There are a lot of inaccuracies in your memories. The original iMac came out in 1998. At the time Apple’s G3 processors were very competitive with everything offered by Intel:

http://macspeedzone.com/archive/4.0/WinvsMacbytemark.html

And you were not getting double performance machines for half the price of the iMac. You are also incorrect about the performance situation one year later which would be 1999. Yes, that was the year the top of the line Pentium III came out at 600 MHz, but it was also the year that the top of the line Power Mac G4 came out at 500 MHz (a machine I owned after it was delayed because yields on the 500 model were poor (we were offered a 450 MHz part as a replacement)). The G4 500 was superior to the P3 600 in many benchmarks, and crushed it in others thanks to the altivec vector unit:

http://macspeedzone.com/archive/4.0/g4vspent3signal.html

I just linked to one pretty biased site here (first that came up on Google). But your extraordinary claims need some source because I don’t remember it like that at all in 1998 and 1999. PPC was a solid contender throughout the late 90s.

Also the original iMac had a 15” screen, not 13”. I had an iMac too.

So yeah, the G4 was perhaps winning the IPC battle, Intel and AMD were more than making up for it with higher frequencies.

Somehow the various RISC vendors managed to remain competitive for a bit longer at the top end[0] - maybe it was easier to compete at the high workstation/server/supercomputer end than at consumer hardware?

[0] https://en.wikipedia.org/wiki/Supercomputer#/media/File:Proc... - x86 only started dominating well into the 2000s.

"We are starting to see some great games come back to the Mac, but this is one of the coolest I've ever seen...this is the first time anybody has ever seen it, the first time they've debuted it...Halo is the name of this game, and we're going to see, for the first time: Halo."

-Steve Jobs, MacWorld 1999

https://www.youtube.com/watch?v=6eZ2yvWl9nQ

I think it’s Moore’s law approaching its limits and the direction of chip improvements isn’t core speed but the number of them, power usage, etc which make a difference but for most folk it doesn’t appear like an improvement like say doubling the frequency every 18 months. I have an old laptop and it keeps up quite nicely after 8 years...

The first 14nm processors started shipping in the 15" in 2015 - 5 years ago.

My current 8 year old macbook has a 22nm processor. I would never have thought 8 years ago that Intel would only have managed a single node shrink since then.

I'm jealous and genuinely curious where do you guys work that your employers can afford to get everyone such expensive machines.

I've been a dev in the EU for 8 years now and at most places I've worked or interviewed(not FAANG) the machines you get are some cheapo HP/Lenovo/Dell with only the executives having Apple hardware.

I never understood why companies in the west cheap out on hardware so much since compared to the cost of office rent and employee salaries that's a drop in the ocean, they could buy everyone MacBooks or Ryzen towers and it wouldn't even dent their bottom line.

This is pretty typical here.

Heck we swap build servers every 6-12 months based on test speed. We buy one of every new CPU and do a test run of auto build and ptest. If it is reasonably faster we order a new rack and replace the old boxes. Power and cooling is way more $$ then the HW. Every month we do not need a new cage is a win. We are deploying AMD Epyc now with 10G-T + 4x1G (not network limited in our test, just segmentation to test DUTs) with a core of 100G for fileservers and 25G for services. File servers are TrueNAS SSD shelves for test and with spinning rust for build artifacts. We run ~1000 containers on a single server in ptest (scaling that was fun to figure out ... hint you have to play with networking stack ARP timers - strace is your friend).

In a place where talent is as competitive as the bay, you wouldn’t survive making people use subpar machines.

But most businesses here are the same, you get lucky to get any nice feature over the 'same laptop that sales gets' which is barely more than a Chromebook. And getting an external monitor that's not the cheapest bulk-buy model was also pretty hard to do (I had a friend in marketing who helped me get a larger display with better colors at that place).

Frankly with the cost of hiring you, spending $3k to make you even a few percent more productive if money well spent.

A high performance SSD, capable of over 3GB/s, is less than $200 for 1TB.

But seriously, a 970 Evo Plus is a high performance SSD suitable for a workstation and that's $400 for two.

Though I would gladly dump the MacBook Pro 16" I currently have to use for work in an instant for a high-end Lenovo/Dell. Apart from MacOS being extremely flakey these days (why does Spotlight only seem to pop up 50% of the time), I don't understand why they don't provide proper ISO and instead give us some form of ANSI that has the dancing alien (§) dedicated key, and why they hide the # which as a Pro developer I use all the time.

That it also spends its entire time overheating so it burns my lap is just the icing on the cake.

But when I used a MacBook Pro about 10 years ago the machine overheated all the time and burnt my lap. They are just a shitty design, but they look pretty.

Plus it seems to disable the switch between windows in the same app.

I think if you have any way of escalating then the best thing is to come up with numbers, such as "a faster computer would let me compile the program in 20 less seconds, which is this much time earned", or "a better screen would not require me to have an external monitor".

Now, I think that to get a 16" MBP for work requires quite a specific use case, because it's really a machine one should use only when it's the only computer they have. For the same price I think you could get a faster desktop + a more portable laptop.

The only difference? I can receive encrypted emails only on the Windows laptop due to some software not being available for the Mac.

I do all my development on macOS, and if spending $4k on a machine means I am more productive, can get work done faster, it's a return on investment that pays back multiple times.

Do note that laptop refreshes in my past companies (current is ~2 years) have been on average every 2.5 year... so it's not like I get a new laptop yearly.

That Dell Latitude is frustrating to use. The trackpad is absolutely atrocious, the display is dim, the keyboard is really mushy and causes pain in my hands when I use it for short periods of time...

I'm in the US, but I've been in a similar position to you; my last job gave me a tower with 8GB of ram and a Core 2 Duo running Windows 7 32-bit. Utterly useless. I had to sneak Ubuntu onto it when nobody was looking. They couldn't even tell the difference.

I ended up having the last laugh when every computer in the office got wiped because somebody decided running the mail server on our ActiveDirectory server was a good idea, and also thought that leaving ports open on the mail server was a good idea.

Last place I worked I got an HP Windows desktop and told to remote in from a 12" laptop whenever I needed to work from home or show off something in a meeting. That place also wanted to knock down three single-person offices so they could fit 10 developers in the same space. And for the last month or so I had a developer working next to me on my own desk since management didn't prioritise us developers over HR.

My current workplace and my previous are both in the public sector in Norway.

On the other hand if you like it, you like it, and cheap is beautiful.

Recently I replaced 4 Asus monitors with led backlights that were produced in 2014 and 2015. Asus says 300 nits. Tested them when I was calibrating their replacements and they were 110-120 at full brightness.

Monitors color shift and dim as they age... that’s why hardware calibrators exist.

So if we want increase our chip's performance any further, we need a fundamental change in our technology. And since silicon atom is 210 pm big and carbon atom (graphene is made from carbon) is 170 pm big, we need to change our architecture as shrinking transistors will not be possible anymore. I mean CISC/RISC, drop x86 support and so on.

Eh, Intel's 14nm is 37M transistors/mm2. TSMC and SS are both up to 52M/mm2 at 10nm, and 92M/mm2 at 7nm. Both Apple and AMD's latest gen stuff is on TSMC's 7nm process _today_. Yes, Intel's 10nm is at 101M/mm2, but until they can get mass production on that they're falling substantially behind.

(I stand by "most", but it's nice to hear that someone's making actual progress again.)

Intel and AMD might be able to deliver somewhat higher performance by throwing a whole bunch of cores at the problem, but they do so at a much higher cost in power requirements. And it would be easy enough to design ARM machines with an equal number of cores (or even way more), and still have much lower power requirements.

Intel stagnated, and has high power requirements. ARM has caught up, and has much lower power requirements.

The reasoning seems pretty simple to me.

ARM hasn't. Apple has. Apple's implementation of the ISA since A6 is unrelated to ARM's.

Also Intel and AMD have not delivered higher performance via more cores; they delivered lower price for (say) 64 cores worth of performance, by putting them all on the same chunk of silicon (edit: or at least in the same package). (There are some slight improvements in inter-processor interrupt and cache-forwarding latency, but if that's a performance bottleneck, the problem is bad parallelization at the code level.)

Have you looked at the encoding of Thumb-2 (T32) and particularly A64 (their newly designed instruction set for 64 bit)? Their instruction encoding is in my opinion much more convoluted than x86.

Arguably AMD did the exact opposite - lower prices via splitting a processor into multiple pieces of silicon. (Chip price scale exponentially with area at the high end)

    Sure, but that's them having a competent 
    (well, less incompetent) ISA and microarchitecture

Note that this is still Moore's law. You're thinking of Dennard scaling. That basically ran out 12 years ago.

Actually, it was about nobody wanting to deliver a Northbridge for Apple.

I interviewed at Apple in this timeframe and was stunned that they used Northbridge ASICs with synchronizers everywhere. No clock forwarding to be found.

This kills memory and graphics performance dead.

On top of that the support ASICs were using more power than the CPU!

Once Apple switched to x86, they could use the Northbridge and Southbridge chips that everybody in the universe was using.

My memory isn't so good, and I never owned a P4, but according to Wikipedia in August 2001, they released one at 2GHz on 180 nm. Almost 20 years later, my laptop i7 is running at...2.1GHz (base)? And 14 nm. That's kind of mind boggling. I think it would be interesting to read/write an article comparing the two in depth and what performance benefits you get from the newer chip.

https://www.cpubenchmark.net/year-on-year.html

The frequencies may not have increased beyond 2-3 GHz but the speed has still got faster because modern processors are much smarter and are able to do more work per cycle. They have all sorts of fancy tricks to do that - speculative execution, hyperthreading, branch prediction, etc.

(More cores aren't a performance improvement; if you were willing to deal with non-serial execution, you could have just bought 32 Pentium Fours; putting them all on the same chip is convenient (and cheap), but as a price/performance improvement, it's all price, no performance.)

That's only true if you only consider ALU throughput for performance, but in terms of real world performance, where the interconnect between cores and memory is hugely significant, a multicore processor has many advantages over a rack of otherwise equivalent single-core NUMA nodes.

My guess is that there are now a lot of forms of hardware acceleration of specific things that make your daily experience seem faster, but I haven't seen them catalogued and put in perspective with measurements.

I would think that and cache size are large differences between a P4 and the i7 running at about the same clock frequency.

I haven't read about P4 and NetBurst in a long t ime, but is mu memory serve me right, P4 is usually less than 1ipc due to its very long pipeline (31 stages IIRC) that is very prone to pipeline stall. Modern processor can also do many thing faster. I think P4 takes ~110 cycles for integer division, while modern cpu can do in ~30-40 cycles. And IIRC, P4 cannot flush the division unit so if branch/jump prediction is wrong and division is speculated, it was to wait until the division unit finish computation before it can resume execution.

All the modern research suggests very much that the decode stage isn't a significant difference nowadays; instruction density is increasingly significant as CPUs become ever faster compared with memory access.

x86 makes it really awkward to pull out several instructions at a time for parallel decoding.

x86 instruction size ranges from 8 bits to 120 bits (1-15 bytes). Since common instructions often fit in just 8 or 16 bits, there are power savings to be had due to smaller I-cache size per instruction. That comes at a severe decode cost though as every single byte must be checked to see if it terminates the instruction. After the length is determined, then it must decide how many micro-ops the instruction really translates into so they can be handed off to the scheduler.

ARM breaks down into v7 and v8. The original thumb instructions were slow, but saved I-cache. Thumb 2 was faster with some I-cache savings, but basically required 3 different decodes. ARMv8 in 64-bit mode has NO 16-bit instructions. This reduces the decoder overhead, but obliterates the potential I-cache savings. No doubt that this is the reason their I-cache size doubled.

RISC-V is not being discussed here, but is the most interesting IMO. The top 2 bits in an instruction tag it as 32-bit or 16-bit (there are reserved bit schemes to allow longer instructions, but I don't believe those are implemented or are likely to be implemented any time soon). This fixed bit pattern means that length is statically analyzable. 3 of the 4 patterns are reserved for 16-bit use which reduces the instruction size penalty (effectively making them 15-bit instructions). The result is something 3-5% less dense than thumb 2, but around 15% MORE dense than x86 all without the huge decode penalties of x86 or the multi-modes and mode-switching of thumb. In addition, the effect of adding RVC instructions reduces I-cache misses almost as much as doubling the I-cache size which is another huge power consumption win while not having a negative impact on overall performance either (in fact, performance should usually increase for the same I-cache size).

I think this is the key reasoning. There's something really interesting happening here. In the past, when Apple transitioned from Motorola to PowerPC Apple wasn't big enough to design and fab their own chips, this was also true when they moved from PowerPC to Intel.

However, Apple has some choices here, and I think the decision comes down to long term supply-chain risk:

1) Switch to AMD. Their processors are blowing the doors off of Intel, at better prices. They aren't having the same process problems and their high-end components are fantastic. However, history has shown AMD surge ahead for a bit, then Intel, and back again. Apple probably doesn't want to risk this happening after they engage in some huge volume discount contract. More importantly, neither Intel nor AMD are winning in the lower power vs performance segments.

2) Become their own fabless designer. The risks are enormous. What if they can't keep pushing the performance envelope against Intel/AMD? What if their fab partners can't keep their processes moving forward? What if they fail to make this architecture jump (again)? But if gives them better supply chain control and increases their vertical integration.

In some sense it points to a weakness of highly vertically integrated companies...as a model it makes their entire product lines dependent on every component being able to progress. If any component lags, the entire product line suffers. So outside suppliers, who have multiple customers to please, become key sources of risk and it will become the instinct of the company to move the riskiest parts of its supply chain in-house.

If Apple is unable to keep advancing ARM chips in terms of performance (regardless of power) it will be a problem for them. But one final advantage of building their own, is that they can obfuscate this component from the rest of the market and make comparisons on cores/Ghz/etc virtually impossible. It's a bit like how Apple really doesn't even advertise how much RAM is in their portable devices.

Note there were rumours going around about whether or not Apple was going to buy PA Semi (or at least contact them) for their PWRficient CPU (implementing PPC).

Ultimately they did buy PA Semi in 2008, though for the skills and they've since made all of the iPhone/iPad CPUs.

I suspect that Apple wants way bigger caches. And chip size (especially RAM) directly impacts yield and cost.

Apple is probably more than happy to double the cost of the CPU if the extra cache gets them 20% more battery life.

Nobody in the commodity CPU market would ever make that trade.

This sentence makes no sense: Intel (and AMD) sell very expensive chips if you ask them to take your money.

Look, for example, at the prices listed at https://www.cpubenchmark.net/high_end_cpus.html

Just to list some arbitrary expensive examples from this list:

  AMD EPYC 7702: $6,200.00
  AMD EPYC 7742: $7,416.51
  Intel Xeon W-3275M: $7,453.00
  Intel Xeon Platinum 8168 @ 2.70GHz: $5,890.00
  Intel Xeon W-3265 @ 2.70GHz: $4,551.37
  Intel Xeon Platinum 8268 @ 2.90GHz: $6,005.98

But who wants an expensive low power chip? Only those who make phones. aka Samsung and Apple.

The fact that our low-power chips also happen to be lower die size is an artifact of path dependence. The primary market for lower power was battery powered devices of which phones are by far the most numerous. So, the lower power chips started there and didn't have much to do. As those have gained sophistication with time, they have also grown in die size.

Xeon and server chips generally want to maximize memory bandwidth--and they make a whole series of architectural tradeoffs to accommodate that.

Phone chips generally want to maximize power efficiency and basically don't care about memory bandwidth at all. They effectively don't want to turn on the system memory or flash, period, if they can avoid it. One way to do that is to cache things completely in local on-chip RAM.

Computer architects will make completely different tradeoffs for the two domains.

A big problem was that IBM was not designing low-power chips for laptops, and laptops were (and are) a major part of Apple's business.

Power6 (not PowerPC) hit 5 GHz in 2007 and Power has remained competitive - Power10 will be described at Hot Chips in August. Of course (perhaps consistent with "Hot Chips") these are not low-power architectures.

Don’t know how much I would agree with that. I just took a look at the Megahertz Myth portion of the 2001 Macworld Expo video¹ where Apple compares an 867 MHz G4 processor with a 1.7 GHz Pentium 4 processor, and while I don’t know how valid Apple’s argument is in that video (don’t know enough about how processors work to judge), from the information they give, it does seem plausible that the 867 MHz G4 could outperform the 1.7 GHz Pentium 4 in some scenarios. Frequency certainly isn’t everything, especially when we’re comparing two different CPU architectures.

――――――

¹ — https://www.youtube.com/watch?v=TJN5GfZuVog

Take the 867MHz G4 you mentioned. There might have been some applications where the G4 was beating a 1700MHz Pentium 4, but at the time of the demo the top-of-the-line from Intel was 1800MHz and they released a 2GHz only a few days later. A year later Intel was shipping 2.8GHz parts, and Apple was selling 1.25GHz G4s. So whatever architectural lead the G4 enjoyed, Intel was eroding it with faster clock scaling.

This does not even mention the mobile space, where in 2003 Intel was shipping the Pentium M, not the Pentium 4, and it was the Pentium M which derived from the Pentium Pro/II/III and foreshadowed the Core product line. The G4 had no architectural advantages over the Pentium M. Apple's mobile products were stuck on the dead-end G4 for years.

I owned a Mac of some kind throughput the PowerPC era but it was only because I had to run Mac applications. There wasn't anything good about them except on rare occasions you got to see the AltiVec unit really go crazy. Most of the time you just got to marvel at how slow and expensive they were compared to the other PC on your desk.

Now the tables have turned though. MacOS software has suffered greatly while MS has embraced Linux and the Terminal. Win10+WSL+Windows Terminal+VS Code today is the superior toolchain IMO because it gives you access to the package managers that will also run on your target servers.

Moreover, the enterprise market had lost that war already. Who cares if PowerPC was really faster? Intel was a cheaper chip that did the job. Everyone in my customer service, accounting, HR, etc department can have a PC for much cheaper than Apple.

My memory is more like they leapfrogged each other from time to time. The first generation of PowerMacs such as the 6100 absolutely spanked contemporary PCs (100 Mhz 486s and 50 Mhz Pentiums if I remember correctly). It was by no means obvious at that time that Intel would catch up.

What killed PPC was the interested parties (IBM, Motorola, Apple) squabbling over CHRP, and Motorola being unwilling to work on a part that would fit the thermal envelope Apple wanted for laptops. The fundamental architecture of PPC is sound, or at least, sounder than that of x86.

Back then it was because the monitor and hardware were carefully calibrated - the gamma was strange but the colors were spot on (assuming correct ambient lighting). An Apple made a lot of sense for visual artists of all disciplines, they still generally have this edge today.

Apple is already working with AMD when it comes to dedicated graphics chips, and people have done Hackintoshes with AMD parts for ages... so why go the (risky) ARM route instead?

https://www.eetimes.com/tsmc-to-keep-supplying-chips-to-huaw...

Same issue here - they’ll switch to ARM for more control. AMD has some good hardware, but they’d just end up switching one outside chip vendor for another. So while a move to AMD would be “safer”, it still wouldn’t offer any more control over their products.

That’s my guess.

It's ultimately an intermediary step. Apple already has the tech for making multi-arch software, it's already a part of the toolkit, and they already design and sell ARM chips in their other product lines. It's not a no-risk scenario, but it's a low risk transition from a company that's managed two architecture migrations before.

The Mach-O object file format supports fat binaries and has run working executables on ARM, SPARC, PA-RISC, PowerPC 32-bit, PowerPC 64-bit, x86 and x86_64 first under NeXT and eventually under Apple. It's what every implementation of Mac OS X, iOS and its derivatives use today and there's nothing stopping Apple from support other architectures down the line other than they probably don't want to or need to. If they decided they wanted to revitalize Mac OS X server and ship them with POWER9 or RISC-V CPUs, they could do that. Not sure why they want to, but it's an option.

If they're going to end a working business relationship with a supplier anyway and they have CPUs that outclass their current supplier in CPU performance (an A13 in a $400 iPhone SE outclasses the Xeons in their $6K Mac Pros in single threaded performance), they might as well go the whole hog, skip the intermediary step of switching suppliers (which might mean signing some kind of multi-year deal during which Intel might start outclassing AMD, so that's not without risk either) and run their own designs through TSMC fabs.

Apple has always been that company that likes to own the whole widget, or as much of it as possible. If they didn't, they might have switched to Windows NT in the late 90s rather than buying NeXT. They don't make everything that goes into a Mac or iPhone, they don't even fab the CPUs in their iPhones, just design them, but they figured out which parts they can control the designs for and which they can source from others and made it work by progressively integrating more of the design work in-house.

That said, I keep virtual machines of operating systems that interest me up to date. I'm between 9front and Haiku OS as my eventual replacement, and I still might switch to a ThinkPad running 9front for my next computer regardless of what Apple announces. I'll still have much of what I value on a Mac in my iPad, and my laptop is essentially for writing, programming and backing up other hardware.

The 90's Apple, like everyone else, had its own vertical integration in software programming stack, network protocols and hardware. Also not every model had internal expansion slots, what we bought was what we got for the device's lifetime.

Naturally the more expensive LC and Quadras had enough internal bays, given their business purposes.

So if somewhere down the line Apple makes the CPUs in all of their Macs and can't compete, they have no one to blame but themselves. Right now they're not really failing to compete though, as their computers are stuck in the same holding pattern everyone else who depends on Intel is. They could switch to AMD parts, but they're trading one horse out for a horse of the same breed that's maybe a bit younger and prettier.

So if they're angling to axe their relationship with Intel down the line, at least for CPUs, why bother switching suppliers temporarily when they can switch their existing supplier out for something they designed in-house when they're ready and skip the intermediary step?

All reports I've seen over the past month seem to indicate that time is upon us. There's enough smoke that if Apple hasn't quietly passed a memo to the WSJ mentioning that they're not announcing anything of the sort on Monday by the close of yesterday, then the time is most likely now (well, Monday for the transition announcement, likely 2021 for the first shipping products).

The far more likely explanation for all of this is changes to Intel’s anti competitive rebate regime.