iMac I bought from the Apple Refurbished store when it was 3 months old. But in 1990 when my home studio consisted of two synthesizers and a drum machine MIDI'd to a Commodore Amiga, it's unthinkable that I would have even considered using a computer from 1979. It would have been completely impossible.

It's happening now with smartphones. Ten years ago you didn't find people using any smartphone more than two years old because it was too obsolete. In 2022, I don't have to look far round the office I sit in everyday to see people using phones 4-5 years old perfectly happily.

 

[iPadified]

Very interesting! Personally, I think an edited version of this would fit the front-page of MR.

 

[uller6]

To be more precise, the Core i5-4260U is also used in the 2013/2014 MBA. The low base clock is partially mitigated by high turbo boost clocks: 2.7 GHz for one core, 2.4 GHz for both cores, and assuming the iMac's or Mac mini's cooling solution can deal with heat better than the MBA's, turbo speeds may be sustained for longer periods of time. It's still clocked lower than the Ivy Bridge configuration it replaced as you say though.

My mother-in law has the low end 1.4 GHz dual core 2014 21.5" iMac. The cooling system is great, so the machine is always running at max turbo boost speeds no matter the workload. In fact, the cooling system is so good (designed for the quad core CPUs) that it's one of the quietest Macs I've ever heard. That said, it is still not a fast computer.

 

[Analog Kid]

Pedantry aside, OK, Analog Kid.


EDIT to add: The 1.4GHz Core i5 was designed by Intel for mobile devices.

Even if clocked officially at 1.4GHz, Intel downclocked a nominally faster, by design, iteration of the fourth-gen, 22nm Core i5 (the Haswell) to offer that product for mobile hardware, where cooling was constrained and ultra-low power consumption was needed.

At least three faster, cheaper desktop CPUs from the i3 line released the same quarter as the 1.4Ghz i5 (i.e., faster than the Ivy Bridge 3.3GHz Core i3 in the base 2013 iMac and faster than the 1.4GHz i5-4260U in the base 2014 iMac) could have been offered by Apple for an entry iMac.

Perceptually and practically speaking, the mid-2014 iMac at 1.4GHz was downclocked, relative to its direct predecessor (3.3 to 1.4GHz); slower than its predecessor (in Geekbench testing); and by design, relied on a CPU underclocked by Intel for use in said low-power, limited cooling mobile hardware (which the 2014 base iMac and the 2014 Mac mini were not).

You keep saying "underclocked", or "downclocked". It was neither. It was running at it's specified speed. It was slower, but that's not the same thing. "Underclocked" is the opposite of "overclocked" and both imply running a device at a clock different than it is specified to run at. Overclocked means you're pushing it beyond its design limits. Underclocked could mean you're being conservative for some reason, such as reliability, but given your tone it's meant to imply sabotage or intentionally degrading the performance below what the component can provide.

My point was only, and simply, that Steve Jobs was perfectly happy to ship products at reduced performance relative to the rest of the line if it met his goals of providing a product he could sell to education at a margin he was happy with.

Calling one "entry level as their placement would indicate" and one "deliberately underclocked" is only coloring factual data with your subjective and unsubstantiated suspicions of intent.

 

[ahurst]

As Apple itself is advocating the use of mobile devices also for computationally demanding tasks, what do you think about a similar visual analysis of iOS device specs?

Okay, so this isn't *perfect* but it gives a pretty good look at Mac vs iOS performance over time. I'm using Geekbench 4 scores here since they cover the widest range of iOS devices (iPad 2/iPhone 4S and up). A side-effect is that Apple Silicon scores are handicapped a bit, since Geekbench 4 isn't ARM-native and is thus running through Rosetta 2. I'm also excluding workstation scores from the Mac data again (Xserve, Mac Pro, iMac Pro) since they sort of obscure the trends for regular Mac models:

As you can see, iPhone/iPad multicore performance kept improving at a constant "Core 2 Duo to Sandy Bridge" rate from the iPhone 4S/iPad 2 up to around 2017, where the average iOS multicore caught up to "lower-midrange" Mac multicore and has kept pace ever since. The mid-2010's Intel slump looks even more pronounced against its rate of progress. As for single-core scores, well...

Again, remember that the M1/M1 Pro Mac scores are penalized by Rosetta 2 here, but holy heck did Apple's single-core CPU performance grow at an impressive rate. The rate of improvement's slowed a bit since 2017, but still improving fast enough to reach midrange Mac levels in ~2017 and overtake the fastest Intel Macs in 2020. I'll be interesting to see whether they can keep it up!

By today's standards, the analogy would be having 3-4 year old G3s for everything except for the high-end Power Mac. Old Apple was putting M1 Pro into all their products, even their cheapest, and saved just the M1 Max for their highest end machine.

To be fair, PowerPC Macs also offered a range of different CPU clock speeds for each product and model year around the same cutting-edge chips, just like the M1/Pro/Max offer different CPU core configurations and GPU configurations for different models. I'm still sad they only offered my favourite iMac G3 (slot-loading Blueberry) with the lowest-spec 350 MHz chip...

Very interesting! Personally, I think an edited version of this would fit the front-page of MR.

Hey, if any of the editors are reading this, feel free to send me a PM!

 

[B S Magnet]

You keep saying "underclocked", or "downclocked". It was neither. It was running at it's specified speed. It was slower, but that's not the same thing. "Underclocked" is the opposite of "overclocked" and both imply running a device at a clock different than it is specified to run at. Overclocked means you're pushing it beyond its design limits. Underclocked could mean you're being conservative for some reason, such as reliability, but given your tone it's meant to imply sabotage or intentionally degrading the performance below what the component can provide.

Dang, you must be a joy at parties.

My point was only, and simply, that Steve Jobs was perfectly happy to ship products at reduced performance relative to the rest of the line if it met his goals of providing a product he could sell to education at a margin he was happy with.

Steve Jobs had been dead for nearly three years when Apple opted to produce these consciously underpowered models, coincident with the period in @ahurst’s data plots when speed improvement across the entire Apple Mac product line plateaued and, for the desktop models, even slipped slightly — no doubt dragged down by this product release decision.

That was the point of mentioning these products. Corporate spin and fanboy spin, Analog Kid, doesn’t alter the outcome. They were downclocked products relative to their immediate and direct generational predecessors. Their performance scores actually dropped relative to their immediate, entry-level predecessors — something unprecedented in Apple’s product release history. And Apple’s practice of this kind of product release was, at least this time, halted by 2018.

Calling one "entry level as their placement would indicate" and one "deliberately underclocked" is only coloring factual data with your subjective and unsubstantiated suspicions of intent.

Referenced citations and supporting technical descriptions from company web sites are difficult for you. I get it, and hey, that’s OK. You can continue to enjoy many other venues and forums across the internet where you can indulge in your bubble of semantic reasoning. But on here, you will enjoy no such bubble.

 

[mectojic]

To be fair, PowerPC Macs also offered a range of different CPU clock speeds for each product and model year around the same cutting-edge chips, just like the M1/Pro/Max offer different CPU core configurations and GPU configurations for different models. I'm still sad they only offered my favourite iMac G3 (slot-loading Blueberry) with the lowest-spec 350 MHz chip...

Incorrect. The Blueberry slot-load could be bought with the 400MHz speed, with airport and firewire https://everymac.com/systems/apple/imac/specs/imac_dv_400.html