Why is Apple not at least using PCIe4 in the new Mac Pro since what I've been reading is :
1. The Mac Pro was probably originally designed and planned to finish in late 2018. There were no roadmaps of PCI-e v4 processors in late 2018. AMD or Intel. IBM Power was still going to be about the only player at that point when looking at projections back in late 2016 - early 2017.
That the Mac Pro is sliding into 2019. It would have slide even more if they had to wait for 2019 targeted stuff engineering samples to get started. For a project that was 3-6 years late (depending upon perspective) that would have been an extremely dubious move. Complicated projects typically work much better when first get something solidly working done and then more on to incrementally more capable system. The Mac Pro had been completely dead-in-the-water for years.
If Apple intends to put steady work into future Mac Pro then they'll merge in PCI-e bump on the next iteration. The primary issue was there was zero movement for years.
2. PCIv4 is faster, but also more problematical. The lengths you can run PCIev4 lanes is shorter ( without having to re-transmit the signal. Which is more complicated board development.... which again more complicated typically leads to longer development time. )
The new Mac Pro places the biggest bandwidth sockets the furtherest from the CPU. That is actually backwards from the normative that PCIe v4 constraints lean toward. It isn't a "show stopper" but it would have been yet another issue to deal with.
3. Lots of folks were hoping that PCIe v5 would "clean up" some of the issues that v4 introduced. So that is why most folks waiting until v5 was closer to wrapping up before deploying v4. In the 2016-17 context v5 was still in early stages and could have possibly ended up getting stuck in committee ( few wanted that to happen but sometimes 'stuff happens' ) . Waiting on v5 was a significant risk increase.
4. Intel was the 'safer bet' for the CPU. Yes their execution was problematical in 2016-2017 and the Meltdown, Spectre , and MDS security defect fest (in 2017-2018) is a problem for a company whose major marketing focus is security/privacy . But AMD had also shot themselves in the foot ( e.g., the GPU evolution sequence and the Mac Pro 2013 design). Intel didn't plan to do an aggressive rollout of PCIe v4 and when Apple picked them for the new Mac Pro , the Mac Pro was locked into that.
Apple could have picked AMD and their PCIe v4 implementations, but it would have more so a "bet the farm" move. It would have guarantee that would slide into 2019 ( so even bigger group of impatient, pissed of user base). If AMD stumbled then could have ended up with components that were not competitive ( that's death at the 100+% increased entry price point they decided to shoot for).
PCIev4 "everywhere" was going to put more thermal pressure on more components ( e.g., almost all of these AMD X570 boards have active coolers on the PCH. that isn't a "show stopper" but it is yet another placement and design problem to work around. )
[doublepost=1560005325][/doublepost]
PCIe 4.0 will probably be scrapped.
PCIe v4 isn't going to get scrapped any more than v3 , v2 or v1 were.
PCIe v4 will probably stick around for a while because the main I/O hub chips (PCH) probably aren't going to go to v5. Even going to v4 is proving to be a stretch for the X570 chipsets (now needing active cooling. That will unwind a bit at the PCH more to better fab process and updated designs but v5 would just put that pressure back).
It has so serious attenuation problems that today's AMD solution only works with one PCIe slot and needs to be exactly side by side with the CPU socket.
The SerDes updates that v5 brings can be run at v4 speeds too. How to get to having something that actually needs v5 speeds in the vast commercial space is a problem. There are v4 PCIe M.2 drives but they all have to throw on more NAND chips ( 8 ) just to crank up the aggregate bandwidth from the NAND chips to fill the increased v4 headroom. Gaming doesn't saturate x16 v3 GPUs. x16 v4 is even less. There are some high end cases that have wins > 100GbE and shared/remote memory message/content passing. .... but much of that is smaller markets ( which is why hardly anybody followed IBM in implementing v4. Power9 was introduced back in 2017. It has been two years since first implementation and about three years since IBM rolled out beta machines to customers to test. )
[doublepost=1560005644][/doublepost]
...and the number of PCIe 4.0 devices likely to ship in quantity this year can be counted on the fingers of one hand (and the thumb is not a finger).
By different general product type or by number? By number all the 2019 AMD GPUs will have it. That is why AMD aggressively pushed it out the door on their CPUs to create bigger perceived synergy. (most of those will probably go into PCI-e v3 systems though). AMD is going a decent sales job of "we have it and they don't" .
If dive into the "don't care how much it costs" Power9 cards can find a few there that have been shipping for a while.
"by product type (in any price category)" . I wouldn't bet on that (less than 6). By late Fall there will probably be substantially more flow.
