Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
What's Happening With the Mac Pro?
- Thread starter MacRumors
- Start date
- Sort by reaction score
It does, when you tell people who owns the machine that they are wrong
To prove our point, giving the iMac Pro as illustration
eshop.macsales.com
You can even double the storage
512GB Memory Upgrade Kit for iMac Pro
Get everything you need to upgrade your iMac Pro memory up to 512GB. Installation by a skilled technician is required.
eshop.macsales.com
You can even double the storage
Or you can go for a Mac Studio instead of relying on software developers supporting Tahoe in the long run?
Last edited:
You can order a 512 GB Mac Studio mate.
And to those chipping in about Xeons going to 2 TB or more.
Cool. What's the workload for it outside of virtualisation? because they've got nowhere near enough memory bandwidth to make using 2TB practical for any sort of heavy compute workload. The reason you don't see them configured that way is because it doesn't make any real world sense, most of the time you get better performance by scaling to more machines before configuring a single box with 2T.
It's like saying I can hook my 2007 Mac mini up to my 1.5 PB enterprise storage array. Cool. there's no way for it to make practical use of that much storage with 1GbE network, 2GB of RAM, and a dual core cpu from 20 years ago.
It would literally take the box 90-100 days of 24/7 100% network throughput to get 1PB on or off it, and has no ability to store that much locally, or generate that much data to upload to it in any case.
If you're only using the machine 8 hours per day, you're talking a solid working year to copy the data, assuming it isn't being copied from somewhere else or (lol) generated on device.
Last edited:
Oh yeah go price up a h100 to get even anywhere close to 128GB of VRAM (only 80GB each, so you'd need two).
If you can even buy them in that small quantity right now.
I wasn’t talking about old Mac Xeons. A modern Xeon system has orders of magnitude more system bandwidth than M chips.
They still do but no idea if anyone buy
:1. High-Speed Storage (NVMe RAID & SSD Adapters)
Cards for expanding internal storage capacity and extreme speed (up to 26,000 MB/s).- Sonnet M.2 4x4 PCIe Card (Silent or Fan version)
- OWC Accelsior 8M2 (Holds up to 8 NVMe SSDs)
- OWC Accelsior 4M2
- HighPoint SSD7749E (E1.S NVMe RAID Controller)
- HighPoint SSD7540 (M.2 NVMe RAID Controller)
2. Video Capture & Playback (Video I/O)
Cards for SDI/HDMI input and output, used for color grading, broadcast, and streaming.- Blackmagic Design DeckLink Series:
- DeckLink 8K Pro G2
- DeckLink Quad HDMI Recorder
- DeckLink Duo 2
- DeckLink 4K Extreme 12G
- AJA Video Systems KONA Series:
- AJA KONA 5 (12G-SDI I/O)
- AJA KONA X (Streaming & I/O)
- AJA Corvid Series (often used by developers/OEMs)
3. Professional Audio & DSP
Cards for recording studios, massive track counts, and specialized audio processing.- Avid Pro Tools | HDX PCIe Card (The industry standard for studios)
- RME HDSPe Series:
- RME HDSPe MADI FX
- RME HDSPe RayDAT
- RME HDSPe AIO Pro
- Focusrite RedNet PCIeR Card (Dante Audio Interface)
- Lynx Studio Technology:
- Lynx AES16e (AES/EBU Interface)
- Marian Audio: (Clara E, Clara D - Seraph series)
4. Networking & Connectivity (HBAs & NICs)
Cards for connecting to SAN/NAS storage, high-speed internal networks, or adding extra ports.- ATTO Celerity Series(Fibre Channel HBAs):
- Celerity FC-322E (32Gb Gen 6)
- Celerity FC-162E (16Gb Gen 6)
- ATTO FastFrame Series(Ethernet NICs):
- FastFrame 3 (10/25/40/50/100GbE)
- Sonnet Allegro Series(USB Expansion):
- Sonnet Allegro USB-C 4-Port PCIe
- Sonnet Allegro Pro USB 3.2 Type A
Connecting Scientific & Industrial Equipment to Mac Pro (M2 Ultra)
You hit the nail on the head. In the modern Apple ecosystem, the vast majority of scientific instruments and industrial machinery connect via USB or Ethernet.The era of installing specialized, proprietary PCIe "control cards" (common in the PC/Windows world) is effectively over for the Mac, primarily due to driver availability. Manufacturers of industrial hardware (PLCs, motion controllers) rarely write drivers for macOS, let alone for the ARM (Apple Silicon) architecture.
Here is the breakdown of how these devices connect to the Mac Pro M2 Ultra today:
1. Microscopes & Machine Vision
Gone are the days of legacy PCIe "Frame Grabber" cards.- USB 3.0 / USB-C: Most modern electron and optical microscopes send data directly via high-speed USB. The Mac Pro has plenty of ports, and you can add more via PCIe USB expansion cards (e.g., Sonnet) if needed.
- GigE Vision (Ethernet): Industrial cameras and high-end scanners often use the Ethernet protocol. This is where PCIe Network Interface Cards (NICs) come in handy—you can install a 4-port 10GbE/25GbE card to ingest data from multiple cameras simultaneously.
- HDMI / SDI Feed: If the microscope outputs a direct video signal, you use a Blackmagic DeckLink PCIe card to capture the feed in real-time.
2. Robotics, CNC & Industrial Machines
The Mac Pro typically acts as a supervisor, data manager, or design station, rather than a real-time PLC controller.- Ethernet (TCP/IP): The industry standard. The Mac connects to the machine's local network to send CAD/CAM files or monitor telemetry.
- Serial Connections (RS-232 / RS-422): For legacy machinery or lab scales. While PCIe serial cards exist, it is far more common and reliable to use high-quality USB-to-Serial adapters (e.g., FTDI chipset) or Ethernet-to-Serial servers. Finding native ARM macOS drivers for obscure PCIe serial cards is often a headache.
3. Laboratory Instrumentation (Oscilloscopes, Spectrometers)
- LXI (LAN eXtensions for Instrumentation): Modern instruments plug directly into the network via Ethernet. The Mac communicates with them via standard protocols.
- GPIB (IEEE 488): The classic lab standard. Instead of internal PCI cards, labs now use USB-to-GPIB adapters (e.g., from National Instruments), though driver support for Apple Silicon should always be verified with the manufacturer.
When do you actually use the PCIe slots for Science?
You use the PCIe slots when the data throughput exceeds what USB or standard Ethernet can handle. The slots are not for controlling the machine, but for saving the massive data it generates.- Genomics / DNA Sequencing: These machines generate terabytes of data. You need PCIe Fibre Channel cards (ATTO) to offload that data to large SAN storage arrays.
- Particle Physics / Astronomy: Radio telescopes requiring extreme bandwidth. Here, you would populate the Mac Pro with 100GbE Network Cards to capture the stream.
For a scientific Mac Pro setup:
- Use PCIe slots for: Massive Storage (NVMe/SAN) and High-Speed Networking.
- Connect machines via: External ports (USB/Ethernet). Do not expect to find internal PCIe cards for direct hardware control (motion control, DAQ) due to the lack of macOS drivers.
"Orders of magnitude". You know that means increments of 10^x yes?
Provide link for the sustained bandwidth over 512GB of memory or more.
I'm no intel website expert or product number nerd, but the highest end xeon I can find on their site has ~360GB/sec of memory bandwidth (8x DDR5-5600 channels), or less than my dinky little M4 max laptop. Less than an M1 Max even:
Intel® Xeon® Platinum 8592+ Processor (320M Cache, 1.90 GHz) - Product Specifications | Intel
Intel® Xeon® Platinum 8592+ Processor (320M Cache, 1.90 GHz) quick reference with specifications, features, and technologies.
www.intel.com
And that CPU alone costs ~$11000 in units of 1000. Emerald rapids.
Show me the intel CPU that has over 8TB/sec of memory IO (one order of magnitude more than M3 Ultra at 800GB/sec).
If you can't, you are demonstrably full of it.
The M3 ultra has a memory bus twice as wide as that Xeon and uses faster memory modules.
Last edited:
I know it’s not likely but for me the case etc don’t need updating. Only the motherboard/chips. If Apple would just update the motherboard/chips and allow user upgrade then it wouldn’t put a lot of stress on the product development team, not cost too much and keep the small section of users that need this machine happy. Maybe upgrade the shell every 5-8 years
Last edited:
If it isn't the Mac Pro, I don't see why they didn't simply release an M3 Ultra Mac Pro.
Mac Pro (2019) memory specifications - Apple Support
Learn about the types of memory (RAM) used in Mac Pro (2019).support.apple.com
Wrong, the CPU of Mac Pro itself can only support up to 1.5TB.
The smallest CPU is the W-3223, which only supports up to 1 TiB.
The largest, however, is the W-3275M, which supports up to 2.
What Apple's support pages say about RAM is never "what does the CPU/chipset/board support" but "what are we officially supporting".
I do, yes. The statement speaks for itself. M processors have very few PCIe lanes to interact with the outside world, while Xeon systems can have more than 10x the number.
Not sure if you didn’t understand, didn’t do the very quick research, or are just a die hard fanboy, but the numbers are pretty clear.
Apple make consumer grade chips for consumer devices. Intel, Nvidia etc. do that too but also make pro grade kit with vastly more capability. That’s not a bad thing, Apple are great at what they do, but if you want to get into AI and ML then they just aren’t the right place to run that workload. You can still use a Macbook to interact with a cloud platform if you can’t use other UIs
This — an upgrade kit with logic board and I/O boards would be a very fine thing.
"Orders of magnitude" suggests multiple. the Xeon 6788P has 96; the EPYC 9755 has 128. The M3 Ultra appears to have 32.
It's a wild exaggeration to call four times as much "orders of magnitude more".
I think someone else said this a few pages ago. Once you're upgrading the SoC, you probably also need to upgrade the board and RAM. At that point, the most expensive parts of the system are getting replaced anyway. It sounds good on paper, but it's rarely worth it.
I never said four times as much, but Xeon isn’t limited to one chip in a system, is it? People working with real systems know that kind of stuff.
That's true. You could put up to eight 6788Ps in one system. Then you would have 768 PCIe lanes, or 24 times as many. I still think calling that "orders of magnitude" is a stretch. It's not 100x or 1,000x.
24 vs 768 is a significant difference no matter how you cut it. Being able to plug things in to those 768 lanes is also a huge advantage if you want Converged Ethernet for stuff like storage or rDMA. Using 768 lanes for GPUs allows for much larger single instance footprints which can be very important.
Here’s one way to understand it:
[1] [tl;dr] If there had been an M3 Max Mac Studio, then there would have been an M3 Ultra Mac Pro.
[2] Every generation of Ultra has come five months after the launch of the Max. M1/M2 Ultra came five months after M1/M2 Max. M3 Ultra came five months after M4 Max. This isn’t because Ultra takes five months longer to produce, it’s because the release of the Max has to be staggered — Apple can’t produce enough Max to launch both the MacBook Pro and the Mac Studio at the same time.
I’m not going to speculate on *why* there was no M3 Max Mac Studio (and thus no M3 Ultra Mac Pro), it doesn’t matter, it’s just a fact. Same goes for the probable reason(s) why Apple built M3 Ultra instead of M4 Ultra.
[3] Assuming M5 Ultra comes in June 2026, five months after the M5 Max in January 2026, then that will be a year and a quarter after the M3 Ultra launched in the Mac Studio alongside the M4 Max. That’s an acceptable cadence for the Mac Studio, but not for the Mac Pro. Instead, assuming the Mac Pro is not dead, we will get the M5 Ultra Mac Pro exactly three years after the M2 Ultra Mac Pro.
Last edited:
Well, the M4 Max, yes. But compared to the M3 Max, it came five quarters later, when previously, that delay was only one quarter.
Right.