This. Building a PC, or using a tower at all is exculsively the realm of gamers. And even they hate the state of the hardware business as it dies off to low volume. By 2030 even most gaming "towers" will be based on SoCs, there will be no way to make a system with discrete RAM, CPU and GPU competitive with what a single high bandwidth, low latency package can do. It's not even that iGPUs will catch up to dGPU performance, it's that the concept of a dGPU will hit a performance ceiling and only putting the GPU, CPU and RAM on package will allow the envelope to be pushed further. No matter your budget in time, money and space you could not build a modern PC out of vacuum tubes. Or standalone transistors. Or simple ICs. And soon, there will be no way a GPU or RAM module connected by a hundred or so pins over a ~12" copper trace can compete with what can be done by putting everything nanometers apart. RAM stacked directly over the die, CPU and GPU cores separated only by nm scale memory controllers, etc. This is always been coming. Hardware always evolves into more tightly integrated components as the miniaturization of manufacturing allows.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro 2023 = Another Mac Pro 2013 crisis and failure.
- Thread starter sunny5
- Start date
- Sort by reaction score
Spot on. Current trends would indicate that all servers will concentrate into a few commercial cloud computing providers over the next decade or so. And those will continue to get more and more specialized, tasking nVidia, AMD, intel, TSMC, global foundries, etc. to design and or build hardware to their specs. Kinda like an exact parallel of what apple has done on the consumer side. Funny, almost like they have a plan and understand where the market is going.
Long word in short: Mac Pro 2023 is a proof that Mac Pro market is failing since they cant even make a powerful chip. Nothing else to say.
the mistake here is assuming that 5 gallons = 1 bucket.
unfortunately, buckets aren't standardized in the SI or SAE world
(buckets for excavators and bobcats are much larger than 5 gallons)
Completely not my point at all. The Workstation market as a whole no longer exists. Laptops do more than 95% of 2003's workstation customers could ever dream of. Phones have more CPU and GPU power than 99% of people ever actually use. Apple could build a 100+ core CPU, an add in card with hundreds of GPU cores, and give them away, and almost no one would take them up on it. Because there is no market for such a product. Just us few weird nerds who still dream about adding expansion cards to our towers. And maybe a few audio engineers or similar who see a bit of benefit to some specific PCIe interfaces. Believe me, I'm on team workstation. I'm on team, a few PCIe slots never hurt anyone. I'm on team "If someone wants to buy a 4090 just to maybe speed up a final cut export, let them!" But I'm also aware of how few people want or need those things. There are reasons CRTs were better than LCDs, but they still died because those niche reasons were more than outweighed (literally!) by the advantages of flat panels. Time marches on, and the expandable tower is on its last legs. Nothing to do with Apple's engineering or marketing decsions.
Look at this way. Sitting next to me is a 2003 Titanium Powerbook. 1ghz, 1GB Ram, etc. When new, the best tower I could have purchased was the Dual 2.0GHz G5. 20 years ago, the fastest Apple laptop theoretically had about 20% the compute power of the fasted desktop. Both started at $3000. In opting for portabilty, you gave up 80% of the power, and any chance of doing heavy duty computing, for no savings. There was a reason most buyers opted for a tower. They could get a LOT more for their money, or spend a lot more money to meet their needs. Fast forward to today, and even on the PC side, it's hard to justify a desktop over a laptop. Spec out a tower plus decent monitor, and the desktop with the same specs will often cost more unless you build it yourself. At the very pointy end, the desktop can be more powerful, but if you were looking for anything other than a 7950x/13900k + 4090 machine, getting a laptop is the smarter choice in every way. Unless I have a very specific use case that requires internal PCIe cards, or am a weird nerd who likes building and tinkering with my PC (I am, but we're rare), the laptop is the smart choice.
Point being, the tower market lost the majority of its buyers to laptops and phones on the low end, and the rise of cloud computing and commonplace fast networking or internet means it lost most of the high end to Server racks and now off site virtual server providers. The tiny shred of market left, nerds, small audio/video businesses, maybe some home lab guys, etc. Just can't justify the expense of tooling up an M2Max sized CPU only die with 48 cores and 4 way interconnect for 192 cores. It can't justify a chip with a DDR5 controller for the tiny set of people who need more than 192GB RAM but can't be served by a super fast internal NVMe RAID (Seriously, it wouldn't be that hard to build a 32TB, 64GB/s capable RAID in the 2023 MacPro, that's apples to oranges but faster than many speeds of Dual Channel DDR5).
If you came to Apple with 10s of Billions of dollars in guaranteed orders, they'd build that chip in a heartbeat. But in reality, when you and half a dozen other guys are the whole market and only want to pay like $3k for the whole tower, that custome HEDT/workstation ain't happening.
Last edited:
Nope. Not a mistake. Just an assumption, clearly stated. There's a difference between the two. That's how to approach these problems.
Besides, if I were to choose a much larger bucket (say, 500 gallons), that would support my point even more strongly, since then a "drop in the bucket" would be only 1/40,000,000 of Apple's annual revenue, which would be only ~$10,000 annually. In that case, Apple's Mac Pro sales are even more obviously more than a "drop in the bucket".
Of course, you go in the other direction, and make the bucket tiny -- like the bucket on a Tonka Truck excavator. But that doesn't reasonably fit most people's picture of a "bucket" when the expression "a drop in a bucket" is used. So I think I've got it covered.
As far as technology is concerned, workstations, servers, and data centers are the same market. If a single company is unwilling on incapable of engaging in the entire market, it doesn't mean that the market (or any part of it) is dying.
Fun fact: my 2023 MacBook Pro has lower memory latency than my 2020 iMac when working set size is below ~100 MB but higher latency above that. Tighter integration is not a silver bullet that makes the hardware unambiguously better.
Pure conjecture. You could say the same thing about the higher end mini that came much later with M2 Mini Pro.
It's not "pretty clear". It was a rumor based on a rumor and mostly wishful thinking by Apple Silicon fanboys.
And I guess that also explains why Intel, and to a lesser extent AMD, have been cranking up the performance by throwing insane amounts of power at their high-end desktop chips. If you care about power consumption, you're overwhelmingly buying a laptop, or if you're a crazy person building a desktop, well, you probably want 250W TDP and some mad liquid cooling scenario so you can get 10 more FPS.
Believe me, as a fellow 'weird nerd who still dreams about adding expansion cards [and drives!] to [his] towers', this makes me sad, but I think you're right. What it comes down is a couple of things:
1) everything is mature enough that there isn't going to be some new connectivity thingy in a year or two that you'll really want to add with a PCI card, 5.25/3.5" drive bay, etc.
2) with the help of smartphones, Ethernet/wifi has become the universal port. Want storage? A good NAS will outcompete any removable media on price per TB and reliability. Want a printer? It's going to have network ability. Same with scanners.
3) if you somehow find something you can't communicate with over Ethernet, that's what USB is for. Your basic USB 3.0 is fast enough.
4) increasingly few people have any interest in upgrading anything, which means that things like removable RAM are increasingly falling out the wayside
Apple anticipated most of these trends ahead of others, but these trends are now industry-wide. As I said in my earlier post, go to a computer store looking for non-RGB non-gamery parts to assemble a Windows desktop, and... they're basically not there. Laptops outcompete desktops on performance per dollar, whether in Windowsland or Apple-land - I haven't run the numbers, but if you compare, say, an M2 Pro Mac mini + studio display + keyboard/mouse set with an M2 MacBook Pro, I suspect the laptop wins.
And you probably also explained another mystery to me - the death of computer speaker systems. 20 years ago you could get great 2.1/5.1 speaker systems at the computer store with fantastic audio quality per dollar. Today, there might be two models left, but it's all headphones, gaming headsets, bluetooth things, or 2.0 bookshelf speakers. And it's probably the same thing you described - there are fewer desktops out there, many of those desktops are used for gaming with headsets, no one is going to buy a 5.1 speaker set to plug into a laptop, speakers can probably last forever unless you get unlucky, net result, there is no market for a $100-250 5.1 speaker sound set.
The only question I would ask, though - if the plan was always to use the same chip for the high-end Mac studio configuration and the Mac Pro, why not launch an M1 Ultra Mac Pro last year? Is there something unique in the Mac Pro, its software, etc that could have required another year and a bit of R&D? Is there something unique in the M2, other than I suppose increased RAM capacity, that made the M1 Ultra unsuitable for the Mac Pro?
If not, then it does suggest there might have been a plan for a much funkier chip, that chip got abandoned at a point where going to M1 Ultra made no sense, so... here we are.
I think many of us have a feeling that 2019 Mac Pro owners are going to get screwed, e.g. by being left without major macOS updates in a year or two. Why would they have left that machine on the market to screw those buyers for a year if the plan was always a version of the studio's Mx Ultra chip?
most people use laptops, or laptop class hardware for their primary machine in all but a few industries. apple silicon is great for laptops so apple is well positioned there ( though likely competition will start using these "mobile phone" processors in their laptops down the road and eventually hit sufficiently competitive speeds.)
but i agree that this is evidence that apple didn't have anything up their sleeve as far as a typical workstation class machine (with lots of ram and a top end performing and upgradable gpu.)
They pieced something together for the 2023 mac pro but the paradigm for apple hardware will be generally limited to what SoC can do (with SoC type ram modules) all the way up and down.
but i agree that this is evidence that apple didn't have anything up their sleeve as far as a typical workstation class machine (with lots of ram and a top end performing and upgradable gpu.)
They pieced something together for the 2023 mac pro but the paradigm for apple hardware will be generally limited to what SoC can do (with SoC type ram modules) all the way up and down.
That could also be because you went to a computer store. Specialty retail stores have been slowly dying for ~25 years, because it's hard for them to compete against online stores that have lower costs and a wider selection of products. The ones that remain focus on most popular items, on serving less sophisticated customers, and on providing services that benefit from in-person contact.
I don't think that's true in all geographic locations.
Here (Canada) where shipping is expensive unless you are Amazon, Dell, Apple, etc, brick and mortar computer stores remain quite popular (though the mom-and-pop players are gone, replaced by larger outfits with locations in multiple provinces). There isn't really much in the way of online-only stores like you describe other than maybe Newegg; the two leading brick and mortar stores also have online operations, but when you see their shipping cost and time frames, I think many people are likely to just click the "pick up in store for free" option and drop by the store.
From what I had heard years ago from friends in the U.S., the situation there is quite different and other than maybe Microcenter, there weren't really others in a similar market position. Along with cheaper shipping, lower taxes on out of state/online purchases may have also played a part.
In a way it's surprising, because you should be right - GPUs, motherboards, RAM, SSDs, etc should be more easily sold online. It's not something less sophisticated customers want (and those stores are not really trying to serve them), it doesn't benefit from in-person contact (I'm sure 90% of the people walking in have read 5 reviews on enthusiast sites and know EXACTLY what they want), etc. But somehow, in the bigger cities in this country, that's just... not the norm and the enthusiast-centric computer stores with walls of motherboards and GPUs and display cases full of RAM and CPUs remain a thing. And if anything, I would get the impression that they've been making less attempts to serve less sophisticated customers than they did 10-15 years ago, when they at least would build those customers a boring desktop machine or sell them a boring non-gamer laptop. Now it's all about the RGB gaming systems...
I should, just for fun, try to see how much they'd want for shipping a full system's worth of parts - case, motherboard, Windows licence, RAM, CPU, etc. Wouldn't surprise me if it got into the three figures even on the slower options. And so... I think most people are not going to spend that kind of money on shipping when they can just pick up the stuff and get it 3 days sooner.
Consider that there are a ton of studio monitors now that didn't really exist back then, and they're very popular for desktop setups these days, paired with an audio interface. I'd say they've replaced computer speakers for the most part.
Also with so much time spent on voice (discord, zoom, etc), it's not that great to have speakers that can be heard in your mic.
Finally, like you say, they last forever, so it's not very often you'll need a new set. I'm still using a mid-00's-era Sony mini-system and the sound quality is outstanding. I do miss the old Altec Lansing 2.1 and 4.1 sets I had, though... easy to use and sounded quite good, if I recall.
This was bound to happen. Those crazy people who dreamed up an Extreme quad-chip, ignoring all the concurrency problems of synchronising data from four high-performance CPUs, were always going to disappoint themselves and blaming Apple for not meeting their flawed expectations. We got exactly the Mac Pro that was to be expected. And everybody could've predicted well in advance, that the Mac Studio would be the far better solution for the vast majority of pro customers.
And yes, this probably is the last Mac Pro only for those few use cases, which haven't yet switched from expansion cards to thunderbolt peripherals. It's funny how people insist that a large screen iMac is no longer necessary, because the Studio Display + Mac Studio have replaced it at twice the price. But nobody wants to admit that the Mac Studio actually replaced the Mac Pro for half its price.
Welcome to Reality! 😊
Epyc and Threadripper are similar but not the same. Same with all other Silicon makers. Geforce has almost nothing to do with nVidia's main commercial/data center products anymore. Etc.
Workstation is commenly understood to mean "High End Desktop" or "HEDT," parts using a larger CPU socket, usually with more cores, wider memory bus and more PCIe lanes. Often sharing some design features (but rarely being socket or mainboard compatible anymore) with server parts.
Point being, a Data center ML accelerator or EPYC Genoa 96 core CPU are different beasts than even a Quadro (RIP) and threadripper Pro, let alone a 4090 and 16 core Threadripper. HEDT grew out of server parts, but server parts also grew out of standard PC parts. It was common into the 2000s for "Servers" to just be higher end standard towers with different software.
They're not the same market in short. Related, sure, but not the same hardware wise. Also, not at all the same sales and marketing wise.
The HEDT/workstation market is dying. The expandable tower PC is dying. The discrete GPU is dying. Soon computers will be laptops, cellphones, headsets and cloud computing resources you can purchase from places like AWS or Microsoft. Nothing but a few niche machines will exist in the middle. You don't have to like it (I don't really like it) but that doesn't make it not true. One of Apple's strengths over the last 25 years has been a willingness to kill things before they die. They've been ahead of the curve on the path to every consumer device becoming little more than an SoC, but the world is catching up fast.
That is a fun fact! It's probably fair to conclude that Apple's use of on package RAM is mostly for power saving, not absolute peak performance. But they do manage insane bandwith for a CPU as well as insane power savings. Feel like the latency issue is like the ProMotion display, It's 120Hz, but the refresh rate is pretty garbage, so it can't please everyone or every use case. It's important to understand that I'm not trying to be out here defending Apple, just stating my opnion that the move to single chip computers is the next step in the march towards smaller and more efficient packaging that started with Vacuum tubes all those years ago.
Everybody is doing that.
Its just Intel and AMD are scaling from low-end to HPC, whereas Nvidia will either stick with just HPC, or also scale down, to low-end.
Somehow, of course...
the customers clearly exist, but use commodity/standard hardware to accomplish what they need. apple would be fine with just ipads and iphones of course anyway if you want to go down that path.
but there were a certain group of optimists/fans who thought apple should be the best in all classes of machine (i speculate this based on a lot of the powerpc apple marketing of 1990s and 2000s perhaps and the trauma of switching to intel and hope of the return to glory)
i think this shows what SoC is suited for and what it isn’t (good for what most people want)
Last edited:
The Mac Pro got all of 2 minutes 20 seconds in Apple's over 2 hours long WWDC event on Monday. That pretty much says it all about what this Mac Pro is to Apple.
They "Exist" to the tune of maybe a couple million per year globally, or maybe ~150k annual for Mac. With the cost of RnD these days, that means, yes, using hardware largely developed for other use cases. That or exorbitant pricing.
If Apple branches from giant monolithic dies to a tile based approach, there would be a bit more flexibility. If making a MacPro that could accept DIMM memory required only an updated I/O die/memory controller, not an entirely new huge SoC, it might happen. But they'll never go that route for the MacPro, only if it helps make iPhones, iPads, headsets and MacBook Airs cheaper.
The last time Apple tried to compete across the majority of PC categories they almost went bankrupt. Apple 25 years ago implemented a '4 quadrant" product philosophy. In 1997-1999 when they were filling those in, one of the quadrants was "Expandable Tower." It was one of the most popular types of computer, and a single motherboard and case design could serve the ~$1500-$5000 market well enough.
Today, the 4 quadrants of computing are Smartphone, Tablet, Wearable, and Laptop. Desktops at all are uncommon and expandable towers uncommon even among desktops. People who think Apple is unique in abandoning the tower haven't tried to buy a PC lately. The market for any tower, let alone a Workstation tower like the 2006-2012 Mac Pro was is tiny.
Personally, if I was leading the Mac team, I'd kill off the 8 PCIe slot Pro and the 0 PCIe slot Studio and release a beefier Mac Mini with front I/O, upgradeable storage and support for the MxMax SoC. Then a "Pro" that supports everything from base M series to ultra SoCs, has 4 x16 PCIe slots, 3 or 4 front "Minibays" (M.2 or custom PCIe x4) for storage or I/O, and 4 3.5" SATA/U.2 bays. I don't think it'd be feasible to add back DIMM slots on this generation, but I'd fight for fitting DDR5 controllers onto later Pro/Max/Ultra dies to use as "SuperSwap," faster and more long lasting than using SSD swap, but a layer above the unified RAM.
There is a very very small niche market for building a workstation and upgrading components. Frys made a killing in build your own PC/workstation space, they never moved past that and closed the stores. Microcenter is probably the last chain still catering to the building enthusiasts.
I wouldn’t be surprised if MacPro barely makes any money for Apple. With the SOC and unified memory, it’s an odd device.
Right. And internal SATA. And more TB ports. But that means if you use lots of storage, or PCIe cards of any sort, it can quickly become cheaper and faster for your use case. For example, You can easily add 16TB of high end NVMe storage in a super fast RAID for barely $1k to a Pro. Doing that to a Studio would require an expensive box and to use one of the Studio's TB ports all to get 1/8th the potential bandwidth of PCIe 4.0x16. There is no way for any price to build a Studio with an internal 32GB/s NVMe RAID. Or for any sort of internal storage larger than 8TB. With a Pro, I can put 40TB or more of HDD inside (2x20TB, really wish it had 4 SATA bays, but I'm weird like that), or use just 25% of the PCIe slots to add up to 96TB of NVMe storage (use the squid 6x NVMe card and 12x8TB drives).
Point being if you think there's no difference between the two, the Pro isn't something you're in the market for. That's why it's so expensive and they will sell so few. Even most people who have a want/need/use for an M2 Ultra have no use case for the Pro's expansion. You'd have to be someone who truly can't live without more than 8TB of super fast storage. Or who makes a living using a PCIe card that just doesn't work right through a thunderbolt box. Or, as one commenter here described, have a mad setup that requires 8TB ports without daisy chaining. It's a super niche product, but if you need it, you need it
Thinking about it more, some truly crazy person could use every PCIe slot for NVMe storage and do some absolute insanity. 8 cards with 6 8TB drives per card. 384TB of PCIe 4.0 storage. Even making 1/6 of the cards parity becuase you'll have some failures for sure, that's 320TB of storage. Rack mount the thing, use one of the front panel TB ports for data ingest, the other for display out (or just use a VisionPro and look at the Mac for wireless display right?) Then connect 2x10gig ethernet and you've got some sort of mad scientist's Mega server from hell all running MacOS if you want to.
Or, more realistically, rack mount a Mac Pro with 4 or more 8k capture cards to do live 8k multicam editing or broadcast. Not many people doing work like that on a Mac, but it's something only a MacPro could do.
The list is, well, not endless, but pretty broad. Again, if none of these scenarios are you, then buy a Studio.
Last edited: