Apple Unveils New Mac Pro With M2 Ultra Chip and More

[skippermonkey]

I have a 2019 Mac Pro with three GPUs in it. There‘s no way the Mac Studio Apple Silicon competes with this for GPU rendering (at least, I doubt it - I’ve yet to find definitive benchmarks.) And now the new Mac Pro doesn’t support discrete GPUs at all. But my system also includes an SSD and two enormous hard drives, one for assets plus a Time Machine backup. So, thanks to Apple, neither the Mac Studio nor the Mac Pro is a viable alternative (unless I spend £7K on the top-end Studio and then spend even more on external storage.)

I just checked out the specs for the M2 Ultra and the baseline for GPU rendering/interaction etc is the Radeon Pro W5500X - a 13-year-old card that you can pick up today for less than £250. The card should never have seen the inside of a Mac Pro in the first place - being 4x or 6x faster is nothing to brag about. Apple is so cynical when it comes to its marketing.

 

[Boil]

I have a 2019 Mac Pro with three GPUs in it. There‘s no way the Mac Studio Apple Silicon competes with this for GPU rendering (at least, I doubt it - I’ve yet to find definitive benchmarks.) And now the new Mac Pro doesn’t support discrete GPUs at all. But my system also includes an SSD and two enormous hard drives, one for assets plus a Time Machine backup. So, thanks to Apple, neither the Mac Studio nor the Mac Pro is a viable alternative (unless I spend £7K on the top-end Studio and then spend even more on external storage.)

With three GPUs in your 2019 Mac Pro, one can only assume the SSD & "two enormous hard drives" are using the internal SATA ports with a Sonnet bracket/wiring...?

The 2023 Mac Pro has those same SATA ports and that same drive cage hookup capability, you can just move your SSD/HDDs over...

 

[ectoplasmosis]

I get it. However, one of today's great mysteries is where the new Mac Pro gets its PCIe lanes from and what the actual bandwidth and number of available lanes is for those slots - so far all we know is that the physical size of those slots. The Sonnet adapters offer 16x and 8x slots but (as you pointed out) that doesn't mean they get 16x or 8x bandwidth.

As far as we knew, the Mx Ultra's i/o consists of enough PCIe to run the ethernet, SD slot and USB-A ports plus 8 TB4 ports. Either the M2 series has been keeping a shedload of unused PCIe lanes under a bushel or those PCIe slots are somehow sharing bandwidth with the TB4 ports. The "worst case" solution would be that Apple have just built a TB4 to PCIe bridge into the MP...

Yep, I’d love to know this too.

I have a sinking feeling, based on the images of the mainboard, that they’re using a PCIe multiplex chip and the aggregate bandwidth is shared between the slots and TB4 ports.

Otherwise, the chip will need to support 128 lanes to feed 6x 16-lane slots and 8x 4-lane ports. Unless it’s a new revision in a larger package with loads more pins, then bandwidth will be gimped.

 

[Boil]

Yep, I’d love to know this too.

I have a sinking feeling, based on the images of the mainboard, that they’re using a PCIe multiplex chip and the aggregate bandwidth is shared between the slots and TB4 ports.

Otherwise, the chip will need to support 128 lanes to feed 6x 16-lane slots and 8x 4-lane ports. Unless it’s a new revision in a larger package with loads more pins, then bandwidth will be gimped.

 

[skippermonkey]

With three GPUs in your 2019 Mc Pro, one can only assume the SSD & "two enormous hard drives" are using the internal SATA ports with a Sonnet bracket/wiring...?

The 2023 Mac Pro has those same SATA ports and that same drive cage hookup capability, you can just move your SSD/HDDs over...

The drives aren't an issue; the new Mac Pro doesn't support discrete GPUs (I have an RX5700 XT and a W6800X Duo MPX unit). So it's useless for me.

 

[Spectrum]

What is the point of this product? It seems bizarre to have alongside the MacStudio.
I mean: What is the use case compared to the MacStudio?
What kind of PCIe cards would it take? The prior-gen Afterburner ones seem to have been mothballed too.

Surely this thing can in no way compete with a discrete multi-GPU-solution? And the RAM limit is crazy low too for those that need it.

 

[ectoplasmosis]

View attachment 2213588

Thanks; this doesn’t, however, address whether the TB4 ports share bandwidth with the PCIe slots…

 

[ectoplasmosis]

What is the point of this product? It seems bizarre to have alongside the MacStudio.
I mean: What is the use case compared to the MacStudio?
What kind of PCIe cards would it take? The prior-gen Afterburner ones seem to have been mothballed too.

What is the point of this product? It seems bizarre to have alongside the MacStudio.
I mean: What is the use case compared to the MacStudio?
What kind of PCIe cards would it take? The prior-gen Afterburner ones seem to have been mothballed too.

Surely this thing can in no way compete with a discrete multi-GPU-solution? And the RAM limit is crazy low too for those that need it.

Failure of imagination much? There are many examples of real-world use cases that are opened up/enhanced by being able to use 8-lane+ PCIe cards.

As already mentioned, our current real-world use case involves multiple 8-lane video capture and 25Gb/s Ethernet cards, both of which are crippled when used in Thunderbolt PCIe chassis with Mac Studio. We will be buying new Mac Pros to replace this setup.

 

[JippaLippa]

Pretty cool and unavoidable, given apple's direction with the M chips.
The only thing I don't understand is the ram...
Not the fact it cannot support more than 192GB natively, rather the fact there seem to be no way of increasing it.
It's pretty weird to go from a machine that was cabale of holding 5 times the ram, to one limited ti 192GB (which is a lot, sure, but not the most one could want).

 

[sharpie000]

It’s hilarious how many people were disappointed when Apple kept postponing the announcement of an Apple Silicon Mac Pro and when it finally happened, people are still complaining.

I was concerned about my 2019 Mac Pro, but after reading a number of comments in this thread, I feel more confident that my Mac will last me for at least a few more years with its upgradability options.

 

[ThomasJL]

Lol @ everyone thinking Apple wouldn't want Studio and Mac Pro competing against eachother. The specs are basically identical, aside from drastically different pricing for PCIe (v4 no less). Makes the studio look like a BEAST in that package.

This just shows how bold Apple is...they dgaf.

The mental gymnastics that Tim Cook supporters will do to justify his mediocrity are pathetic.

 

[jazzerd8]

Apple today updated its Mac Pro desktop tower with the all-new M2 Ultra chip, which features a 24-core CPU, up to a 76-core GPU, and support for up to 192GB of memory. Apple says the new Mac Pro is up to 3x faster than the fastest Intel-based Mac Pro.

The new Mac Pro has the same overall design as the previous Intel-based model. However, the tower is now equipped with eight Thunderbolt 4 ports instead of four, two higher-bandwidth HDMI ports that support up to an 8K display at 240Hz, seven PCIe expansion slots, dual 10 Gigabit Ethernet ports, and a headphone jack that supports impedance headphones. The computer can be configured with up to 8TB of SSD storage.

Like the new Mac Studio, and the latest MacBook Pro and Mac mini models, the Mac Pro now supports faster Wi-Fi 6E and Bluetooth 5.3.

The new Mac Pro is available to order starting today, and it will begin arriving to customers and launch in stores on Tuesday, June 13. In the U.S., pricing starts at $6,999, whereas the previous Intel-based Mac Pro started at $5,999. A rack-mounted version of the Mac Pro remains available from $7,499.

Article Link: Apple Unveils New Mac Pro With M2 Ultra Chip, Extra Ports, and More

Been anxiously awaiting news of the Mac Pro for almost a year since my 2017 retina iMac is beginning to have trouble with certain functions. To my surprise we saw the Mac Pro announced, but to my dismay you can’t add RAM yourself, which is where the savings is.

So now I am likely going with the maxed out Mac Studio but it is so expensive man! And it pisses me off that Apple screws you by not including ANYTHING with it, not even a freakin mouse! Apple abandoned target display mode so we can’t use the 2017 iMac as a monitor either, so there’s more $$.

I paid $3k for a pretty maxed out iMac in 2017 and wish they had another 27” iMac now bc this one will be $7k plus the monitor. Crazy $, and yes I need it for production work. So I’m conflicted both ethically and financially and also concerned they will release the iMac or a cheaper alternative in 6 months. I know the Studio will be blazing but damn, we are getting nailed and blackballed by Apple continuously upselling by way of discontinuing.

What are you all doing, esp people in my situation? Thanks!!

 

[DearthnVader]

I guess if you have a scientific need you can daisy chain enough of these together to get 1.5TB of RAM.🤣

 

[tny]

> As a Webdeveloper 32 GB RAM is the bare minimum these days. I can’t image people working with 8K+ video and RAW phots having enough RAM.

As a very experienced backend developer who handles a lot of data and has done front-end work, your dev box should have no more than 16 GB, unless you're running multiple containers for some reason (if you are, you should be using a cloud provider for that), or you're a designer using Adobe or similar image /video editing / composing products. If your dev box is too powerful, you won't notice that you have written a giant RAM sink that will crawl on your users' machines.

 

[Anaxarxes]

I have a 2019 Mac Pro with three GPUs in it. There‘s no way the Mac Studio Apple Silicon competes with this for GPU rendering (at least, I doubt it - I’ve yet to find definitive benchmarks.) And now the new Mac Pro doesn’t support discrete GPUs at all. But my system also includes an SSD and two enormous hard drives, one for assets plus a Time Machine backup. So, thanks to Apple, neither the Mac Studio nor the Mac Pro is a viable alternative (unless I spend £7K on the top-end Studio and then spend even more on external storage.)

I just checked out the specs for the M2 Ultra and the baseline for GPU rendering/interaction etc is the Radeon Pro W5500X - a 13-year-old card that you can pick up today for less than £250. The card should never have seen the inside of a Mac Pro in the first place - being 4x or 6x faster is nothing to brag about. Apple is so cynical when it comes to its marketing.

Where did you read for a fact that the new MP does not support external gpus?

 

[theluggage]

Otherwise, the chip will need to support 128 lanes to feed 6x 16-lane slots and 8x 4-lane ports.

(You probably already know this but - for the discussion Bear in mind that the Xeon-W in the 2019 Mac Pro only had 64 PCIe lanes - you probably won't find a machine that can give full dedicated bandwidth to all of its slots at the same time - PCIe equipped Mac Pros have always had an expansion slot configuration utility to tune how the bandwidth is allocated to the slots (https://support.apple.com/en-gb/HT210104 - be interesting to see if that document appears for the new Mac Pro).

View attachment 2213588

Yup - and this Thunderbolt-to-PCIe expansion box from Sonnet is advertised as having one x16 and two x8 slots.

Echo III Desktop (3 PCIe slots for full size cards)

Desktop Thunderbolt 3 adapter box with three PCIe expansion slots for audio, video, fibre, RAID and ethernet PCIe cards.

www.sonnetstore.com

...except in reality they're all sharing a single x4 pipe from the Thunderbolt port.

Worst-case scenario is that Apple have just thrown in a couple of Thunderbolt to PCIe bridges and what you're effectively getting is two of those Sonnet adapters each driving one x16 and two x8 slots... Just spitballing and I don't think Apple would be that lame, but that math is horribly compelling!

 

[Apple Knowledge Navigator]

Personally, I think they (Apple) wanted to kick-off with M3, but felt that the time gap was too much between summer and winter.

I’ve no doubt M3 Ultra will be a major upgrade.

 

[toke lahti]

Maybe but I doubt it - LPDDR5X should allow them to do 256 GB next year, and if the M3 Extreme can be realized (and not cancelled like the M2 Extreme) hopefully we will see up to 512 GB… NVIDIA doesn’t offer expandable memory on their grace hopper super chip - instead they mix LPDDR and HBM but they too are limited in the max amount of LPDDR.

I'd say it's not about mac RAM, it's about that you have to buy max RAM if you are not sure that you DON'T need it.
Is Nvidia becoming a player in workstation business?

 

[kemal]

Notice that when they went through the list of PCIe card types the could be plugged in, graphics car was not mentioned?
Could be there are no cables from the power supply to power the card and it might be the PCIe lanes are multiplexed to something like four lanes, which is pretty slim for a modern card.

 

[CWallace]

Notice that when they went through the list of PCIe card types the could be plugged in, graphics car was not mentioned?
Could be there are no cables from the power supply to power the card and it might be the PCIe lanes are multiplexed to something like four lanes, which is pretty slim for a modern card.

The actual reason is that, at the moment, either the M family of SoCs does not have the necessary hardware to interface with an off-SOC GPU or Apple does not wish to write - or allow third party - drivers to support non-Apple off-SOC GPUs.

 

[mrothroc]

You know what, you're absolutely right about all that. It's a shame that some folks out there think those things ain't a big deal or "not important." Yeah, sure, we can argue all day about whether anyone really needs that much RAM or VRAM. I mean, I vaguely remember someone in the Stone Age saying that 16K of memory was more than enough (was it actually 16K?). But regardless of the viewpoint, my observation of Apple remains unchanged. I believe they're falling behind the trends of the times, especially in the AI/ML field. Now, I can't quite say it's sunset time for them yet, but AI/ML is definitely a rocket that's shooting up and has already reached great heights. And for some known or unknown reasons, Apple is clearly struggling in the hardware department. From Nvidia's perspective, the A100 specs are already outdated, but even with that kind of product, Apple can't even keep up, let alone compete with the H100 or the popular RTX in terms of computational power for AI/ML.

Yeah, if we don't focus on that aspect, then whether it's the M1, M2, or M3, it hardly makes a difference. As long as Apple doesn't intend to hop on that train, it won't have much impact. But hey, I'm not gonna write them off completely. They've never been the first leg in a relay race. Also, I still have hope that Apple will step up their game in their own GPU and AI/ML computation. Because instead of waiting for macOS to reintroduce compatible Intel/Nvidia/AMD AI/ML accelerators, it's better to wait for Apple to make some breakthroughs. Well, as long as folks still want to comfortably use macOS for work and develop AI/ML, okay, I guess we can make do, especially since the house we're living in is pretty cozy.

This is a good discussion, but I want to point out:
1) The H100 doesn't exist yet. It is still in development. There are issues with sanctions on China that may keep it in development for a while.
2) Once the H100 is finally shipping, there won't be enough to go around. The cloud providers are going buy them in large quantities, so no one else is going to get them for many, many months.
3) The A100, which is what they are currently selling is on back order. Again, if you order one now, you probably won't get it until next year.

And, just to make sure we're comparing the right things, how much does an A100 with 80GB cost (if you could get it) vs. an M2 Ultra with 192GB (imminently shipping)?

(Edit to add an answer: A100 80GB new is $14-17k on eBay right now.)

 

[avkills]

The actual reason is that, at the moment, either the M family of SoCs does not have the necessary hardware to interface with an off-SOC GPU or Apple does not wish to write - or allow third party - drivers to support non-Apple off-SOC GPUs.

I do not think it is a communication problem since Intel laptops had GPUs on the CPU package and the laptops also discrete GPUs. The software foundation is in place already; although there could be a larger switch that prevents that if Apple Silicon is present.

I for one will not be doing an upgrade; going to wait to see what M3s bring.

 

[CWallace]

I do not think it is a communication problem since Intel laptops had GPUs on the CPU package and the laptops also discrete GPUs.

Yes, because Intel designed that functionality into their CPUs. Apple very well could have chosen to not design that functionality into theirs.

 

[smulji]

Personally, I think they (Apple) wanted to kick-off with M3, but felt that the time gap was too much between summer and winter.

I’ve no doubt M3 Ultra will be a major upgrade.

I'm not expecting M3 to come out until Spring 2024 at the earliest

 

[smulji]

Notice that when they went through the list of PCIe card types the could be plugged in, graphics car was not mentioned?
Could be there are no cables from the power supply to power the card and it might be the PCIe lanes are multiplexed to something like four lanes, which is pretty slim for a modern card.

That's because RAM and GPU can't be upgraded after the fact.