MP 7,1 Mac Pro Apple Silicon is going to eat our lunch

[Melbourne Park]

I get that, but I wanted to know if I can go back and forth with the pci drives. This is not my main drives. Just samples.

Black Friday is not tomorrow

Sorry, Black Friday has been creeping into Australia, but its not got the same impact here as in the USA.

By the way, when the 7,1 came out there was concern that Apple had built in protection about not only the main drive, but also for the CPU. But Apple's protection leaves the CPU alone. So you can spend $US650 and put in a Intel Xeon W-3245M QS QRSM 3.2GHz 16-Core LGA3647 CPU which has 16 cores. More than double that cost for 24 cores. $US1,600-$1800? There's more choices too. So I'd be feeling great about $4k for a 7,1. If it's its not as quick as future solutions, if it gets work done reliably that is the key IMO. Which ironically, could well be the opposite direction Apple will go with the new Mac Pros.

And when people talk about energy costs - I often wonder, do they include the monitors? And all the external devices? I think at full bore a single Xeon uses 55watts. A 32 inch monitor when being used, 45 watts. If reckon external drives would use more power than internal drives. I suspect thought that winchester hard drives would use less power than SSD drives. So a MacPro with some low cost winchester drives is more power friendly than notebook's SSDs. Which are bottlenecked by their cable tech, and cost a lot if they have T3 transfer speeds.

The big power drain though is the GPUs in desktops. And Apple is doing something about that it seems, via GPU and unified memory and no add on GPUs in their recent Macbook Pro Ms - and Apple may do so - IMO they will - with the coming M Mac Pros.

 

[Pressure]

Intel Marketing seems like driving their pistons at full steam lately. Looks like the real thing will be much better than the experimental results from the youtuber above.

Leaked i7-12700H (6P+8E at 45W?):
View attachment 1913932
Full story: https://www.notebookcheck.net/First...-3-and-Apple-M1-Max-in-the-dust.579828.0.html



Not in the sense that people can't use for what they want if they're willing to spend time and develop the software, and cope with potential limits Apple will impose on expandability. I meant if the new Mac Pro has no user upgradable memory, storage, networking & etc, that limits its flexibility and hence general applicability to different usages. And I think it's not a problem for Apple because they perhaps primarily intend the new Mac Pro for a subset of video/graphics production crowds.

I wouldn't worry much about Cinebench as it uses Intel Embree kernels. It does a bad job of testing across architectures like x86 and Aarch64. It's so heavily skewed that the workload isn't even the same.

They are optimised for x86. Apple are currently working on optimising the kernels for Apple Silicon but the pull request is still not merged into the main git.

As it stands now AVX is not available on Apple Silicon and it defaults to the slower SSE2 instruction set. It doesn't even use the specialised Neon SIMD (4x128-bit) available on Apple Silicon to speed things up.

It doesn't seem like an accident the leaks heavily focus on Cinebench though.

 

[4wdwrx]

Sorry, Black Friday has been creeping into Australia, but its not got the same impact here as in the USA.

By the way, when the 7,1 came out there was concern that Apple had built in protection about not only the main drive, but also for the CPU. But Apple's protection leaves the CPU alone. So you can spend $US650 and put in a Intel Xeon W-3245M QS QRSM 3.2GHz 16-Core LGA3647 CPU which has 16 cores. More than double that cost for 24 cores. $US1,600-$1800? There's more choices too. So I'd be feeling great about $4k for a 7,1. If it's its not as quick as future solutions, if it gets work done reliably that is the key IMO. Which ironically, could well be the opposite direction Apple will go with the new Mac Pros.

And when people talk about energy costs - I often wonder, do they include the monitors? And all the external devices? I think at full bore a single Xeon uses 55watts. A 32 inch monitor when being used, 45 watts. If reckon external drives would use more power than internal drives. I suspect thought that winchester hard drives would use less power than SSD drives. So a MacPro with some low cost winchester drives is more power friendly than notebook's SSDs. Which are bottlenecked by their cable tech, and cost a lot if they have T3 transfer speeds.

The big power drain though is the GPUs in desktops. And Apple is doing something about that it seems, via GPU and unified memory and no add on GPUs in their recent Macbook Pro Ms - and Apple may do so - IMO they will - with the coming M Mac Pros.

The standard 8-Core Xeon I believe has 165W TDP and can pull near 200 watts at 100% load. CPUs are power hungry and hot, especially Intel CPUs.

Alder Lake suppose to be more power efficient at normal tasks, although it pulls more at 100%, near 225 Watt.

 

[kvic]

Today Micron announced first validated SoC (from MediaTek) with their LPDDR5X DRAM offering. Interesting news.

I think Micron is the 2nd vendor who made it to LPDDR5X. The first one was Samsung (?): https://www.anandtech.com/show/17058/samsung-announces-lpddr5x-at-85gbps

What's exciting for this thread IMO: Samsung had started manufacturing LPDDR5X, and will be able to offer up to 64GB in a single package. So in a M1 Max like SoC, Apple can ship up to 256GB memory in not so distant future. In a stitched dual M1 Max like SoC, it'll be 512GB. In quad config, it can be a whopping 1TB.

Given the potential release date of New & Smaller Mac Pro between mid and late 2022, it's likely the SoCs will pair with LPDDR5X to offer greater capacity & 33% additional uplift in bandwidth over LPDDR5. That'll be interesting to look forward to and sort of set itself further apart from laptops.

 

[singhs.apps]

It’s given that 64 GB modules were on the way.

Between now and whenever the Mac Pro ships, I gathered 1 TB wasn’t a far fetched idea with such capacities, enough to go way past any other Mac in the line up today and the near future.

It will raise the upfront cost significantly tough if Apple sticks to a SOC based Mac Pro with zero expansions.

Yet I am hoping the mpx modules aren’t a one time phenomenon ( or reserved for pure intel systems ).

There is no need to stick to a 400w envelope for a Mac Pro.

 

[Boil]

Today Micron announced first validated SoC (from MediaTek) with their LPDDR5X DRAM offering. Interesting news.

I think Micron is the 2nd vendor who made it to LPDDR5X. The first one was Samsung (?): https://www.anandtech.com/show/17058/samsung-announces-lpddr5x-at-85gbps

What's exciting for this thread IMO: Samsung had started manufacturing LPDDR5X, and will be able to offer up to 64GB in a single package. So in a M1 Max like SoC, Apple can ship up to 256GB memory in not so distant future. In a stitched dual M1 Max like SoC, it'll be 512GB. In quad config, it can be a whopping 1TB.

Given the potential release date of New & Smaller Mac Pro between mid and late 2022, it's likely the SoCs will pair with LPDDR5X to offer greater capacity & 33% additional uplift in bandwidth over LPDDR5. That'll be interesting to look forward to and sort of set itself further apart from laptops.

Having followed nothing in relation to RAM manufacturing, I have been thinking LPDDR5X might be the way Apple goes with the Mac Pro lineup...!

Memory bandwidths with LPDDR5X:

• M1 Max - 500GB/s
• M1 Max Duo - 1TB/s
• M1 Max Quadro - 2TB/s

Is it WWDC 2022 yet...?!? ;^p

 

[4wdwrx]

Having followed nothing in relation to RAM manufacturing, I have been thinking LPDDR5X might be the way Apple goes with the Mac Pro lineup...!

Memory bandwidths with LPDDR5X:

• M1 Max - 500GB/s
• M1 Max Duo - 1TB/s
• M1 Max Quadro - 2TB/s

Is it WWDC 2022 yet...?!? ;^p

I hope they go full DDR5 for Mac Pro. The low power DDR was never impressive.

 

[Boil]

I hope they go full DDR5 for Mac Pro. The low power DDR was never impressive.

I wonder as to the possibility of a double-sided motherboard (much like the 2019 Mac Pro mobo) allowing DDR5 RAM on the opposite side as the M1 Max SoCs...?

 

[kvic]

It will raise the upfront cost significantly tough if Apple sticks to a SOC based Mac Pro with zero expansions.

Indeed. Increase repair cost significantly too. So I believe more people will buy Apple insurance to extend warranty to perhaps five or more years for their new Mac Pro's with Apple chips.

Also, if some people still haven't realised it, moving Mac Pro to Apple silicon means the end of benefits from democratisation brought to you by the PC world in the past 15 years. Hence, no more inexpensive alternatives for major components such as CPU, GPU and DIMMs.

I hope they go full DDR5 for Mac Pro. The low power DDR was never impressive.

Apple needs LPDDR for the integrated GPU. LPDDR and its memory controllers must be there. So in order to support DDR DIMMs, Apple has to additionally add DDR memory controllers. Just for the sake of a tiny volume of Mac Pro sales. How likely will it be?

Also, imagine in their design review committees, all sorts of arguments on how additional DDR DIMMs destroy the "beauty, elegance, simplicity" of their "unified memory architecture."

 

[Melbourne Park]

...

Also, if some people still haven't realised it, moving Mac Pro to Apple silicon means the end of benefits from democratisation brought to you by the PC world in the past 15 years. Hence, no more inexpensive alternatives for major components such as CPU, GPU and DIMMs.

The most powerful effect on pricing is the effect of the market. If Apple overprices, they won't sell their product. It's always been that way. It's possible to do work on PCs with different software than what Apple users have found productive in Apple's niche apps.

As for democratisation of CPU, GPU and DIMMS, Apple's Xeons were never cheap IMO. And with PCs, GPUs are costly at the moment, due now to producers artificially reducing GPU production to increase GPU prices and hence GPU makers profits ... Apple is offering an alternative route to good graphical performance.

 

[Boil]

The most powerful effect on pricing is the effect of the market. If Apple overprices, they won't sell their product. It's always been that way.

With non-upgradable CPU, GPU, RAM, or SSD; Apple needs to bring the "Apple Tax" down a notch or two; no one will want a top end Mac Pro (Quad M1 Max / 1TB RAM / 8TB SSD) to cost five or six times as much as a base model...!

It's possible to do work on PCs with different software than what Apple users have found productive in Apple's niche apps.

But then you gotta use something beside macOS...? ;^p

As for democratisation of CPU, GPU and DIMMS, Apple's Xeons were never cheap IMO.

And they never will be...! ;^p

And with PCs, GPUs are costly at the moment, due now to producers artificially reducing GPU production to increase GPU prices and hence GPU makers profits ... Apple is offering an alternative route to good graphical performance.

Discrete GPU pricing has gone CRAZY the last three or four years, as has the power usage for an "Gamer/Enthusiast/HEDT PC", 1kW PSUs are becoming the norm for most rigs...?

 

[kvic]

The most powerful effect on pricing is the effect of the market. If Apple overprices, they won't sell their product. It's always been that way. It's possible to do work on PCs with different software than what Apple users have found productive in Apple's niche apps.

As for democratisation of CPU, GPU and DIMMS, Apple's Xeons were never cheap IMO. And with PCs, GPUs are costly at the moment, due now to producers artificially reducing GPU production to increase GPU prices and hence GPU makers profits ... Apple is offering an alternative route to good graphical performance.

With non-upgradable CPU, GPU, RAM, or SSD; Apple needs to bring the "Apple Tax" down a notch or two; no one will want a top end Mac Pro (Quad M1 Max / 1TB RAM / 8TB SSD) to cost five or six times as much as a base model...!

Brand new PC workstations (such as HP Z) aren't cheap either. I was more thinking about retail consumers who buy a new baseline Mac Pro and upgrade memory, CPU & GPU shortly afterwards or in the following years. Then there is the 2nd hand market where new owners propping up the re-sale value of Mac Pro expect a relatively cheap cost to upgrade their machines. Such a playbook may change in the era of Apple silicon Mac Pro. How was it like back in the Power Mac days? Perhaps the coming decade will look somewhat between Power Mac and Intel Mac periods but more like Power Mac days.

Yet I am hoping the mpx modules aren’t a one time phenomenon ( or reserved for pure intel systems ).

There is no need to stick to a 400w envelope for a Mac Pro.

I would think MPX as-is ends with Intel Mac Pro (which btw will continue for a couple more years). On the other hand, I still think MPX-like daughter boards with Apple SoCs are likely to happen in the new Mac Pro. One MPX-like daughter board can potentially house two Jade-4C packages in future. That's already a whopping 80 CPU cores and I'd lost count of how many GPU and NPU cores.

Socketed NVMe blades (less the controller; i.e. Apple proprietary blades) are likely too. This could happen in addition to soldered NVMe drive on the MPX-like daughter boards. Let's not forget a Jade-4C (say stitched from 4 M1 Max) will have up to _four_ NVMe controllers built-in. If I were Apple, I would think it's a brilliant idea to connect all or some of those controllers to sockets on the motherboard

 

[Melbourne Park]

Brand new PC workstations (such as HP Z) aren't cheap either. I was more thinking about retail consumers who buy a new baseline Mac Pro and upgrade memory, CPU & GPU shortly afterwards or in the following years. Then there is the 2nd hand market where new owners propping up the re-sale value of Mac Pro expect a relatively cheap cost to upgrade their machines. Such a playbook may change in the era of Apple silicon Mac Pro. How was it like back in the Power Mac days? Perhaps the coming decade will look somewhat between Power Mac and Intel Mac periods but more like Power Mac days.

With Apple in the Mac Pro era, you had to know what you were doing, and take a gamble on the future. When I bought my 5,1 Mac Pro, it was pretty cheap, and I bought it new. The reason was that Apple introduced a new 5,1 in 2012 or early 2013 - which the base model had an increased clock rate CCP (or twin CPUs). That was the only difference between the models. So I bought the 2010 twin CPU 2.4 four core CPU combination, which was hugely discounted. For the same price, I could have bought a single CPU 6 core, also a 2010, also hugely depreciated. All because of I recall 2.4 versus 2.67 GHz and some extra CPU cache. The gamble back then was that CPU chips of higher cores and clock rates would come down in price. The prices did fall eventually, and hugely.

This time is different. For one, the reason Xeon fast processors became cheap, was I reckon because Xeons were being either thrown out or their CPUs upgraded, which brought the prices way way down. Retail prices also came down. The same with RAM, and increased supply as the RAM became superseded.

These new machines will be quite different. The only chance of replacement CPUs will be if Apple sells them cheaply. People might sell their CPUs though, if they are easy to replace by new faster ones. Hopefully, if we can upgrade from a single CPU to a duel or to a Quad configuration, and that it's affordable to do so, that would be nirvana.

Somehow I think it won't, and I won't find myself in a higher state of perfection - or maybe I'll fined myself in heaven if that actually happens. Along with Bill Gates using a Mac?

Memory is a distance hope IMO. Drives though - I hope we can add some. I suspect with PCI expansion slots, that the majority of buyers only use a few of the slots while under warranty.

But Apple should also remember, that building in upgradability actually increases sales. Firstly it lowers the risk of a purchase decision. Secondly a good second hand market encourages Pro users to upgrade their machines. They can sell the machine for its depreciated value, and just role over the finance cost into a faster piece of hardware. And Pro users love the latest - a new machine is great for morale, besides the better productivity.

A good second hand market depends on affordable upgradability IMO.

 

[kvic]

A good second hand market depends on affordable upgradability IMO.

All valid points. I'm afraid though people's behaviour need change for better or worse in the new Apple silicon era. Cheap & abundant CPU, DIMMs and GPUs from the PC world that people took for granted in the past 15 years will no longer apply in the new Apple era. So far it seems people have greater concern regarding DIMMs and GPUs. Less concern of CPUs. Not sure if it's because of past experience or future expectations. Personally I agree performance uplifts of future compute will mainly come from accelerators and one major form of accelerators is known as GPUs. Their annual technological advancement has already outpaced CPUs. So for a workstation platform, it seems silly to tie up CPU and GPU and _both_ have to be replaced for an upgrade. Perhaps even worse _all_ CPU, GPU & system memory have to be replaced for an upgrade. Love it or hate it, likely that'll the new Mac Pro's. Anyway, if SoCs & RAMs indeed come in a MPX-like daughter board, at least repair, replacement, upgrade, and a variety of SKUs will appear to be easier. Also make things in new Mac Pro's more fungible and hence tradable in the 2nd hand market.

I don't know how the second hand market will play out in the Apple silicon era. Perhaps it depends on how widely institutions and corporations will pick up the new Mac Pro. For example, I believe Apple silicon Macs (though not necessarily Mac Pro) may find a growing number of users in science & technology research and education. But if Apple offer bigger discount on Mac Pro to incentivise..? Yet I think it's not that difficult to foresee replacement parts will be harder and more expensive to source if supply is limited, not to mention again that cheap abundance of the PC world mostly become irrelevant. That certainly won't be healthy to build up a flourishing used market. What do people think?

 

[Boil]

Obviously the following is if you have a need for such distributed processing...

This "disposable computing device" "trend" makes something like a Mn Max-powered Mac mini the sweet spot, especially if Apple offers a model with a dual SoC...

When you "upgrade" by buying an entirely new computer, you just plug the "old" one into a 10Gb network & send assorted jobs to the cluster...? ;^p

 

[deconstruct60]

Today Micron announced first validated SoC (from MediaTek) with their LPDDR5X DRAM offering. Interesting news.

I think Micron is the 2nd vendor who made it to LPDDR5X. The first one was Samsung (?): https://www.anandtech.com/show/17058/samsung-announces-lpddr5x-at-85gbps

What's exciting for this thread IMO: Samsung had started manufacturing LPDDR5X, and will be able to offer up to 64GB in a single package. So in a M1 Max like SoC, Apple can ship up to 256GB memory in not so distant future. In a stitched dual M1 Max like SoC, it'll be 512GB. In quad config, it can be a whopping 1TB.

Samsung got to 64GB how in the package? by adding mulitple "banks" of RAM dies stacked higher, doesn't necessarily sit with Apple's model of running multiple, concurrently active memory channels into a package. ( the active channel versus "bank" fan out from Apple's Soc and vast majority of the smartphone SoC are on different paths. What Samsung probably has there is more for standard Android phone fanout than anything Apple can use. )

The plain M1 and the M1 Pro/Max and multi dies probably has the same baseline design for a memory controller. Just different in number of controllers. And in case of the M1 run at slower speeds agains LPDDDR4 dies used in close to LPDDR5 channel fashion. The die capacity mapping that needs to be mapped by the controller for LPDDR4 is less than that of LPDDR5. So it is not hard to do generation back mappings.

The M1 being in more cost sensitive , higher volume , and earlier shipping systems would be much less risky for Apple to do with LPDDR4. Apple had more options for suppliers and needed semi-custom packages.


LPDDR5X is out on the bleeding edge. Unlikely that standard was built into the designs back in 2018-19 when started the M1 series. Needed semi-custom packages on a bleeding edge memory standard also isn't "low risk" (even before pandemic. which it is worse now).

M3 Max (and bigger ) gets LPDDR5X ? Probably. M2? toss-up (wouldn't be on it. May not be an M2 "better than max" anyway).


Going to > 512GB without ECC is dubious. With just 16-bit channels they'd probably need to "double pump" to get checksum data through ( without side channels). However, that runs opposite of Apple's blazing bandwidth message. Somewhat doubtful Apple is trying to go that large anytime soon. Apple will take the clock bumps but probably not the capacity extremes over the next several iterations.

Given the potential release date of New & Smaller Mac Pro between mid and late 2022, it's likely the SoCs will pair with LPDDR5X to offer greater capacity & 33% additional uplift in bandwidth over LPDDR5. That'll be interesting to look forward to and sort of set itself further apart from laptops.

Release date doesn't matter as much as design date. Pretty doubtful anything with a "M1" prefix is going to come out in late 2022. Anything in the first half of 2022 is very likely old enough to predate LPDDR5X finalization.

A Intel Xeon 3300 series CPU doesn't need "DDR5" suffix to get to 1TB RAM capacities. Depending upon just how much "smaller" a M1 powered system is may not really merit "mac pro" as the system name. ( of so much smaller that squeeze out all the slots , then really should use another name. )

 

[deconstruct60]

Obviously the following is if you have a need for such distributed processing...

This "disposable computing device" "trend" makes something like a Mn Max-powered Mac mini the sweet spot, especially if Apple offers a model with a dual SoC...

Dual SoC is an bit of an oxymoron. If it is a "system on a chip" then don't really need two systems.

Apple can have dual or quad die (chiplets ) that form a single SoC.

The Max die as presented by Apple so far doesn't scale. There will be something other than Max. Whether Apple goes to another name or slaps adjectives after 'Max' as a "modifier" is up in the air.


As for distributing processing. One master hub system with a 4 port 10GbE card would be better distrinuted processing to four Mini' with 10GbE than running all of that single port systems and a switch. Aggregate 40GbE distributed network versus 10GbE.

It is more than just socketed CPU packages and GPU on cards that are helpful upgrades.

When you "upgrade" by buying an entirely new computer, you just plug the "old" one into a 10Gb network & send assorted jobs to the cluster...? ;^p

That presumes that Apple doesn't kneecap the next revisions of the Mini so that only have 1GbE by thinning them out 24" iMac style.

But this also presumes can chop the persistently stored work data into shared and/or distributed pieces. Shipping the job to another box 'costs' just as much as getting the data to the box doing the 'job dispatching' from.

 

[Boil]

Depending upon just how much "smaller" a M1 powered system is may not really merit "mac pro" as the system name. ( of so much smaller that squeeze out all the slots , then really should use another name. )

So no "Pro", just Mac Cube...! ;^p

Dual SoC is an bit of an oxymoron. If it is a "system on a chip" then don't really need two systems.

Apple can have dual or quad die (chiplets ) that form a single SoC.

The Max die as presented by Apple so far doesn't scale. There will be something other than Max. Whether Apple goes to another name or slaps adjectives after 'Max' as a "modifier" is up in the air.

I was more referring to the rumored Jade-2C (dual M1 Max SoCs "stitched" together)...?

As for distributing processing. One master hub system with a 4 port 10GbE card would be better distributed processing to four Mini' with 10GbE than running all of that single port systems and a switch. Aggregate 40GbE distributed network versus 10GbE.

So maybe something like a Mac Cube (assuming it might have a single expansion slot) as front-end workstation & four Mac minis as back-end renderfarm...?

The single expansion slot (PCIe Gen5 x16) would be primarily intended for use with a PCIe expansion chassis; those who need more slots use the expansion slots, those who do not need slots don't have to pay for the extra slots or space...?

 

[kvic]

Based on latest leak on early November from the publication Information, which I believe was one of Apple officially orchestrated leaks again, more realistically I expect next year's new Mac Pro will only come with two M1 Max stitched together. To be released in first half of next year or right after WWDC.

Soldered memory not necessarily LPDDR5X, it may well be officially "overclocked" LPDDR5 with LPDDR5X characteristics, say, 15% higher bandwidth, and 100% higher density in a single package. This will allow Apple to equip the new Mac Prow with 256GB memory in a 2X stitched M1 Max configuration. Come in a MPX-like daughterboard if the new chassis to allow a couple of PCIe slots for internal expansion. Or soldered on motherboard if the new chassis is more like 2013 Trashcan.

The SoCs will come with two NVMe controllers. I think Apple could wire both to sockets on the daughterboard or the motherboard for user replaceable Apple proprietary NVMe blades. Or perhaps one drive soldered one drive wired to a socket.

Modular designs will also allow Apple to easily offer a single M1 Max config with beefed up LPDDR memory at lower price points. That itself sets it apart from the laptop counterpart. Further sweeten up the form factor with possibly internal expansion. Not to blur the line between Mac Mini and Mac Pro, Internal expansions will be distinctive difference. Hence, I believe a redux of 2013 Trashcan is unlikely for the new Mac Pro next year.

So Intel Mac Pro refresh should come out early 2022. Or else no matter from which side you look at it, it'll be kinda an embarrassment for Apple Marketing. On PC side, 2022 is going to be exciting for workstations and high-end desktops too.

Apple is going to support Xeon Mac Pro for a long time. It's doubtful at the moment CPUs will be updated beyond next year's refresh. GPUs as in MPX are likely to be updated for a few more generations. AMD has very exciting flagship RDNA3 next year. So I would think a new MPX based on RX 7900 Extreme with 15,000 cores is very possible in early 2023.

One explanation for Apple to continue updating MPX GPUs is simply to tap into AMD's roadmap for future designs and hence somewhat indirectly the roadmap of Nvidia. I very much doubt it Apple will compete head on with AMD and Nvidia for GPU leadership. It's waste of resource for Apple if they aren't serious about workstations, servers, high-performance compute in data centres and national supercomputer labs. By the same token, somewhat similar can be said for CPUs.

Anyway, Apple will offer unique and interesting high performance devices with great power efficiency for your desks, laps, wrists, hands, and soon heads.

 

[4wdwrx]

I wonder if they will do the Mac Pro like the Mac Mini stack concepts.

A new connector on on top and bottom, derived from MPX, which allows PCIE 5.0 16x, power, and other I/Os.

One M1 Max SoC, maybe M1 "Plaid", per Mac Pro Mini. The more you stack, the software cluster them together. Sort of like their vision for the Mac Mini using ethernet, but instead a direct PCIE x16 connection.

The only way to beyond Max is Plaid.

 

[kvic]

I wonder if they will do the Mac Pro like the Mac Mini stack concepts.

A new connector on on top and bottom, derived from MPX, which allows PCIE 5.0 16x, power, and other I/Os.

One M1 Max SoC, maybe M1 "Plaid", per Mac Pro Mini. The more you stack, the software cluster them together. Sort of like their vision for the Mac Mini using ethernet, but instead a direct PCIE x16 connection.

The only way to beyond Max is Plaid.

A stacked cluster of Mac Mini surely sounds fun, looks cool & dramatic. Will be very nice props for Sci Fi dramas. I believe using PCIe 5 for external interconnects will be very challenging and costly to perform reliably. People who build small form-factor computers usually sandwich the motherboard and GPU around the case's middle backplate. They have to extend the PCIe slot through "ribbon cable" to reach the GPU. Can you imagine how stiff, thick such a cable for PCIe 5 will look like? Now imagine how you're going to design it for connecting Mac mini's

Your idea reminds me of Beowulf Cluster of early 2000s. That later morphed into Blade Servers. Both buzz words had lost their vogue for a long time. But the basic idea of clustered compute has stayed and morphed into better and more efficient form factors. This begs the question why people will need a clustered Mac mini or Mac Pro at home and in office cubicles? Why a single, efficient, powerful box with modular internals won't serve the same purpose..

 

[mattspace]

I wonder if they will do the Mac Pro like the Mac Mini stack concepts.

After all the hullabalo over how we can't possibly have user-serviceable slots for ram, storage and graphics, because the interconnects make everything too slow and laggy...

Can you imagine how stiff, thick such a cable for PCIe 5 will look like? Now imagine how you're going to design it for connecting Mac mini's

Presenting 1994's SCSI cables *gesture to the raising curtain, as the live gameshow audience has a religious experience*

 

[4wdwrx]

A stacked cluster of Mac Mini surely sounds fun, looks cool & dramatic. Will be very nice props for Sci Fi dramas. I believe using PCIe 5 for external interconnects will be very challenging and costly to perform reliably. People who build small form-factor computers usually sandwich the motherboard and GPU around the case's middle backplate. They have to extend the PCIe slot through "ribbon cable" to reach the GPU. Can you imagine how stiff, thick such a cable for PCIe 5 will look like? Now imagine how you're going to design it for connecting Mac mini's

Your idea reminds me of Beowulf Cluster of early 2000s. That later morphed into Blade Servers. Both buzz words had lost their vogue for a long time. But the basic idea of clustered compute has stayed and morphed into better and more efficient form factors. This begs the question why people will need a clustered Mac mini or Mac Pro at home and in office cubicles? Why a single, efficient, powerful box with modular internals won't serve the same purpose..

There are PCIe riser cables that can extend PCIe slot with minimal signal quality degrade, also can extend longer using redrivers for longer lengths or use thinner cables like miniSAS or Oculink. The form factor will be small so there should not require much extension.

A Mac Pro stack would make it easy to expand capabilities while also fulfill Apples profit, since they can make all sorts of stack modules.

The problem with the Mac Mini cluster, was all the power cables needed for each unit and ethernet cords. This stack Mac Pro would pass the power and data through the connector. The stack will be more integrated than a cluster that use ethernet. The PCIE will be direct from M1 to M1, thus more like dual SoC/CPU. It would be like the Cloud but at the edge.

A stacked Mac Pro would be much easier for users to upgrade and service. The current Mac Pro, while nice to have standard PCIE slots, the non-Apple upgrades never work right if not officially supported by Apple.

I've been trying find a solution to use my RTX 3080 while retain sleep functions. I was thinking of relocating the RTX 3080 to external, by extending the PCIe slot x16 using extension riser cables or Oculink. This would allow me to easily disconnect the GPU when not needed, so the Mac can sleep.

 

[kvic]

The problem with the Mac Mini cluster, was all the power cables needed for each unit and ethernet cords. This stack Mac Pro would pass the power and data through the connector. The stack will be more integrated than a cluster that use ethernet. The PCIE will be direct from M1 to M1, thus more like dual SoC/CPU. It would be like the Cloud but at the edge.

A stacked Mac Pro would be much easier for users to upgrade and service. The current Mac Pro, while nice to have standard PCIE slots, the non-Apple upgrades never work right if not officially supported by Apple.

I could not image what kind of form factors such ideas will turn out to be.. The essence of Beowulf Clusters in 2000s was connecting commodity/off-the-shelf PC with fast ethernet. Your suggestion is very similar. Mac replaces PC. External PCIe 5 (+power delivery capability over multiple of 500W) replaces fast ethernet. The problem is Beowulf Clusters are history for a long time. You may argue the Raspberry Pi crowd is re-inventing/re-experiencing Beowulf Clusters (in tiny footprint) but that's very different from productivity requirements. People no longer build Beowulf Clusters for productivity because there are more efficient way to achieve clustered compute.

Twisting your idea a bit. Also dreamed of by another commentator before in this thread. I think a future full-tower Mac Pro with clustered compute nodes internally is surely more plausible. The years long journey can begin with next year's Smaller Mac Pro with a single MPX-like SoC daughterboard (that I've been suggesting) and a couple PCIe slots. In the following years Apple could expand on number of supported MPX-like SoC daughterboards. They can also choose to expand number of SoCs on a single daughterboard (or/and in addition to adding more cores on a single die). In a couple of years, you basically get an extremely powerful box with clustered nodes inside a full-tower box.

Lay down the box horizontally and you get a 5U rack. Racking up a few of these in a standard rack cabinet you get all the power a proprietor or small to medium video production house will ever need..

I've been trying find a solution to use my RTX 3080 while retain sleep functions. I was thinking of relocating the RTX 3080 to external, by extending the PCIe slot x16 using extension riser cables or Oculink. This would allow me to easily disconnect the GPU when not needed, so the Mac can sleep.

I've seen on Github a guy modified his GPU a bit and used Arduino to switch on/off the power supply to the GPU inside a standard PC. I couldn't locate the page with a quick google.

 

[4wdwrx]

I could not image what kind of form factors such ideas will turn out to be.. The essence of Beowulf Clusters in 2000s was connecting commodity/off-the-shelf PC with fast ethernet. Your suggestion is very similar. Mac replaces PC. External PCIe 5 (+power delivery capability over multiple of 500W) replaces fast ethernet. The problem is Beowulf Clusters are history for a long time. You may argue the Raspberry Pi crowd is re-inventing/re-experiencing Beowulf Clusters (in tiny footprint) but that's very different from productivity requirements. People no longer build Beowulf Clusters for productivity because there are more efficient way to achieve clustered compute.

Twisting your idea a bit. Also dreamed of by another commentator before in this thread. I think a future full-tower Mac Pro with clustered compute nodes internally is surely more plausible. The years long journey can begin with next year's Smaller Mac Pro with a single MPX-like SoC daughterboard (that I've been suggesting) and a couple PCIe slots. In the following years Apple could expand on number of supported MPX-like SoC daughterboards. They can also choose to expand number of SoCs on a single daughterboard (or/and in addition to adding more cores on a single die). In a couple of years, you basically get an extremely powerful box with clustered nodes inside a full-tower box.

Lay down the box horizontally and you get a 5U rack. Racking up a few of these in a standard rack cabinet you get all the power a proprietor or small to medium video production house will ever need..



I've seen on Github a guy modified his GPU a bit and used Arduino to switch on/off the power supply to the GPU inside a standard PC. I couldn't locate the page with a quick google.

I like the idea of MPX "NUC compute unit" module with the M1 SoC. The stacked Mac Pro will essentially the same idea but done externally. Referring to a "stack" instead of a cluster, since the Beowulf cluster is essentially discrete computers networked together. PCIe has come a long way and having a direct CPU connection means it can interface multiple SoC directly without a separate protocol.

The stack would be more like you describing, MPX PCIe compute units, but externally. Why external? form factor and price. With a large box, like the current Mac Pro, there is a high starting price and the large form factor also turn off some buyers. The current Mac Pro is too big and expensive for many home users, too expensive for enthusiasts and Pro market, but too small and limiting for enterprise. It really only suitable for small studios and companies.

With a stack, where each Mac Pro, lets say similar or to twice volume as a Mac Mini, it opens the market for larger group of buyers.

ie. one Mac Pro for home use, then enthusiast would have 2-3 stack, then small studio 3 and more. I know it will not fill enterprise use case, but I don't think Apple planned to anyways.


Interesting with the Arduino on/off switch, I'll have to look that up. That being said though, that is the problem with Mac expansion, it always require a workaround or rigging it up, so i think as much as people hate, should as well go proprietary since PC industrial standards do not work as intended.