MP 7,1 Mac Pro 8.1 - WHAT IF...

[maikerukun]

6800x Duos x2 with infinity link.

It must be a software thing then. Whatever softwares you’re running I’m guessing are only utilizing one of the 4 GPU’s available to it in that scenario. As an A6000 in Octane is comparable to an RTX3090, which is 33% slower than 1 w6800x duo. But again, only when speaking to software that takes advantage of it. So I’m fairly certain in the right software that a4000 would likely be equivalent to a single 6800. I’m hypothesizing here but nevertheless I think that’s what’s causing your comparison to happen.

 

[maikerukun]

After a bit more messing about I noted that:

The A4000 on my interior scene is slower when using Optix for LookDev IPR. Use ODIN instead and it’s essentially real-time. Loading textures and them showing in the IPR is nearly instant.

Using a single 6800 out of the four with ODIN is comparable in speed. Use more than one and it is slower. Updating textures is a little slower.

My feeling here is that this is more of a software thing than anything else. It seems that things are more optimized on Windows than Metal on MacOS.

Up until now I hadn’t used OPTIX that much. It’s certainly cool but it’s not fairy dust. My test renders showed how it can over de-noise particularly on small details. To be fair, Noice on Arnold is the same (agin on difficult interiors). ODIN on the other hand always seems to do a good job.

Despite these teething issues, I still marvel at the design of the machine. For kicks I tried to put the Nvidia card in a spare PC. Well nope, that didn’t go well. Didn’t have a power cable long enough, way too many wires etc etc.

But with the 7,1 I can swap out multiple gpus, add in non MPX gpus along with the power connectors in less than 2 mins. It’s so easy and practical and every time I do it I always wonder whether this will be the last Mac you’ll ever be able to do this on and if so, what a shame.

Well…yeah lololol.

Ps - any idea what is the most powerful windows side GPU we can place in a 7.1 Mac Pro?

 

[maikerukun]

I've owned my 7'1 for just over a year and it's been a weird experience. I've owned many a cool thing but I still remember that sense of awe when the I first unboxed it, there really isn't a computer like it. On the other hand I haven't enjoyed the constant bashing from sections of the internet telling me how terrible it is nor having any direction from Apple on their plans for pro's as a company.

Personally I think that some of the criticism of the 7'1 is unfair and it seems to mostly come from people who've never owned one (maybe I'm wrong here?). If it had shipped with a stronger CPU would people have been as critical? That seems its main weakness. I think Apple dropped the ball by not offering a more compelling base model.

When I was building my daughter a PC it was easy to get sucked into the PC's are cheaper and more performant thing. I.e., a $550 I7 CPU performing the same as my 16 core Xeon on Cinebench. Then you start thinking WTF Apple! Until I asked myself does it have 64 PCIE lanes? No. Can I stuff 5 GPU's in it? No. Can it take 8 cards? No. Does it have more than a single Thunderbolt port? No. Is it pretty lol? No. Can it be two computers in one, i.e. Windows and Mac? No.



So maybe it's not that bad after all then?

Bingo.

The overwhelming majority of noise about it comes from non-owners. And noise from owners is almost exclusively owners on the CPU side of things that don’t have the most powerful GPUS in theirs.

You got guys like Max Tech saying the M2 MacBook Pros make the 2019 Mac Pro I älter which…no they don’t lol. They’re not been close on the GPU side as I’ve demonstrated elsewhere in this thread. I love Maxtech but he and others crapping on the 7.1 do so because they have very mid versions of the machine. And to be fair, the CPU is trash 😂😂😂.

Had the CPU been anything north or even near threadrippertown and I don’t care what Apple released next, wouldn’t be upgrading lololol.

 

[maikerukun]

An M-series afterburner type card for an x86-64 machine would seem to be the best solution.

When the afterburner was launched, Apple made a big deal about it being an FPGA and reprogrammable, and I suspect a lot of people thought that meant dynamically, in flight by whatever task was using it.

Just putting a bunch of excess yield A/M series chips on a card would seem to be a win / win / win solution.

This speaks to something I said earlier. Quite frankly, making an afterburner card that gives me the media encoders and the neural engines from the M chips is quite frankly, enough to make my 7.1 an all of a sudden extremely well rounded monster. CPU would literally be the absolute only complaint I have at that point and quite frankly, 1 of these 2 things is all I need. I’m fine not having both. Gimme the m2 afterburner or give me an 8.1 CPU refresh.

 

[maikerukun]

I have a slightly different take on that.

Apple press politely admires it. Tech press acknowledge it.

YouTubers gawk and scratch their heads with awkward smiles and giggles while they reach for various "tests" they don't understand in an effort so that "you and me" (talking to the camera now) can "find out together" what "this thang can do".

Hobbyists whine on forums about how expensive it is and immediately start pointing out the 90% of cases that an Apple Studio Ultra already meets or beats the Mac Pro in and how "no one needs" a Mac Pro.

Professionals mutter in areas of the internet no one even knows how to find anymore: "Hrmf. Mmhmmm. Interesting. I don't know. Will have to test it fist. Hmmrf. Maybe"

Apple boardroom: "Great launch everyone! High fives all around! This sure is one powerhouse! Will be great for pros!" *crickets*


—train of thought continues below in response to smckenzie—


Yep.


Nothing to enjoy about it. But isn't it water off a goose's back for you too?


The way the world turns. People love validating what they have or want by criticising what they can't get.

Despite the high hopes I had for MaxTech when he started his YouTube channel and as far away from it I stay now.... he might be the one exception of a person I know that bought the Mac Pro with high hopes of putting it to good use and found out that it was a bad match for him.

A few valid points of criticism has been covered numerous times in this and other threads. But the loudest voices are from armchair warriors, for sure.


The strengths of a Mac Pro can't always be evaluated in an afternoon. Buying one is more akin to a marathon than a sprint.
An overwhelming majority of people out there—whose voices you hear in relation to reviews and buying guides, and almost everyone the in the first wave of YouTube clips—are buying stuff "out of their own pocket". You'd all be surprised (maybe) how many 'one man bands' out there across small photography, videography and 3D visualising businesses don't know how to charge for services+gear.

People in this category are often critical of seemingly expensive pro gear. For them, a camera is a $2500 expense. A fancy computer costs them $4150 and needs to be saved for, or paid off over a few months.

A pro acquires a $6500 camera that generates $40000 short term. A computer costing $16000 works 24/7 generating many times that cost in revenue. Horses for courses, as it were.

Hilarious. I just posted a response not :30 seconds ago I believe to the exact same thing and used Maxtech as an example lol.

 

[maikerukun]

I for one certainly hope the 8,1 ASi Mac Pro smashes it out of the park, with stonking increases in performance that have pro's salivating for an upgrade. Having waited an age for the 2019, I expect you all rushed out and bought one at launch; by 2023 you're hopefully dying for an upgrade.

With any luck you'll all be looking to offload your 7,1 MPs at once, and the value will go through the floor (hey, your projects have already long paid for them). Only hobbyists like me will be interested in a hulking x86 Mac, but I'd buy one as a (massive) upgrade from my 5,1 - if the price is right. I have no doubt they're beautifully constructed, and I'd like to be able to continue rebooting into / virtualising Windows as desired.

OTOH, if pro's hang on to them as render boxes or file servers, I'll likely head in the direction of a Studio Max.

Btw, I think it very unlikely Apple will release any new x86 Mac at this point. I think they'd rather wait until 2027 to release an ASi 8,1, saying nothing all the time, than force themselves to support x86 macOS a day longer than they need to. Though I'm also sure it won't come to that, and the new MP will be along in the next few months. It's possible too that the 7,1 will hang around for a while, 2018 Mac mini-style, for those who need specific aspects of the 7,1 that aren't replicated by the first-gen ASi MP.

I was just discussing this with my girlfriend who saw this thread and wondered what I was planning to do with my 7.1 if the 8.1 turns out to be a machine that can actually replace this and then some.

My machine is a 28 core CPU, 384 gigs of ram, with 2 w6800x duos in it. I’ll likely be letting it go with all the bells and whistles for around $20k or so.

But the 8.1 would have to not only match this one on EVERY SINGLE LEVEL GPU INCLUDED…but furthermore it would have to be expandable with the ability to completely obliterate my current 7.1.

Fingers crossed.

 

[Lord Blackadder]

But the 8.1 would have to not only match this one on EVERY SINGLE LEVEL GPU INCLUDED…but furthermore it would have to be expandable with the ability to completely obliterate my current 7.1.

If the 8,1 turns out to be a Xeon-based stopgap, I doubt whether it will 'obliterate' the 7,1 except perhaps in a few narrow areas. An Apple Silicon successor is much more of an open question.

 

[mattspace]

If the 8,1 turns out to be a Xeon-based stopgap, I doubt whether it will 'obliterate' the 7,1 except perhaps in a few narrow areas. An Apple Silicon successor is much more of an open question.

Could very well be it's just a new Xeon, pci-5 version of exactly the same machine, with "Afterburner 2", which is fully backwards compatible with the 7,1 as a demonstration to a deep commitment to the Pro crowd - new Xeon Mac Pro at Intel's pace, new Afterburner every 12-18 months. Perfectly salvageable from a PR perspective - the CPU is just one part of the machine, and doesn't matter any more than the wifi chip - what matters is how much AS processing is available for apps to use etc

 

[AlphaCentauri]

You guys are true dreamers. I will shave and eat my beard if there’s another Intel Mac Pro…

 

[prefuse07]

HERE is an interesting read (if true)

 

[Lord Blackadder]

You guys are true dreamers. I will shave and eat my beard if there’s another Intel Mac Pro…

I consider it quite unlikely myself. But not impossible. It makes more sense to just release a refresh to the 7,1 if needed than create a whole stopgap model.

 

[AlphaCentauri]

I consider it quite unlikely myself. But not impossible. It makes more sense to just release a refresh to the 7,1 if needed than create a whole stopgap model.

Yeah, but that’s Apple we’re talking about. Sensible solutions seem to be elusive to them quite often…

 

[prefuse07]

I consider it quite unlikely myself. But not impossible. It makes more sense to just release a refresh to the 7,1 if needed than create a whole stopgap model.

Yup, 100% agree, as I outlined in my previous post -- Releasing the 8,1 as a "refresh" and their very last x86 machine makes a lot of sense, and may even save them some cash since they already have 7,1s to "upgrade", with the most important one being that it buy apple time to mature AS, because where it currently stands doesn't even come close to Nvidia/AMD (and maybe the new Intel cards?).

 

[prefuse07]

Yeah, but that’s Apple we’re talking about. Sensible solutions seem to be elusive to them quite often…

More importantly, it's Timmy who is driving the mothership, so your point is even more valid

 

[mode11]

A new Intel Mac Pro would not only extend the required Intel macOS support by another few years, it would raise the bar for the ASi MP replacement. Quite substantially, if talking 13th gen Intel with up to 56 cores, and 7900XTX-based GPUs.

They want people hungry for the replacement, and for the ASi MP to demonstrate a worthwhile upgrade in at least most respects. The 7,1 will have no successor until the ASi version is ready to ship, even if that takes years. See: previous Mac Pro history.

 

[AndreeOnline]

You guys are true dreamers.

Agreed. The Mac Pro that has been tested has been AS and the whole transition is about releasing the AS Mac Pro, so it's hard to see where the Intel angle comes from.

But, something that I could believe is...

Kuo says it's increasingly unlikely to see an ARVR headset in March. Aaaand that sort of, kind of makes more room for Apple to focus on getting the Mac Pro out the door. I don't think there are too many technical challenges left for this first gen AS Mac Pro.

Let's make the March event a Pro Event!

 

[jmho]

HERE is an interesting read (if true)

If Apple are doing what I think they're doing then I'm incredibly excited.

• Earlier in this thread I posted a WWDC video where Apple daisy-chained together 4 studios with thunderbolt to perform machine learning.
• Apple's Swift language is building up towards version 6 which is aiming to be a radically thread-safe language, and they've also just come out with a feature called Swift Distributed Actors which makes it "easy" to run code across distributed systems.
• There's also the fact that Raspberry Pi has something that is also called a "Compute Module" which is a tiny standalone ARM SoC that plugs into a main board.
• Also the fact that Apple seemingly cancelled the 2xUltra variant makes me think that the future of the Mac Pro is distributed.

To me all these things point to the Mac Pro being a cluster of Apple Silicon SoCs working in parallel like the nVidia DGX A100.

I think contrary to the pessimism of this thread the 8.1 might end up being an absolute monster (for certain workloads)

 

[prefuse07]

If Apple are doing what I think they're doing then I'm incredibly excited.

• Earlier in this thread I posted a WWDC video where Apple daisy-chained together 4 studios with thunderbolt to perform machine learning.
• Apple's Swift language is building up towards version 6 which is aiming to be a radically thread-safe language, and they've also just come out with a feature called Swift Distributed Actors which makes it "easy" to run code across distributed systems.
• There's also the fact that Raspberry Pi has something that is also called a "Compute Module" which is a tiny standalone ARM SoC that plugs into a main board.
• Also the fact that Apple seemingly cancelled the 2xUltra variant makes me think that the future of the Mac Pro is distributed.

To me all these things point to the Mac Pro being a cluster of Apple Silicon SoCs working in parallel like the nVidia DGX A100.

I think contrary to the pessimism of this thread the 8.1 might end up being an absolute monster (for certain workloads)

And that actually sounds super exciting!!!

 

[innerproduct]

Haven’t we discussed the MacStack before? I have a distinct feeling that I myself have brought up that option even lol. Like an “ultra” version of “continuity”.
But it has its own issues. Want dual perf? Pay exactly 2x. Buy memory twice etc. While I like the idea of using all your devices as a a cluster it is a complementary solution and we still need a power mac.

 

[jmho]

Want dual perf? Pay exactly 2x. Buy memory twice etc.

No different from buying 2 GPUs to put in an x86 machine.

It's a good point though that price is everything here, and I wouldn't put it past Apple to sell cut down "Compute Modules" for a huge premium just because they're for the pro market.

I don't believe the rumour about compute modules being Mac Minis / Studios.

 

[innerproduct]

No different from buying 2 GPUs to put in an x86 machine.

It's a good point though that price is everything here, and I wouldn't put it past Apple to sell cut down "Compute Modules" for a huge premium just because they're for the pro market.

I don't believe the rumour about compute modules being Mac Minis / Studios.

The nerd in me really like the idea of effortless compute cluster setup and hiding it behind APIs.
Having 2-10 m2 maxes available in the 2019 chassis seems about right

 

[mode11]

The idea of the Mac Pro being some kind of cluster sounds less due to being an ideal technical solution, and more like "we've got a bunch of laptop-focussed SoCs that we don't want to change too much for a niche machine, what can we cobble together that has an impressive number of total TFLOPS"? It sounds like something that could be great for embarrassingly parallel stuff, but less so for single threaded work. Though admittedly, Apple are currently a leader in the latter.

It also sounds like a very expensive way of reinventing the GPGPU wheel, simply to suit Apple's mobile-first architecture. Makes sense for Apple, but for the customer? A workstation may be a depreciating business asset, but companies don't just throw money away for the sake of it.

 

[Lord Blackadder]

The idea of the Mac Pro being some kind of cluster sounds less due to being an ideal technical solution, and more like "we've got a bunch of laptop-focussed SoCs that we don't want to change too much for a niche machine, what can we cobble together that has an impressive number of total TFLOPS"? It sounds like something that could be great for embarrassingly parallel stuff, but less so for single threaded work. Though admittedly, Apple are currently a leader in the latter.

It also sounds like a very expensive way of reinventing the GPGPU wheel, simply to suit Apple's mobile-first architecture. Makes sense for Apple, but for the customer? A workstation may be a depreciating business asset, but companies don't just throw money away for the sake of it.

Possibly, but in the longer-term it is clear that Apple intends to rely exclusively on its own line of silicon that competes with the leading CPU and GPU makers. So while what happens in the short term with Mac Pro is perhaps debatable, the future seems fairly clear unless the Apple silicon project runs into major obstacles.

 

[mode11]

The better solution may be chiplet-based, with some kind of hub chip that allows 4x Max's to connect. Or perhaps custom (expensive) chiplets that separate CPU and GPU cores, to allow for more flexible pairings. It may just be that the Mac Pro needs to have relatively low margins, even to stay competitive with Windows workstation pricing, and Apple suck up the cost there. They're presumably making huge savings elsewhere from consolidating software / hardware across iOS and macOS.

Whilst on a pure spreadsheet basis, the Mac Pro likely doesn't look too attractive to Apple, it arguably serves an essential role for the macOS platform. It acts as the heavyweight range-topper that stops the bleed of high-end users to rival platforms, which could result in others lower down the stack following suit (i.e. if companies would rather standardise on one platform, and PCs are available at all levels).

 

[deconstruct60]

If Apple are doing what I think they're doing then I'm incredibly excited.

• Earlier in this thread I posted a WWDC video where Apple daisy-chained together 4 studios with thunderbolt to perform machine learning.
• Apple's Swift language is building up towards version 6 which is aiming to be a radically thread-safe language, and they've also just come out with a feature called Swift Distributed Actors which makes it "easy" to run code across distributed systems.
• There's also the fact that Raspberry Pi has something that is also called a "Compute Module" which is a tiny standalone ARM SoC that plugs into a main board.
• Also the fact that Apple seemingly cancelled the 2xUltra variant makes me think that the future of the Mac Pro is distributed.

1. The "four studio via Thunderbolt" is primarily leveraging an existing TBv2+ feature of point-to-point 10GbE for 'free' over TB. The major 'feature' there is cost savings via no 10GbE switch to manage the connectivity. As far as 'distribute over Ethernet (virtual one or not)' that would work just as well with a 10GbE (or better) switch as without one.

2. Swift Distributed Actors

"... This abstraction does not intend to completely hide away the fact that distributed calls are crossing the network, though. In a way, we are doing the opposite and programming assuming that calls may be remote. This small yet crucial observation allows us to build systems primarily intended for distribution and testable in local test clusters that may even efficiently simulate various error scenarios. ..."

Introducing Swift Distributed Actors

We’re thrilled to announce a new open-source package for the Swift on Server ecosystem, Swift Distributed Actors, a complete server-oriented cluster library for the upcoming distributed actor language feature!

www.swift.org

I does not hide the distribution (meaning need code changes). There are lower hurdles if using the concurrency model of Swift already, but need to presume in the application being written that all concurrency calls are remote ( and if just happen to be local , then that is the 'transparent' part. ). This is substantially different in spinning this makes assumed local calls transparent distributed. Applications would have to be changed around this.

At the top of the blog post there is also

" ... a complete server-oriented cluster library for the upcoming distributed actor language feature! ... "

Apple has been pushing Swift as viable non-macOS (i.e., Linux) , server backend language tool. There isn't much macOS there. Apple has killed their "macOS Server" product. Apple's Cloud services largely run on Linux ; not macOS. If Apple is 'eating their own dog food' here, it is far more likely on Linux than on macOS.

A server and a single user , high GUI workstation are two different markets.






3. The Raspberry Pi 4 compute module does not plug into the main regular Pi board.

The plain compute module is a board stripped of some ports and some embedded/industrial features added.

https://www.makeuseof.com/raspberry-pi-4-vs-raspberry-pi-compute-module-4-key-differences/

The 4S Compute Module looks like a DIMM but is electrically not a DIMM. ( main Pi 4 board doesn't have a DIMM slot either).

Raspberry Pi Compute Module 4S is a SODIMM with the same processor as the Raspberry Pi 4 and CM4 - Liliputing

Raspberry Pi Compute Module 4S is a SODIMM with the same processor as the Raspberry Pi 4 and CM4

liliputing.com

[ Note: the I/O ports being provisioned over these DIMM-like connections are no where near the bandwidth levels a M-series would need to hit. ]



4. The rumor is that Apple cancelled the 4 die solution relatively very late in the process. Apple is doing a major pivot isn't particularly likely. They'll just go without that one , of two , SoC options they had planned.

Stuffing 2-4 independent Mac instances onto cards and inside of a Mac Pro really isn't going to completely cover the same space that a much larger SoC would with the same applications. The applications that presume one , unified address space are not really built to be distributed.

To me all these things point to the Mac Pro being a cluster of Apple Silicon SoCs working in parallel like the nVidia DGX A100.

Not like a DGX at all. DGX has very elaborate mechanisms for doing NUMA shared memory. There is nothing in the above that really highlights that at all. iOS as a key/critical foundational element certainly does not .

If Apple was doing a Mac M.o.C. ( Mac on a Card) there is some synergies with their support of the 'rent a Mac in the Cloud for at least a day" marketplace. If put 3-4 Macs on a 75W PCI-e card and then used the host CPU/GPU/etc infrastructure to serve as the core file/compute/etc serve for a group of folks of an individual business that would have some synegiers. A rack version of the Mac Pro is around 18" wide. Say 16" and diviide by 2" means could get 8 Mini / MiniPro in roughtly the same rack space. That is more dense CPU compute than a "cluster inside the box of M2/M2 Pro" but it is much neater and if layer virtual Ethernet over the PCI-e switch backplane ... possibly cheaper also (since skipping 10GbE (or better) switch.

Some other folks could also use it as a "cluster inside the box" in a tower/deskside form. And when M3 Pro , M4 Pro , etc comes out can upgrade nodes in the cluster and keep the same central chassis.

Apple would still need a GPGPU compute accelrator for other workloads though (e.g., MI210 , MI310 ). "Macs on a Card" still doesn't get you ECC for data integrity sensitive HPC workloads. And the ALU density isn't going to keep up either.

I think contrary to the pessimism of this thread the 8.1 might end up being an absolute monster (for certain workloads)

If they are doing a "Mac MoC" on a regular 75W add-in-card it wouldn't necessarily be limited to the Mac Pro. Could put the cluster module in a Thunderbolt PCI-e card enclosure and get still get around a virtual 10GbE connection back to the host MBA , Mini , or Studio.

Doing something that is hyper coupled to a very minor variant of the MPX slot would mean would limit sales just to the Mac Pro. If going to do some kind of compute module it probably would work better to sell more rather than less. It could be a card inside of a Windows back if just connecting back to the host via a virtual Ethernet-over-PCI-e connection.