I blame apple more than Intel. Despite horrendous delays for each release they picked the worst point humanly possible in which to release a Mac Pro. The 7,1 is the latest example. No pci4. Later processors way better. It’s like try to pick the worst point possible.
The W-3300 was 'way better' than the W-3200 ? Really? It was priced better, but in terms of timely technical abilities that Apple was primarily going to use it was better how? delivering in very low volumes 2021 instead of high volumes in 2019 is better how for a product where most of the targeted base had been waiting since 2016 or so for a slotion? Making them wait 5 years instead of 3 years would make things much better? Running 50-80W hotter helps Apple how (since at max recommended electrical code common household wall socket power)? All that does is reduce the GPU power budget (which is also spinning upwards). Slower base clock speed ( Ice Lake has a clock speed regression that Intel only partially offset with better IPC in some areas) helps on a single user workstation how? Dell/HP/Lenovo also skipped the W-3300, it wasn't just an 'Apple' thing. W-3300 has some corner cases where it was better (mainly some narrow server workloads), but holistically it was generally a 'miss' in the single user workstation market. ( Like 'Rocket Lake', Intel will probably drop that product sooner than they usually do. Stopgap products that didn't get much traction. )
iMac Pro started at the beginning of a socket lifecycle. But W-2100 to W-2200 was mostly just a price cut. It was primarily a recycled die with some tweaks. There was no W-3100. The 3x00 basically born at x200 cycle because Intel was 'stuck'. Even if Apple had done a more horizontal shift from Mac Pro 2012 to 'MP 2017' with around 4 slots and lower price points.... it too would have been stuck on the W-2x00 series. That was a contributing factor as to why the iMac Pro didn't get "one last update" at 2H-2020 - 1H-2021 transition.
Perhaps can 'blame Apple' for not switch 'horses' and going with AMD. But if they are getting out of that particular 'horse racing' business it isn't really a 'picked worst point in time' issue. Picking a time when Intel was way off balance and not executing is exactly the right time for Apple. Apple was eventually going to drop x86.
And if they release a Mac Pro now, it will be the same thing. No 3nm. No extreme chip. No ram expansion ability. Lame pci lane ability. No ecc.No ability to support 3rd party graphics (all above if rumors are to be believe…I hope not).
No TSMC N3 isn't for sure. Apple doesn't have to release the M-series in 'smallest to biggest' order on every single cycle. No extreme is a set back but not releasing anything is demonstrably worse.
From back in 2021.
" ... Most notably, The Information says Apple and foundry partner TSMC plan to produce 3-nanometer chips for Macs as soon as 2023. These could feature as many as four dies, with up to 40 CPU cores in total per chip. The three versions of the third-generation chip are reportedly codenamed ‘Ibiza’, ‘Lobos’ and ‘Palma’. ..."
Apple is taking the PC world by storm with its first debut of Apple Silicon chips inside of Macs, taking...
9to5mac.com
the "Ultra and up" packages may be on an only odd-number release schedule. M1 , M3 , M5 , etc. Pretty doubtful that the Extreme chip would be iterated at the same pace at the 'plain' Mn packages will be. The economics are completely different. If there is some "Mac Pro only" chip package it isn't going to evolve at the same pace as the rest of the line up. Apple's track record for Mac Pro update frequency over the last decade is pretty clear. Every 12-15 months highly likely isn't coming.
So if Apple 'nuke' a M1 iteration then a M3 would be the next-man-up.
Lame PCI ability isn't for sure either. There are 3-4 highly straightforward ways of adding some without disrupting the baseline floorplan of the silicon they have. ( 1. some chiplets swap out TB controllers for PCI-e. The TB controllers have PCI-e v3 controllers in them anyway so shifting to v4 while dropping some other stuff wouldn't be a 'moonshot' product. 2. put PCI-e controller die between two chiplets that don't have loads of PCI-e lane provisioning (UltraFusion works fine. 'adding' to a chiplet isn't a moonshot project either. 3. slice the I/O section off the top baseline design and optionally add a more PCI-e lane provisioning, relatively small chiplet were appropriate. etc. ). Apple doesn't have to use exactly the same design as the M1 Ultra to deliver a M2 Ultra. So far the M2 Max doesn't have a UltraFusion connector in pictures. So it doesn't 'have to be' the solution for the M2 Ultra.
I would not be surprised if the crap slots they include are only pcie4 despite 5 being out.
If the PCI-e lane count is delivered by a I/O augment chiplet doing PCI-e v5 wouldn't be out of the question. It isn't on the main die so the PCI-e v4 there isn't a hard constraint. If Apple is trying to keep the chiplet die (or subsection die ) relatively small though then two x16 PCI-e v4 == four x16 PCI-e v3. It isn't a backslide.
Pretty good chance if Apple went to PCI-e v5 that all you'd get would be one x16 PCI-e v5 == four x16 PCi-e v3 on an even smaller die are allocation.
Making the PCI-e lane augment chiplet cheaper to make makes it more likely Apple will do it. Apple would likely throw money at a fancy discrete PCI-e switch to 'fan out' from the SoC. It isn't a 'cheaper for end users' move. End users can slice-and-dice the bandwidth to whatever slots they think are appropriate (like two x16 PCi-e v3 bandwidth of the MP 2019).
PCI-e v5 on a workstation without CXL is going down the wrong track. If Apple is likely missing out on anything it is CXL , not the baseline v5. If Apple can't loop in CXL when they go to v5 then it is missed opportunity. If Apple needs to wait the CXL 2+ to jump in that is probably worth it. Primarily Apple needs to just get something out the door with greater than 'four x1 PCI-e v4' worth of backhaul. Baby steps before start running.
All the above, I bet, would be way easier to deal with via M3. So you’ve delayed this long might as wel, delay till you can get it it “right’.
Delaying longer isn't likely going to 'buy' RAM DIMM slots. TSMC N3 isn't going to be some ginormous panacea either. The notion that Apple is going to 'start over from scratch' on N3 and bring in all the missing trade-offs is likely self delusion. highly unified and homogenous memory is likely going to be there on N3 also.
Apple's corporate policy of not commenting on future product has a built in premise that they are going to ship on a regular basis. They let the deliver of new product do the 'talking'. Squatting more than 3 years on something doesn't help them the longer they deviate from that presumption. If Apple got onto an every 2-2.5 year schedule things would 'work' much better. 3 year maybe if the plain M-series settles on 1.5 for itself.
The 'waiting for perfect' appears to be part of the problem Apple has with the celluar modem offering. Each year they wait Qualcomm gets better and puts a new even snazzier line in the sand for the next year. Keep feature chasing them on every newest shiny and will never ship. (Qualcomm when to N3 so now have to wait for N2 to 'trump card' them. etc. etc. etc. )
Which is exactly why I believe they’ll do the wrong thing and release a half ass pile of s*** machine and then complain how the market doesn’t want a “pro” machine anymore to justify their pulling the plug forever.
Yea, apple has taught me to be that cynical.
Anyway Mac Pro became my main workhorse about mid-snow leopard, and I had every single model of MP, and of course 5,1 beeing the favourite one.
