Workstation was always a term for a very powerful computer that had very narrow purpose. Think of Silicon Graphics as the most famous one but there were others especially in early 90s. Basically every application in visuals industry had a dedicated computer/workstation and they cost a lot. I think one for motion graphics/ video compositing running predecessor of Flame cost around $400,000 in late 80s/early 90s forgot the name. Then there were a lot of 90s game SDK computers that also image sprite manipulation.
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
Workstation was always a term for a very powerful computer that had very narrow purpose. Think of Silicon Graphics as the most famous one but there were others especially in early 90s. Basically every application in visuals industry had a dedicated computer/workstation and they cost a lot. I think one for motion graphics/ video compositing running predecessor of Flame cost around $400,000 in late 80s/early 90s forgot the name. Then there were a lot of 90s game SDK computers that also image sprite manipulation.
This snazzy lab video claims the two main market segments that could use a Mac pro are moving or have mostly moved on to thunder bolt based solutions anyway rather than PCI express cards, so the market for the pro is even smaller than you think. I don't know how true this is. Maybe some real pros can chime in
(Jump to 7 minute mark)
(Jump to 7 minute mark)
apple knows so the $3K markup is to make up for the loss of apple storage upgrades vs PCI-e m.2 sticks.
There are still a few things that can't be done by Thunderbolt because of the limited bandwidth; so I am going to disagree with that statement. 40Mbps isn't fast compared to PCIe or other higher end connections like FibreChannel and such.
Specifically Mentions Multiscreen Capture Cards and HDX Cards for Audio moving from PCIe to TB.
Also mentions that not everything is available.
TB also have video eating up part of that bandwidth.
what bus is the HDMI ports on? is each TB port it's own bus?
Listening to the latest "Upgrade" podcast and an "anonymous Apple Silicon engineer" wrote in to say that there are no plans to revisit the "Extreme" SoC through at least "M7". They claimed that the "M1 Extreme" was intended to be the only product of this category as the market is just not there to support regular development (which is probably why the 2019 Mac Pro never saw the follow-on generation of Intel W Xeons).
Said engineer did note that with "M8", Apple intends to move to Multi-Die Packaging which would allow for fabrication of independent CPU and GPU dies that could then be packaged together to allow for large-scale CPU or GPU core counts.
Said engineer did note that with "M8", Apple intends to move to Multi-Die Packaging which would allow for fabrication of independent CPU and GPU dies that could then be packaged together to allow for large-scale CPU or GPU core counts.
I checked that podcast and it's quite shocking. They admitted that Apple Silicon was never meant for high-end or high performance chip.
Which frankly makes sense as it seems clearly optimized for efficiency and performance with macOS and the mainstream workflows consumer and prosumer Macs are used for.
Indeed. When I first heard about the limitations of Apple Silicon (SoC), I suspected as much. With Apple trying to fit a square peg into a round hole, I'm not sure if Apple will be able to build workstation class computers in the near future.
richmlow
There are still a lot of pro users. PC isn't different from that as majority are consumers.
I honestly don't think they care about that market because even when they were using Intel's kit, the sales figures are just too low.
I mean at least they have the option of putting in the new generation of Ultra SoC each year if they want, but I could see them doing an "every other generation" swap considering...
And the significant majority of them are ably served with the Ultra (be it in the Studio or the Mac Pro).
For the "edge cases" who need Intel or discrete GPUs from specific OEMs or 512GB or more of RAM, they will join those who have already moved on to PC/Linux workstations.
Tell that to all Pro users using Mac Pro. Clearly, you are defending Apple's action. This is why a lot of pro users are disappointed.
99.99% of the disappointment I have seen on this forum is from people who don't use Mac Pros.
Yes I have, because I know a fair number of Pro users using actual 2019 Mac Pros for actual Pro work under macOS with macOS applications and the majority of them have already moved on to MacBook Pros or Mac Studios because it does the work faster and significantly more efficiently. The (relative) few who are going to move to the 2023 Mac Pro are doing so because they have five-figures worth of non-GPU PCIe cards in their 2019 Mac Pros and they either don't want multiple Thunderbolt breakout boxes cluttering up their workspace if they went with a Mac Studio Ultra or they have
one or more PCIe cards that would be bandwidth-choked over a TB4 connection.
The Pro users who use Mac Pros are per Apple's own claims in the very low single digit percentages of macOS users. Apple designed Apple Silicon for the other 95%+ of us macOS users who were tired of having Intel and AMD space heaters in our laptops and desktops that needed fast and loud fans to provide sufficient Air Volume Flow Rate to cool them (and often failing at that, requiring the CPUs and GPUs to throttle themselves) and which burned through even 100Wh batteries in a handful of hours. Now we benefit from cool and fast machines on our desks and in our laps and the latter can go to the nearside of 24 hours on a charge.
One of the big problems that the Mac Pro has is that back in 2009-2012 what would call the golden age of the Mac Pro then there was a LARGE gap between the general Mac ie iMac, Mini, MBP etc and the Mac Pro
I got to Mac in 2008 so started a 2008MBP and a 2009 mini so
Core2 Duo and 8Gb RAM.
2009 Mac Pro = Quad Xeon 16Gb RAM / Dual Quad Xeon 32Gb (shipping)
2010 Mac Pro = Quad Xeon 16Gb RAM / Dual HexaCore Xeon 32Gb (shipping)
Obviously RAM limits went up with larger DIMM Capacity but a large gap between the Mac Pro and the Consumer boxes.
However after that then the iMac Specs started to impede more and more on the Entry Mac Pro market.
iMac 2010 27" = Quad i7 16Gb RAM so starting to impede on lower end ie Entry Mac Pro levels
iMac 2013 27" = Quad i7 32Gb RAM so again moving into more Entry Mac Pro levels
Come 2019/2020 and have
Mac Pro 2019 ENTRY - Eight Core Xeon 32Gb RAM and 580x GPU.
(yes can build higher but point is that there overlap between ENTRY model Mac Pro and Higher End iMac/MBP)
iMac 2020 27" = 10 Core i9 128Gb RAM 5700XT with 16Gb VRAM
A lot of the people that bought Mac Pro's as needed more then the iMac/Mini/MBP offered found could move away from them to the iMac/MBP. I understand a lot of wedding photographers went to the 5K iMac instead whereas previously used Mac Pro. They had bought Mac Pro as offered more power then the iMac/mini etc not that they needed the expandability/upgradeability or the high end grunt of upgraded models. Instead upgraded iMac etc could meet there needs.
I bought a second hand 2.8Quad with 5750 and 12Gb RAM and 1Tb HDD in it in2013 that had been offloaded from company that had written off after 3 years. I would be surprised if bought that in 2010 if they bought 2013 trashcans to replace it. I upgraded it and now have a Studio with a Sonnet TB3 box and Highpoint SSD card and quicker, cooler (temp wise), quiter and much lower electricity usage
I suspect will actually replace the Studio with a Mini with an Mx Pro SoC in it when come to replace it.
PC land even better/worse in that now in Consumer Land can build a Z790 system with 128Gb RAM and a 24 Core i9 13900 CPU. Slap in a GTX4090 or 2 and have a powerhouse that not a Workstation grade machine.
There is a reason why the Mac Pro 2019 moved upwards as had to justify it's existence however had Intel with the Xeon that could spread a lot of the cost around. With ASi then there is no-one else to spread the cost with and not going to be able to ship lower entry level models to shift more as the Studio will cover those lower entry Mac Pro's that sold in the 2009/2010 generation. There has been quite a few YouTubers unpleasantly shocked that they can get there Channel work done on ProSumer hardware and don't need to spank out the top end money for a Mac Pro. (and salty tears when realised how much money they wasted)
If that PodCast is correct then effectively for people that bought the High End or bought low end and upgraded themselves then there isn't a future on Mac for you. Same as people that need CUDA software. Would need GPU Support and Nvidia which cannot see either giving way on and don't see the Multi-Boot improving much either.
Looking at min of 5 years for M7/8 shipping if Apple do a 12 monthly Mx processor life on a HOPE and a PRAYER that Apple bring in no SoC GPU Support and if do will it include Nvidia in Apples Metal Only world etc
I got to Mac in 2008 so started a 2008MBP and a 2009 mini so
Core2 Duo and 8Gb RAM.
2009 Mac Pro = Quad Xeon 16Gb RAM / Dual Quad Xeon 32Gb (shipping)
2010 Mac Pro = Quad Xeon 16Gb RAM / Dual HexaCore Xeon 32Gb (shipping)
Obviously RAM limits went up with larger DIMM Capacity but a large gap between the Mac Pro and the Consumer boxes.
However after that then the iMac Specs started to impede more and more on the Entry Mac Pro market.
iMac 2010 27" = Quad i7 16Gb RAM so starting to impede on lower end ie Entry Mac Pro levels
iMac 2013 27" = Quad i7 32Gb RAM so again moving into more Entry Mac Pro levels
Come 2019/2020 and have
Mac Pro 2019 ENTRY - Eight Core Xeon 32Gb RAM and 580x GPU.
(yes can build higher but point is that there overlap between ENTRY model Mac Pro and Higher End iMac/MBP)
iMac 2020 27" = 10 Core i9 128Gb RAM 5700XT with 16Gb VRAM
A lot of the people that bought Mac Pro's as needed more then the iMac/Mini/MBP offered found could move away from them to the iMac/MBP. I understand a lot of wedding photographers went to the 5K iMac instead whereas previously used Mac Pro. They had bought Mac Pro as offered more power then the iMac/mini etc not that they needed the expandability/upgradeability or the high end grunt of upgraded models. Instead upgraded iMac etc could meet there needs.
I bought a second hand 2.8Quad with 5750 and 12Gb RAM and 1Tb HDD in it in2013 that had been offloaded from company that had written off after 3 years. I would be surprised if bought that in 2010 if they bought 2013 trashcans to replace it. I upgraded it and now have a Studio with a Sonnet TB3 box and Highpoint SSD card and quicker, cooler (temp wise), quiter and much lower electricity usage
I suspect will actually replace the Studio with a Mini with an Mx Pro SoC in it when come to replace it.
PC land even better/worse in that now in Consumer Land can build a Z790 system with 128Gb RAM and a 24 Core i9 13900 CPU. Slap in a GTX4090 or 2 and have a powerhouse that not a Workstation grade machine.
There is a reason why the Mac Pro 2019 moved upwards as had to justify it's existence however had Intel with the Xeon that could spread a lot of the cost around. With ASi then there is no-one else to spread the cost with and not going to be able to ship lower entry level models to shift more as the Studio will cover those lower entry Mac Pro's that sold in the 2009/2010 generation. There has been quite a few YouTubers unpleasantly shocked that they can get there Channel work done on ProSumer hardware and don't need to spank out the top end money for a Mac Pro. (and salty tears when realised how much money they wasted)
If that PodCast is correct then effectively for people that bought the High End or bought low end and upgraded themselves then there isn't a future on Mac for you. Same as people that need CUDA software. Would need GPU Support and Nvidia which cannot see either giving way on and don't see the Multi-Boot improving much either.
Looking at min of 5 years for M7/8 shipping if Apple do a 12 monthly Mx processor life on a HOPE and a PRAYER that Apple bring in no SoC GPU Support and if do will it include Nvidia in Apples Metal Only world etc
There are tons of videos and articles about pro's disappointment.
So what? Mac Pro 2023 killed so many pro markets even for video because of stupid hardware with a lot of limitation. The percentage is lower because Apple keep killing the pro market. Beside, PC isn't different from Mac as majority of users aren't pro at all. That's such a poor excuse.
Gotta get those clicks for the algorithm and ad revenue and complaining about Apple always works for that.
If you are using ProRes, there is no reason you should not be on Apple Silicon considering how fast the hardware encoding and decoding is even on a Max, much less an Ultra. The 2019 Mac Pro had to offer a $2000 accelerator card ("Afterburner") because even the mighty dual Pro Vega II choked on it, in comparison.
And if you are not using ProRes, you probably should not be using a Mac.
Not all Mac users are using FCPX and ProRes. In reality, there are more PP users. MKBHD is a great example. This is why professional FCPX users already had petitions about getting more supports and advertisements for FCPX directly to Apple before. Beside, you still need high GPU performance for After Effect if you need graphics on videos even for FCPX users. Don't say Motion cause it sucks.
Seriously, you don't know anything about the video market.
That claim is a bit odd. First, it is coming from someone who works on the GPU team. Yes, the GPU is going to hooked into the whole chip package organization , but it is also the case that the GPU shouldn't really have to 'care' that much about stuff on the other side of a UltraFusion link.
Did UltraFusion throw the GPU design way , way , way off for the Ultra. From the commentary it doesn't seem like so. Apple already has a multiple die package offering in the Ultra. And they really don't necessarily need 4 (compute) dies. Just something > 2.
The comments the engineer made seems far more applicable to the future fab process divergence when might want to pick a different fab process for CPU dies than for GPU dies ( both have different leverage they place on cache and/or I/O which could 'fork' keeping them on the same process in future set ups. )
It smacks as if Apple won't really commit to chiplet focused design until the laptops 'have to' commit to chiplet design. ( until going to separate some 'computing' type cores from each other... just don't bother. )
There really was no Intel Xeon to follow up the MP 2019 to prior to the transition (or even early in the transition). The Xeon W-6300 series was skipped by Dell/HP/Lenevo also. The W-6400 series also had tons of quirks (went throw over 10 steppings to even get out the door in quirky state. ) You think Apple was going to touch that with a 10-foot pole? Nope. That half-baked , completely untimely products was one of the contributing factors why Apple left Intel.
Without a very long term macOS on Intel instance, there is no rational justification for a 'new' x86 Mac Pro past 2023 ( the end of the transition 2020->2022 ). Even if Intel cleaned up their act, it has little to do with the x86-64 SoC.
What the 'rumor' actually said was that: ' My upstanding the quad required too much effort for too small a market ' .
What is being dismissed here is a quad specifically. Not multiple dies. It is more so the notion of trying to pound a round peg ( a die primarily designed to be monothitic) into a square hole ( scale by fundalmentally baking disaggregation of significant chunks into the core baseline design). Quad monothothic chips is kind of dumb... so no wonder it is 'too expensive' (after applying copious Rube Golderberg gyrations to make it something that it is not).
They already have mulitple die packaging. The issues is far more so is when do the vast majority of the M-series SoC also all get multiple die packaging. Then Apple will care about doing chiplet design. When the laptops are doing it , then they'll let it 'trickle down' to the desktop SoCs.
Apple could scale past 2 before M8.
[ I/O ] skim the I/O off a 'Max' top. (put a ultra fusion connector)
[CPU
GPU ] -- 'the rest' Max (ultra fusion on both ends )
[ GPU
CPU ] -- a pair up current ultra style 'bottom to bottom' .
[ CPU
GPU ] -- another tile pair up 'top to top'
[ I/O ]
Still get an ultra by pulling the middle chiplet. could get a desktop Max by dropping bottom three above (or sticking two I/O on a 'rest' compute chiplet. ( not really a good reason why the "max" and "ultra" Mac Studio should have to switch socket types. 6 TB ports for both would be reasonable and consistent. Both the Studio and Mac Pro are hobbled a bit with the forced reuse of the laptop die. )
No Quad laptop optimized monolithic die involved at all. Nor any deep need to decouple the GPU and CPU cores onto different fab processes for the dies. [ there could be some upside for letting the I/O focused ones stay on N4 or affordable N3 variant, but they all could be the same process. ] What Apple mainly needs to do is disaggregate the I/O from the compute; not the compute from the other specialized compute. Take the laptop 'Max' die and pull it apart so have decent chunks to use a chiplets. [ Aa oppose to brute force pounding the laptop die into perpetrating that it is a chiplet. ]
.
Podcast doesn't really say that at all. Apple pulled the plug on the Extreme because they thought the folks left in the target market were too small in number to match the additional cost of doing a much bigger SoC package. If they were goin to pour $500M into the effort and only get $500M back ( or less) then why bother. If the payoff had been spend $500M and get $2000M the 'bang for the buck" would have payoff would have greenlit the effort.
[ Similarly if $500M into the Rn series get them $1000 back and but only $600M back for a Mn Extreme ... guess which one loses access to a finite amount of resources ? ]
However, the standard high end performance configuration that Apple sold 5-10 years ago, Apple siliicon is absolutely out to do better than that. Folks ready to retire hardware they have been squatting on for a long, long time, these new systems are candidates. Apple is picking segments of high performance where there is a substantial pay off fro the effort they put in. It is not a sell everything to everybody effort.
The title on the Upgrade blog for the one before this one is : "The Least Important Mac"
There is partially where there is a disconnect. Some folks think the Mac Pro is the "Most important Mac". Apple probably does not. That "Extreme" SoC has to justify itself. It isn't going to get done "just because". Or Apple should just throw money at it like a drunken sailor on shore leave in a strip club.
What makes you think Apple wouldn't incorporate ECC if they went to ≈1 TB? That's what NVIDIA did with the (up to) 960 GB LPDDR5x RAM in its Grace Hopper Superchip.
Indeed, Apple recently published a patent about the use of ECC in LPDDR RAM. While the patent makes a passing mention of on-die ECC, most of it focuses on their claims of improvements to link ECC:
[ https://patentscope.wipo.int/search/en/detail.jsf?docId=US403904970&_cid=P12-LLBQX0-80508-2 ]
Expanding on this:
AIUI, there are two broad types of ECC: (1) traditional "transmission" ECC (which corrects transmission errors between the RAM and the processor; when RAM is labelled "ECC", they're referring to this); and (2) on-die ECC, a new type of ECC that was introduced with DDR5 to address, as the name indicates, on-die memory errors that DDR5's higher memory density makes more prevalent.
Transmission ECC comes in two flavors: side-band (typically used in DDR RAM) and inline (used in LPDDR RAM). The variant of inline ECC that JEDEC introduced for LPDDR5 is called link ECC (I don't know if inline ECC was available to earlier generations of LPDDR).
But that leaves me with these specific questions:
1) Are there differences in the robustness of error correction between the JEDEC DDR5 standard's "transmission" ECC, and the JEDEC LPDDR5 standard's link ECC?
2) How exactly does Apple's patent claim to improve upon link ECC? It would be nice if it came with a clear abstract that said: "This is how the patent improves upon relevant existing technologies and prior art: ..." It does say that existing ECC used in servers consumes too much power for many applications, but doesn't summarize how this is an improvement over the relevant existing tech, which is link ECC (and which, as it's designed for LPDDR RAM, likely consumes much less power).
And these more general ones:
1) On-die ECC helps with RAM manufacturing. But is it also used to address memory errors (e.g., on-die bit flips) that occur during use?
2) Does the JEDEC standard for LPDDR5 specify on-die ECC?
3) Is there currently any LPDDR5 link ECC RAM on the market? I can't find any. Or is it the case that *all* LPDDR5 RAM comes with link ECC, but because its error correction is not as robust as that of standard DDR ECC, they decided not to label it as "ECC"? Note I'm referring to LPDDR5, not the LPDDR5x ECC RAM that NVIDIA uses.
For more info.:
ECC in DDR Memories | DesignWare IP | Synopsys
Technical Bulletin: Error Correction Code (ECC) in DDR Memories
LPDDR5 key features | DesignWare IP | Synopsys
Technical Bulletin: Key Features Designers Should Know About LPDDR5
Sheesh, who pissed in your cereal? Really, you should try to be environmentally responsible and not waste question marks. Five is excessive, some might say profligate.
This seems like a complete non-sequitur to my post. Nowhere, in suggesting TB 5, did I rule out DP 2.1 Indeed, when TB5 is released, it should accomodate DP 2.1.
Note also that, while the TB5 standard hasn't been released, it appears it will be able to dynamically rebalance from 80 Gbps each way to 120 Gbps out/40 Gbps in. If Apple's next XDR is, as some rumors say, a 10-bit 7k 120 Hz display, you could run it uncompressed (no DSC) using TB5, as that requires 111 Gbps. DP 2.1, by contrast, is limited to 80 Gbps each way.
Last edited:
Please, gaming rigs are workstation nowadays. They can fit up to 64 CPU cores now. The main advantage of choosing "workstations" was that they provided more computing power than desktop PC's, but that is no longer true at all. ECC doesn't matter for work. People do their work on regular RAM just fine. It's intended for servers that are running 24/7 without any downtime.
And for high-performance computing where you could make a case for ECC RAM, this is done in the cloud as "workstations" are too weak for this.