MP All Models 2023 Mac Pro! [MERGED]

[singhs.apps]

I think we’ll have a large, iPad like pencil enabled display, if not the whole computer.

The ever expanding UI+hardware synergy between macs and iOS devices (specifically iPad Mx chips) points to a touch interface in the macs in the future.

The Apple vision pro might throw a spanner in my speculations but it still a while away before it’s practical enough to be an everyday, all day computing/work device.

 

[mattspace]

Having just watched Humane's AI Pin demo video, I honestly don't think Apple has much of a future in anything.

Humane just threw the entire core of Apple's business - devices to *look at* your digital self (messages, notes, interpersonal relationship management) under a bus, and when you don't need that stuff to be on the device you use to also do "looking at" tasks, that really shifts the lockin gravity away from Apple.

No one "cool" "influencer" or "trendy" is going to be seen dead with an iPhone once this pin is released.

 

[ZombiePhysicist]

Humane Ai Pin | See the World, Not Your Screen. | Humane

Ai Pin is your intelligent companion. With just a touch and your voice, connect and be present without distraction. Explore Ai Pin experiences today.

hu.ma.ne

 

[mode11]

But how will you doomscroll MacRumors on that? It would just be a case of:

“AI Pin: catch me up on the MacRumors forum”

“Everyone’s complaining about Apple prices and their locked down hardware”

Where’s the fun in that?

 

[mode11]

that really shifts the lockin gravity away from Apple.

And on to Humane’s servers?

It does look impressive, albeit straight out of Black Mirror.

 

[treehuggerpro]

Strange device, ready made for dystopian overlords?

 

[jimmy_john]

Can’t take 600w GPUs so clearly not pro.

 

[mattspace]

And on to Humane’s servers?

yup, but humane might just operate as the interface for a heterogenous ecosystem of devices that coordinate through the badge - imagine instead of your music on your big phone, your badge talked directly to an iPod, which can stay in your pocket. Imagine if iCloud was a dedicated device and service that wasnt tethered to any one company.

I think aboul it as if 2007 onwards had been based on iPods expanded to make phone calls, rather than Macs shrunk to fit in pockets - no twitter, no instagram, no pervasive social media.

It does look impressive, albeit straight out of Black Mirror.

I mean the smartphone is a 1984 televisor you can carry in your pocket, and your government can even require you to own one to access services...

 

[ZombiePhysicist]

And on to Humane’s servers?

It does look impressive, albeit straight out of Black Mirror.

I'll get it to play with, but considering it will be hacked most definitely, and it captures all your most intimate details to mine AI poop about you, constantly, by a company with absolutely no track record, and high likelihood to be bought out by google 3 minutes later, yea pass.

I'm in the minority that likes to own and control my own data on my own hardware. I can't believe how quickly people forgot Steve Jobs line of "I dont like renting my music". Now everyone does (except me, I still buy it) and they are "happy owning nothing". Bizarre world that seems incapable of learning from it's own past.

 

[mode11]

I think aboul it as if 2007 onwards had been based on iPods expanded to make phone calls, rather than Macs shrunk to fit in pockets - no twitter, no instagram, no pervasive social media.

Social media has a lot to answer for, though like much wrong with the Internet, the larger problem may be the way it's funded through advertising.

A spoken interface has limits of its own though. What if you want to read something private? You can put in earphones, but that's less convenient than just glancing at a phone. Plus if everyone's doing that (which might be necessary to avoid a cacophony in public spaces), it just isolates people further, similar to VR. And some things are inherently visual; you can take photos and videos on a Pin, but there's no way to review / show them to others (short of potentially lasering them onto your hand in monochrome). Plus, even writing this forum post would be painful if it had to be dictated and repeatedly read back, rather than just editing with a keyboard.

An AI that can summarise emails and messages and tell you anything actually relevant or useful would be pretty awesome though - basically an AI PA. And the general concept of a screenless device is an interesting one. It's stripped of everything but useful information, decluttering your 'information space'.

I mean the smartphone is a 1984 televisor you can carry in your pocket, and your government can even require you to own one to access services...

That's a bit of a stretch. I guess post-Snowdon we can assume that email, SMS and phonecalls are routinely scanned electronically for key words, and I expect governments have backdoors (official and otherwise) into smartphone services, so if you were a 'person of interest' it would be easy enough to record audio / video from your phone without you knowing. But presumably that last level of intrusion requires a warrant, and in any case, people (including myself) go about their lives unaware of any of this - we don't feel like we're living under an oppressive cloud of fear. Though you'd assume that anyone involved in serious criminal activity would use burner phones, and treat any electronic communication as if it were tapped. In any case, having your whole life channelled through one AI service doesn't sound like an improvement in security / surveillance.

 

[mattspace]

But presumably that last level of intrusion requires a warrant, and in any case, people (including myself) go about their lives unaware of any of this - we don't feel like we're living under an oppressive cloud of fear. Though you'd assume that anyone involved in serious criminal activity would use burner phones, and treat any electronic communication as if it were tapped. In any case, having your whole life channelled through one AI service doesn't sound like an improvement in security / surveillance.

In Australia, all the metadata of every form of electronic communication is captured by law, and almost anyone can get unknowable, and un-notified access to it - your local municipal council can use it to find out if you've been knowingly failing to separate your garbage, the animal protection charity which has prosecutorial powers can use it, etc.

But I would imagine Humane is no different to Apple in terms of what they capture and know, to be honest. Apple mines its customers just as much as any company.

This is purely speculation, but I could see a path for Humane as being like a combination of a home automation hub, and device / platform independent iCloud - a thing that connects and coordinates all your "humane aware" devices, so they can talk peer to peer in their local vicinity, with the pin mediating the networking. Your Humane enabled Mont Blanc pen records all of your pen strokes, and they're available in any word processor on any device that can access your humane persona, etc.

 

[mode11]

In Australia, all the metadata of every form of electronic communication is captured by law, and almost anyone can get unknowable, and un-notified access to it - your local municipal council can use it to find out if you've been knowingly failing to separate your garbage, the animal protection charity which has prosecutorial powers can use it, etc.

Does a VPN help with this type of thing? Haven't really looked into it. Don't 'have anything to hide', but do object to being relentlessly datamined.

I would have no trouble believing that Apple mines its customers just as much as any other company, but to what end? They don't insert any noticeable advertising into anything, so how would this activity pay for itself? And I doubt they sell your information, given the hoopla they make about privacy (though when most companies say they 'don't sell your data', it's only because it's far too valuable).

 

[mattspace]

Does a VPN help with this type of thing? Haven't really looked into it. Don't 'have anything to hide', but do object to being relentlessly datamined.

So long as you don't use the VPN provider's local node, I imagine. A key factor of the law is that the government can force any local company to provide backdoors to any piece of hardware here, and the punishment for revealing even the existence of the request are serious jail time.

So any international company is nuts to host anything here, because the *staff* at the facility where the server is hosted can have the order issued against them personally.

Oh yeah, resigning to avoid having to comply with the notice, you better believe that's a paddlin'.

I would have no trouble believing that Apple mines its customers just as much as any other company, but to what end? They don't insert any noticeable advertising into anything, so how would this activity pay for itself? And I doubt they sell your information, given the hoopla they make about privacy (though when most companies say they 'don't sell your data', it's only because it's far too valuable).

The people behind Humane as far as I can see, are folks who left Apple post-Jobs, who perhaps didn't have confidence in the Cook era.

Honestly, they're selling a USD$700 device, with a $25/month cellphone plan that won't use a lot of data because it's not browsing web pages etc, and no doubt there'll be an accessory ecosystem, and fashion tie ins...

They may just be an iPod-era Apple type company who don't need to do advertising revenue.

 

[mode11]

They may just be an iPod-era Apple type company who don't need to do advertising revenue.

I meant why would Apple mine your data? You seemed to think they do.

 

[mattspace]

I meant why would Apple mine your data? You seemed to think they do.

Apple has a pretty significant advertising business *shrug*

 

[danwells]

I found something today that goes into the "how fast is Apple Silicon compared to Wintel workstations" sub-thread.

There is a weird discrepancy between Geekbench 5 and 6... The M3 Max scores about the same (about 21000 on GB6, ~23000 on GB 5, both multi-core scores). Windows systems with huge numbers of cores (anything in the serious workstation market) score much, much higher in GB5 than GB6

The Threadripper Pro 7995WX (96 core monster) scores around 26,000 on GB 6 (easy prey for the M3 Ultra, not even that much faster than the Max), but about 81000 on GB5 (it would take an M3 Extreme (double Ultra) to come close, and maybe not even then)...

The Xeon 3495X (56-core) scores just under 20,000 in GB 6, and GB 5 results are all over the place, but up to 50,000+

Does GB 5 give chips with a ton of cores an unrealistically fast score, is GB 6 unrealistically slow, or a little of both? I can't imagine GB6 is really being fair to big multi-core systems - the top multi-core scores are all desktop chips (13700K and 13900K), almost certainly heavily overclocked. Those numbers are all right about where I'd expect an M3 Ultra to come in (1.5x the M3 Max - doubling cores doesn't scale perfectly).

I'd expect overclocked desktop/gaming chips to dominate single-core scores, because the Xeons and Threadrippers have much less power PER CORE to play with (4x the cores, but only twice the TDP). I wouldn't expect a 16 core chip to beat a 64 core using the same cores at multicore by overclocking, at least not unless you're running a clock speed around TWICE as high.

 

[jimmy_john]

I found something today that goes into the "how fast is Apple Silicon compared to Wintel workstations" sub-thread.

There is a weird discrepancy between Geekbench 5 and 6... The M3 Max scores about the same (about 21000 on GB6, ~23000 on GB 5, both multi-core scores). Windows systems with huge numbers of cores (anything in the serious workstation market) score much, much higher in GB5 than GB6

The Threadripper Pro 7995WX (96 core monster) scores around 26,000 on GB 6 (easy prey for the M3 Ultra, not even that much faster than the Max), but about 81000 on GB5 (it would take an M3 Extreme (double Ultra) to come close, and maybe not even then)...

The Xeon 3495X (56-core) scores just under 20,000 in GB 6, and GB 5 results are all over the place, but up to 50,000+

Does GB 5 give chips with a ton of cores an unrealistically fast score, is GB 6 unrealistically slow, or a little of both? I can't imagine GB6 is really being fair to big multi-core systems - the top multi-core scores are all desktop chips (13700K and 13900K), almost certainly heavily overclocked. Those numbers are all right about where I'd expect an M3 Ultra to come in (1.5x the M3 Max - doubling cores doesn't scale perfectly).

I'd expect overclocked desktop/gaming chips to dominate single-core scores, because the Xeons and Threadrippers have much less power PER CORE to play with (4x the cores, but only twice the TDP). I wouldn't expect a 16 core chip to beat a 64 core using the same cores at multicore by overclocking, at least not unless you're running a clock speed around TWICE as high.

It could be because synthetic benchmarking has always been a terrible way to evaluate system performance, ripe for tinkering by vendors, and not representative of real world usage.

Even within that realm, Geekbench has always been one of the least consistent.

 

[avro707]

I found something today that goes into the "how fast is Apple Silicon compared to Wintel workstations" sub-thread.

There is a weird discrepancy between Geekbench 5 and 6... The M3 Max scores about the same (about 21000 on GB6, ~23000 on GB 5, both multi-core scores). Windows systems with huge numbers of cores (anything in the serious workstation market) score much, much higher in GB5 than GB6

The Threadripper Pro 7995WX (96 core monster) scores around 26,000 on GB 6 (easy prey for the M3 Ultra, not even that much faster than the Max), but about 81000 on GB5 (it would take an M3 Extreme (double Ultra) to come close, and maybe not even then)...

The Xeon 3495X (56-core) scores just under 20,000 in GB 6, and GB 5 results are all over the place, but up to 50,000+

Does GB 5 give chips with a ton of cores an unrealistically fast score, is GB 6 unrealistically slow, or a little of both? I can't imagine GB6 is really being fair to big multi-core systems - the top multi-core scores are all desktop chips (13700K and 13900K), almost certainly heavily overclocked. Those numbers are all right about where I'd expect an M3 Ultra to come in (1.5x the M3 Max - doubling cores doesn't scale perfectly).

I'd expect overclocked desktop/gaming chips to dominate single-core scores, because the Xeons and Threadrippers have much less power PER CORE to play with (4x the cores, but only twice the TDP). I wouldn't expect a 16 core chip to beat a 64 core using the same cores at multicore by overclocking, at least not unless you're running a clock speed around TWICE as high.

I suspect bench-marketing is less valid for those serious high performance massively-multi-core workstations that real world actual work they do.

Yes, I know, they don't do real world work, the Apple Silicon crowd says they are all the domain of cashed up enthusiasts.

 

[Waragainstsleep]

Nope, still keeping it as called. It aint going to happen.

The cost of big panels is not compatible with an iPad-disposable appliance strategy, and the environmental impacts of pulling those panels from service while they still work will never be matched by efficiency in newer generations of hardware. Also, Apple has a new "big screen computing" solution, and it's a headset.

Big iMac is dead.

What if its not an LCD panel, but an AR display? Just a big white rectangle the Vision Pro project onto.

 

[mode11]

What if its not an LCD panel, but an AR display? Just a big white rectangle the Vision Pro project onto.

Then it's got nothing to do with the iMac, so is irrelevant to the discussion?

 

[ondioline]

Does GB 5 give chips with a ton of cores an unrealistically fast score, is GB 6 unrealistically slow, or a little of both?

It’s just a completely different test. GB5 multicore afaik was executing a process on each core in a parallel way. GB6 is doing something closer to concurrent operations which probably resemble something like X amount of threads processing data from a channel with a mutex.

 

[mattspace]

What if its not an LCD panel, but an AR display? Just a big white rectangle the Vision Pro project onto.

I've said that before, that I thought what apples glasses tech would do would be to turn a "dumb" monitor into something else, but I don't think that's going to happen.

There's no reason for Apple to sell you a big white rectangle if you already have glasses on, and the route Apple is going with Vision Pro isn't one I believe has ordinary spectacles with augmentation as an outcome in the next decade.

 

[danwells]

It’s just a completely different test. GB5 multicore afaik was executing a process on each core in a parallel way. GB6 is doing something closer to concurrent operations which probably resemble something like X amount of threads processing data from a channel with a mutex.

If that's the case, GB6 multicore (which shows the Ultra chips and the M3 Max as EXTREMELY powerful - right up there with very big Windows workstations) is probably closer to the truth in an average workstation use case than GB5 multicore, which sounds like it almost requires a multi-user system to generate enough independent threads. A GB5 benchmark sounds closer to accurate for a VAX from 30 or 40 years ago, when you might have had 100 people logged in at once, using text editors and checking e-mail. 100 copies of emacs is a much easier test for a many-core system than one copy of Photoshop...

If schedulers are that bad at using 50+ cores, then the M-series chips have a huge advantage at classic workstation tasks. An M3 Ultra should easily beat even a Threadripper Pro 7995WX at a GB6-like task mix (let's see if that's true on something like a fully non-Rosetta PugetBench). To be fair, all the Xeons and Threadrippers are really server chips - they are designed for workloads that are actually closer to that old VAX - a million little tasks rather than a few big ones...

Even if GB6 is unrealistically optimistic, and real-world creative tasks show it as close to the big Threadripper, but not ahead, remember just how huge a difference there is between the systems in price and livability.

An M3 Ultra is going to start at $4000-5000, and it'll go up to $10,000 or so with high RAM and storage - maybe $12,000 if Apple has some monster options. A full 256 GB of RAM, but with somewhat modest storage (2TB) should be around $7000. Add external SSD storage for a few hundred to a few thousand more (depending on your needs). I wish Apple offered add-it-yourself NVMe, but they don't... $10,000 buys a heck of a system.

$10,000 buys a Threadripper Pro 7995 WX, too - a bare CPU. Building a decent workstation around it is going to be right around $20,000. Maybe less if your work isn't RAM or storage intensive (. Maybe more for super configurations (that the Mac Studio may not be able to match). Just adding up the parts, it's really expensive

CPU: $10000
Motherboard: $1000 (conservative estimate - these aren't out yet, previous generation is $1000)
Case, PSU, cooling: $1000+
GPU:$2000
RAM: $1000 for 256 GB
Storage: as desired (2 TB is only a couple of hundred dollars)

Where the Mac is probably going to be about $7000 with 256 Gb of RAM and a 2 TB system/application drive, a 7995WX will be about $15000 in the same state.

A M3 Extreme (double M3 Ultra) Mac Pro (if such a thing ever exists) will compete with the big Threadripper on price, and if GB6 is correct, may well be twice as fast for tasks that don't love 96 cores...

 

[mode11]

If that's the case, GB6 multicore (which shows the Ultra chips and the M3 Max as EXTREMELY powerful - right up there with very big Windows workstations) is probably closer to the truth in an average workstation use case than GB5 multicore, which sounds like it almost requires a multi-user system to generate enough independent threads. A GB5 benchmark sounds closer to accurate for a VAX from 30 or 40 years ago, when you might have had 100 people logged in at once, using text editors and checking e-mail. 100 copies of emacs is a much easier test for a many-core system than one copy of Photoshop...

If schedulers are that bad at using 50+ cores, then the M-series chips have a huge advantage at classic workstation tasks. An M3 Ultra should easily beat even a Threadripper Pro 7995WX at a GB6-like task mix (let's see if that's true on something like a fully non-Rosetta PugetBench). To be fair, all the Xeons and Threadrippers are really server chips - they are designed for workloads that are actually closer to that old VAX - a million little tasks rather than a few big ones...

Even if GB6 is unrealistically optimistic, and real-world creative tasks show it as close to the big Threadripper, but not ahead, remember just how huge a difference there is between the systems in price and livability.

An M3 Ultra is going to start at $4000-5000, and it'll go up to $10,000 or so with high RAM and storage - maybe $12,000 if Apple has some monster options. A full 256 GB of RAM, but with somewhat modest storage (2TB) should be around $7000. Add external SSD storage for a few hundred to a few thousand more (depending on your needs). I wish Apple offered add-it-yourself NVMe, but they don't... $10,000 buys a heck of a system.

$10,000 buys a Threadripper Pro 7995 WX, too - a bare CPU. Building a decent workstation around it is going to be right around $20,000. Maybe less if your work isn't RAM or storage intensive (. Maybe more for super configurations (that the Mac Studio may not be able to match). Just adding up the parts, it's really expensive

CPU: $10000
Motherboard: $1000 (conservative estimate - these aren't out yet, previous generation is $1000)
Case, PSU, cooling: $1000+
GPU:$2000
RAM: $1000 for 256 GB
Storage: as desired (2 TB is only a couple of hundred dollars)

Where the Mac is probably going to be about $7000 with 256 Gb of RAM and a 2 TB system/application drive, a 7995WX will be about $15000 in the same state.

A M3 Extreme (double M3 Ultra) Mac Pro (if such a thing ever exists) will compete with the big Threadripper on price, and if GB6 is correct, may well be twice as fast for tasks that don't love 96 cores...

I’d have thought the 64 core Threadripper 7980X ($5000) was more likely to be competing with the M3 Ultra. Bit unfair to compare the latter with a TR Pro chip that supports massive amounts of PCIe lanes and RAM capacity, given the AS chip has neither. PC workstations are also often based around multiple high-end GPUs, which is obviously not an option for the M3.

 

[danwells]

Yes, features-wise, the 7980X is a lot closer. It also has a multi-core Geekbench 6 score that is very close to the 7995WX (the 50% extra cores seem to barely matter at all - a few hundred points out of 26000). Neither one is quite as fast in GB6 multicore as some (almost certainly overclocked) desktop CPUs

My suspicion (judging from the fragmentary numbers that are out there) is that the scheduler on the big many-core Intel and AMD chips is VERY server-oriented. They are phenomenally powerful in workloads with a lot of independent threads, but not great at anything where the scheduling is trickier. Apple may have a better scheduler, or they may simply be less dependent on a ton of cores. The result is that the M3 Max (with 12 P-cores) is coming within spitting distance of 56, 64 and even 96-core Intel and AMD chips, which are all P-cores. A 24 P-core version will probably leave them all behind. If you ran independent tasks on all those cores, the M3 Max wouldn't come within a mile.

This makes a lot of sense, because those are all really server chips, meant for a whole bunch of concurrent users throwing jobs at them. They are REALLY fast at that, and they have the I/O for it... What's up for grabs is how fast they are at a single session in Photoshop or DaVinci Resolve.

Apple's scheduler isn't perfect, either. Looking at P-cores only, a 4 P-core M3 scores about half of what a 12 P-core M3 Max manages (if there were no scheduling issues, it would be 1/3). A 10 P-core M3 Max scores 90% of the 12 P-core version (without scheduling, it would be 83%). None of this accounts for whatever part of the benchmark is running on the E-cores (all systems mentioned above have four E-cores, presumably identical). Looking at the previous generation, a 24-core (16 P-core) M2 Ultra should score exactly twice what a 12-core (8 P-core) M2 Max does, if scheduling were perfect. It doesn't - it's around 160% instead of 200%.

I'd love to see some real-world comparisons between Macs and various forms of big PC chips that don't involve Rosetta. I think the current version of PugetBench runs some operations under Rosetta, which isn't really fair. Is there a creative app benchmark that is all native on both sides? Ideally, it would use optimized code paths, too (Metal for GPU tasks for the Mac, whatever's fastest on the PC).

Essentially no gaming benchmark is fair to Macs (unless you actually want to play games, rather than using them as benchmarks - if you do, the sad state of game code on Macs is a VERY fair thing to note, and "nothing is Mac-optimized" becomes a BIG part of your decision), both because Apple has optimized performance (especially GPU performance) around creative apps instead of games, and because games on Macs are a mess of emulations).

From the best numbers I can see, the M3 Max is elite-class on productivity applications on the CPU side (about as good as the best desktop PC chips, beaten only by overclocking and by server-type chips, both by around 20%), and is a very good laptop on the GPU side, although it can be beaten (easily) at GPU tasks by big dedicated desktop GPUs, and even by upper-midrange models. An M3 Ultra should be the best yet on the CPU side (by a significant margin), probably with GPU performance that is competitive with all but a few elite desktop GPUs. Multiple GPUs should beat it easily...