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.
MP 7,1 Where are the lucky ones?
- Thread starter chfilm
- Start date
- Sort by reaction score
It is indeed workstation class. ECC memory support, up-to 32 Cores (64 Cores next year, same socket, no motherboard swap needed). 256GB of RAM maximum (which may not be enough for some people). PCIe 4.0 already today.
Intel doesn't have something comparable because they are stuck on 14nm. These new chips from AMD use TSMC's 7nm process which is comparable to Intels 10nm (which they've yet to get working for high performance chips due to high defect rates).
Essentially since 2016 Intel has been struggling where manufacturing of their chips is concerned. Their designs are still solid but they have no way to manufacture them. This is why they've released chips on what they call 14nm, 14nm+, 14nm++ etc - Refinements of 14nm due to 10nm not being viable for high performance chips yet.
The architecture they have been pushing this whole time is called Skylake. The newer chips such as Kaby Lake, Coffee Lake, Canon Lake, Ice Lake etc are all basically identical. Skylake with some minor tweaks.
Chips such as the 7980XE, 9980XE, 10980XE. All released 1 year apart from each other perform identical because it's the same chip sold 3 times. Literally.
This is what people outside of the tech industry are not that aware of as it's like being able to know the difference between car engines, it's not something people who don't take a keen interest in that stuff will be aware of. The short end of this is simple, Intel is in trouble when it comes to being competitive with AMD right now and for the foreseeable future.
The next socketable XEON from Intel that is bigger than 28 cores will be a 38 Core Ice Lake chip coming at the end of 2020 (November-December time frame) if it doesn't slip further. By then AMD will have already launched their 64 Core Threadripper chips.
EDIT:// There is also some things I left out for brevity such as AMD moving to a chiplet design while Intel has stuck with a monolithic design. This is also a reason Intel can't release such high core count processors as the more transistors in a single die there is the higher the defect rate. AMD mitigates this by creating a chip from multiple smaller dies instead of one giant one like Intel currently is doing. It will take several years before Intel can make a chiplet design of their own due to chip design lead times.
Very interesting, thank you for taking the time to explain it. Now the question is, what is Apple going to do about this when their sole provider of chips is lagging and doesn't seem to be moving forward fast enough? I doubt Apple has any server quality alternative right now and probably won't have for a while so where does this go from here? Any guesses?
Well they have three options.
1. Stick with Intel and simply sell a slower system than all the Windows systems available on the market from HP, Dell and boutique makers like Pugent Systems.
2. Switch to AMD and use EPYC or Threadripper parts (this is AMD's top of the line processors for Servers and Workstations).
3. Switch to using a custom ARM chip they develop internally. Clearly they've cleaned up with their mobile ARM parts. There is certainly the potential for them to surprise us all with a secret high end chip. TSMC is able to fabricate huge chips, even ones larger than Intels biggest dies such as the 21 Billion transistor filled NVIDIA GV100 Volta. Apple currently uses TSMC for their mobile chips.
So those are the options. I personally think Apple will stick with Intel for one more refresh of the Mac Pro which I think will be mid to late 2021. After that I think they'll go custom with an ARM design. But I'm just guessing here.
They could certainly go AMD but it would mean they would have to burn their relationship with Intel and I don't think they want to do that. Intel has given Apple many chips first before other OEM's sometimes only announcing new chips even exist the same day as Apple unveils Macs containing them. Perhaps Apple values that partnership more than having the fastest most capable chips for a few years time.
But Apple does also work very closely with AMD. The D300, D500 and D700's used in the 2013 Mac Pro were re-branded AMD gaming cards. Usually the Pro cards from AMD contain ECC memory. This wasn't done on the 2013 Mac Pro cards and those D300, D500 and D700's were not available on the market to any retailer or OEM other than Apple.
Similarly with the new Mac Pro Apple has gotten a RX 580 (Gamer card) re-branded as a Pro card and AMD seems happy to do that. AMD also works closely with Sony and Microsoft on their custom System on Chip (SoC) designs used in the PS4/XBOX and the upcoming PS5 and XBOX SX. There is certainly potential for Apple to get semi-custom silicon from AMD if they felt it was warranted.
Predicting what Apple will do is the source of an entire industries existence so what I say here is certainly just possibilities, I can't judge which way they'll go, just what I personally think is most likely to happen.
Not literally.
https://en.wikichip.org/wiki/intel/microarchitectures/cascade_lake#New_Technologies
DL Boost No here 9980XE
DL Boost Yes here 10980XE
The changes are small and confined to a narrow subset of instructions, but there are changes.
( also a narrow set of Meltdown and Spetre band-aids and fixes in the chip. )
I was already aware of those things when I said what I did, I just consider them extremely insignificant. Which is why I said "Skylake with some minor tweaks.".
My use of the phase they sold the same chip three times literally wasn't accurate. But for all intense and purposes it might as-well be. If anything the newer versions of this architecture perform worse due to those mitigations. And while they added improvements to some math functions for machine learning it's really questionable if that constitutes a separate distinction for these chips.
It's like years ago they released E5 XEON's which didn't have VT-d then they released revised chips which had it. Same exact model number. I wouldn't be surprised if the older chips (7980XE) also featured these more efficient machine learning operations but had them disabled due to errors. It just so happens they gave them new names.
I also think the extra PCIe lanes these newer chips have is present on the older chips but disabled. I mean I'm splitting hairs here but fact remains they released 3 series of chips that perform identically in 99.9% of tasks when core counts are the same. It's the same architecture at the end of the day with some tweaks each time.
Last edited:
I received my Mac Pro Friday. I ordered December 17th, so it arrived super early, I'm guessing because I live in Texas? It's going under the Christmas tree though, so I'm impatiently waiting to start it up! I did briefly, and gently, open up the box just to make sure it was really there. 
For those curious, it's a 12-core, 48GB RAM, 2TB SSD, Radeon Pro 580X GPU.
For those curious, it's a 12-core, 48GB RAM, 2TB SSD, Radeon Pro 580X GPU.
You were literally wrong
Wow what a credible argument to refute anything I said!
Intel Core i9-10980XE Cascade Lake-X CPU Benchmarks Hit Geekbench
The Intel Core i9-10980XE is one of two Cascade Lake-X CPUs that haven't been released yet but have seemingly been tested on Geekbench.
Cascade Lake - Microarchitectures - Intel - WikiChip
Cascade Lake (CSL/CLX) is Intel's successor to Skylake, a 14 nm microarchitecture for enthusiasts and servers. Cascade Lake is the 'Optimization' phase as part of Intel's PAO model.
Hmmm sure sounds like the exact same architecture to me! - Same I/O, same memory controller, same pin out, same socket, same cache sizes, same core counts.
So what exactly did change?
Gee, even Wikichip says it's largely identical to Skylake.. what a shock.
Now you see Cascade Lake was actually intended to be much different to Skylake. But Intel couldn't get 10nm working so instead of being the real Cascade Lake what we've actually got is Skylake v1.2.
Coffee Lake was v1.1, Cascade Lake is v1.2 and Icelake will be v1.3 - Each with tweaks and improvements to surrounding instructions and with more mitigations but the same IPC (instructions per cock). The only increase in performance comes from increased clock speed.
If you take a Skylake, Coffee Lake, Cascade Lake CPU and clock them identically with the same core counts and cache sizes with the same frequency memory guess what, they perform 1:1 identical.
If you look at Zen however, Zen 1 vs Zen 2 there is around a 10 to 15% IPC increase due to the architectures being significantly different. That is real innovation.
Intel did sell us the same CPU three times for three years in a row. The 7980 XE is the 9980 XE which is the 10980 XE. Just look at some reviews of the 10980 XE from reputable outlets and they will discuss at length how this is just the same CPU rehashed again and again.
Intel aren't the only ones, look at the RX 480 graphics card from AMD which became the RX 580 which became the RX 590. Companies do it all the time when they don't have something new to peddle they have to rebrand and sure sometimes they include minor tweaks like putting lipstick on a pig but we know the truth.
I'm sort of one of the lucky ones. My Mac Pro arrived today, but my XDR display doesn't ship until mid-to-late February. I have an old Apple Thunderbolt display and realized that I can't hook it up to my Mac Pro without a Thunderbolt 2 to Thunderbolt 3 adaptor, which to my dismay costs $50. I called Apple and said, "I bought an $8,000 computer, and the $6,000 monitor for some reason isn't shipping for almost two months. Do I really need to buy a $50 adaptor to temporarily use my old monitor? He is shipping me an adaptor. Ultimately, $50 bucks is a drop in the water when you're talking $14,000, but at the same time it's the principle of it. So, will report back once I can actually use my machine.I thought first users were receiving their units yesterday... where are they? Report please
Wow what a credible argument to refute anything I said!
Intel Core i9-10980XE Cascade Lake-X CPU Benchmarks Hit Geekbench
The Intel Core i9-10980XE is one of two Cascade Lake-X CPUs that haven't been released yet but have seemingly been tested on Geekbench.www.tomshardware.com
Cascade Lake - Microarchitectures - Intel - WikiChip
Cascade Lake (CSL/CLX) is Intel's successor to Skylake, a 14 nm microarchitecture for enthusiasts and servers. Cascade Lake is the 'Optimization' phase as part of Intel's PAO model.en.wikichip.org
Hmmm sure sounds like the exact same architecture to me! - Same I/O, same memory controller, same pin out, same socket, same cache sizes, same core counts.
So what exactly did change?
Gee, even Wikichip says it's largely identical to Skylake.. what a shock.
Now you see Cascade Lake was actually intended to be much different to Skylake. But Intel couldn't get 10nm working so instead of being the real Cascade Lake what we've actually got is Skylake v1.2.
Coffee Lake was v1.1, Cascade Lake is v1.2 and Icelake will be v1.3 - Each with tweaks and improvements to surrounding instructions and with more mitigations but the same IPC (instructions per cock). The only increase in performance comes from increased clock speed.
If you take a Skylake, Coffee Lake, Cascade Lake CPU and clock them identically with the same core counts and cache sizes with the same frequency memory guess what, they perform 1:1 identical.
If you look at Zen however, Zen 1 vs Zen 2 there is around a 10 to 15% IPC increase due to the architectures being significantly different. That is real innovation.
Intel did sell us the same CPU three times for three years in a row. The 7980 XE is the 9980 XE which is the 10980 XE. Just look at some reviews of the 10980 XE from reputable outlets and they will discuss at length how this is just the same CPU rehashed again and again.
Intel aren't the only ones, look at the RX 480 graphics card from AMD which became the RX 580 which became the RX 590. Companies do it all the time when they don't have something new to peddle they have to rebrand and sure sometimes they include minor tweaks like putting lipstick on a pig but we know the truth.
didn't see the emoji at the end? I was responding to your inaccurate use of 'literally'. [lightenupfrancis.gif]
I went ahead and went for the 16 core MP to hit what I see as the "sweet spot" in the current BTO options. Radeon Pro Vega II, 1TB SSD and 32 GB RAM from Apple, then added another 120GB aftermarket and a 2TB NVMe on a PCI card. I agree the prosumer is being left behind with this workstation pricing, but Im willing to pay for it to ensure expendability and future upgrade capabilities.
The way I see it, this makes the machine much more of a viable investment over time. I managed to hold onto my 4,1 since 2009 with iterative updates to RAM, drives, GPU, processors, etc to keep it usable. An iMac is never going to allow that, and will be obsolete much faster as a result.
The way I see it, this makes the machine much more of a viable investment over time. I managed to hold onto my 4,1 since 2009 with iterative updates to RAM, drives, GPU, processors, etc to keep it usable. An iMac is never going to allow that, and will be obsolete much faster as a result.
Who gets to define 'workstation class'? Nobody does. Some people have a $2000 Linux home build they use for modelling and rendering in Blender. It's a workstation to them. You can buy a $50000 Mac Pro and it will perform worse than a much cheaper Linux box that Blender supports much better. That's one example of many.
To get into this 'workstation class' mindset is detrimental. That's what brought SGI down. They thought they could keep using this word 'workstation' to charge thousands of dollars but they got outperformed by PCs that cost 10 times less.
These had IPC, minor feature set changes and clock speed improvements.
Have had my 16 core for about a week. Just finishing getting all the odd cabling, plus external TB3 Drive enclosure for 4 drives. Also going over all the possible software snafu's that can occur.
Hate the fact that Catalina is both a 'young' system and that it can't handle 32bit apps (very irritating as in production you can have older apps that are no longer supported but do amazing things). An interesting issue is whether an older 64 bit app could be using a 32 bit installer.
Also looking at ways of loading an older system (Mojave or even better, High Sierra). Have been reading about bypassing the T2 chip - do not need that annoying chip. I don't encrypt my drives, nor do I care about facetime, secure boot or Touch ID on a Mac Pro. And I don't care if apple co-signs an app. Wondering if I can modify or swap kext files or plists to allow booting (I've done it before). Might see what the Hackintosh lunatics have come up with as a bypass.
So in a production environment there are many complicated things that have to work, and it can take a bit of time. The machine isn't used as a glorified web browser!
Hate the fact that Catalina is both a 'young' system and that it can't handle 32bit apps (very irritating as in production you can have older apps that are no longer supported but do amazing things). An interesting issue is whether an older 64 bit app could be using a 32 bit installer.
Also looking at ways of loading an older system (Mojave or even better, High Sierra). Have been reading about bypassing the T2 chip - do not need that annoying chip. I don't encrypt my drives, nor do I care about facetime, secure boot or Touch ID on a Mac Pro. And I don't care if apple co-signs an app. Wondering if I can modify or swap kext files or plists to allow booting (I've done it before). Might see what the Hackintosh lunatics have come up with as a bypass.
So in a production environment there are many complicated things that have to work, and it can take a bit of time. The machine isn't used as a glorified web browser!
AMD literally defines it as that. They even released a video of how Threadripper was being used by the studio making the new Terminator film where they used the word workstation throughout.
I mean you can pick bones all you want but these are workstation chips just the same as a truck is a truck. Who gets to define it? anyone with common sense does.
If anything they had IPC degradations due to spectre and meltdown mitigations that don't even work due to new attacks. The hardware fixes actually make the attacks easier to perform than before. They had clock speed improvements in that Intel set the multipliers higher.
I own a 7900X, it runs at 4.7GHz in my system all day long even though Intel ships it with a base clock of like 3.5GHz, them turning the clock speed up on the 9 and 10 editions is nothing special at all, really.
It's just like when they shipped Pentium 4's at 4GHz to try and stave off AMD in the old days, everyone remembers the Prescott chips you could cook food on.
Last edited:
AMD has no more right to define what a workstation is than the guy who builds his own college dorm Linux PC for doing 3D work with Blender. Cmon, you know that.
True for the 9x00, there was a slight IPC decrease in some tests, but the 10 series has higher clock speeds and some reviewers have hit 5GHz on the 12 core variant.
Well you can think that, I disagree entirely. The idea of what constitutes workstation class hardware is very well understood and Threadripper processors are definitely in that category as far as I'm concerned.
Higher clock speeds are not IPC. They are two different concepts. Ultimately that is what I'm talking about, IPC. The clock speed thing, slight improvement if you're willing to suffer the 400+ Watts power draw.
At the end of the day, these chips are lousy, no one should be buying the X299 chips anymore, Threadripper is faster and has more cores and in some cases for less money too.
I should be clear though the chips we're discussing (for others reading) are not the same chips used in the Mac Pro. It uses a different socket and goes up-to 28 Cores, the chips we're discussing go only up-to 18 cores.
Feel free to tell everyone who builds their own workstation that their computer isn't a workstation and enjoy the flame war. If you don't know how SGI's 'workstation' business died then this is all news to you.
I didn't say they were, but you're the guy who thinks someone has a right to define what a workstation is so you're a lost cause.
Here is when SGI made a desperate attempt to save their workstation business by switching from MIPS to Intel Pentium III.
SGI Visual Workstation - Wikipedia
Workstation. Consumer CPU just fine. But it was too late for them. They first got outperformed and then they got outpriced.
We can go further with an Apple example. Was the G4 a workstation CPU when it was also in iMacs and iBooks? Was the G5 a workstation CPU when it was also in iMacs?
You see....a workstation is whatever you want it to be, as long as you are working on it.
Last edited by a moderator:
I'm aware of SGI and their demise. I dunno why you keep bringing that up for it's not really relevent to this discussion.
Workstation class hardware: ECC Memory support, high core counts, today that means beyond 8 cores, more memory channels than mainstream (4, 6 or 8), 10Gb networking on the motherboard, hardware RAID support (1/5/6), more PCIe lanes (today that means more than 16 channels from the CPU directly).
Threadripper means all of these things ^
X299 from Intel means all of these things ^
LGA3647 from Intel means all these things ^
AM4 from AMD only has some of these things. Z390 from Intel has none of these things. In my opinion you need atleast 1/3rd to 1/2 half of these features to be considered workstation territory for a CPU and I really think ECC is a must.
I build workstations, usually one a year for the past 15 years.
Last edited by a moderator: