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.
Like it or not, Apple needs Mac Pro.
- Thread starter 1d1otic
- Start date
- Sort by reaction score
maflynn
macrumors Broadwell
The OP is postulating that the apple needs a mac pro computer in the line up, and my question pertains to that. Apple is not returning to x86, so a comparison to xeon/threadripper only makes sense imo, is that apple comes out an Extreme version of the apple silicon which they were rumored to have been working on a few years ago, and that silicon has an architecture more suited for workstation usage.
keksikuningas
macrumors regular
would love to see some real life benchmarks for this, but it seems that M3 Ultra studio is the most powerful ARM workstation? Does anyone know if M chips are useful in embedded (automotive etc) ? I assume all of that runs on Linux?
Highlighting automotive cause I know the Ampere ARM workstations were marketed towards automotive sector
This is wrong on so many levels it will take a day to explain. How much GPU RAM does 4070 or 5090 support compared to Apple silicon Mac Studio? Apple isn’t going to go after high end Nvidia data center chips. It’s well placed in the workstation place that needs lot of GPU memory. And if Apple does need to compete in workstation markets, 5090 or even 4070 are’nt really workstation class GPUs according to Nvidia.
Last edited:
So wrong.
GPU uses VRAM, not RAM. While it is an advantage over PC, does it really gives advantage for workstation? No. Besides, despite having a lot of space, the bandwidth is slower than RTX high-end GPU and workstations. The space is not everything.
How does Mac Studio compare to RTX 5090? It only performs up to RTX 4070TI mate. By your logic, how will it replace workstations when the GPU level isn't even near RTX 5090? How ironic.
The real question is will Apple release an MK Ultra version?
Apple Knowledge Navigator
macrumors 601
No, they dont. After the transition and Mac Pro 2023, they no longer use them.
maflynn
macrumors Broadwell
So wrong
There's a number of GPUs that use dram, including apple. Also people can use the term ram, to mean either dram, or vram, depending on the context. You missed that, I guess
Attachments
Storage extension is a big plus for Mac Pro if you need high end local storage. But then AMD intel provide more bang for the buck if storage is main criteria.
carry on! Ignorance is bliss. Good luck running models that need more than 32 GB on GPU. 5090x is useless when it runs out of memory.
Unfortunately there are no mid tier workstation solutions, and Nvidia gets crazy expensive as you start pushing higher memory requirements. I wanted to like spark, but it is like a beta product. Mac Pro is not gonna make AS faster at fp8.
You have lot more patience than me. With unified memory it doesn’t really matter as they are becoming interchangeable.
There are many tasks for GPU and yet, you only picked LLM and then say Apple is the best. Since when Apple is well known for AI? Ironic.
That’s your problem, I don’t even use it for LLMs. If you wnat to run LLM they are much cheaper cloud options than blowing money on Apple silicon or Nvidia.
If so, why not make Mac Pro then? You are justifying the current situation as if 512GB is the biggest that LLM can go. You logically failed.
maflynn
macrumors Broadwell
I asked this already, what can a mac pro do that the studio cannot? You cannot use discrete GPUs, there's no NVMe expansion with apple silicon, no ram upgrades.
Last edited:
I don’t care about LLM. Not sure what logic you are talking, it makes no sense when LLM is not even in the conversation nor I use them on MBP. Mac Pro outside of extending storage or additional Video cards isn’t any different than Mac Studio. honestly that market has moved on long time ago. My last Mac Pro was 2012, trash can might have just killed Mac Pro.
HouseLannister
macrumors 68040
Mac Pro is a retail product for the retail channel. It doesn't sell well, so it won't be updated. That has nothing to do with their chip plans or what they design for their own internal use needs. If they need to make faster chips to keep up with AI, then they will need to do that. But having a poor-selling retail product to take up inventory space in a warehouse isn't going to drive innovation--the needs of the demanding AI will. Having a product that only gets updated every few years is actually a hinderance to innovation. Why make something better when you have a stockpile of perfectly decent ones that you haven't sold through yet? Things like the iPad Pro or Macbook Air/Pro, which get revised every single year, keep the pace of those improvements much higher than the Mac Pro ever could.
I'm pretty sure we've been under mind control ever since 1984.
I think one has to see the bigger picture here.
Macs account for less than 10% of Apple's revenue - and that's including all Macs, i.e. MacBooks, iMacs, Mac minis etc.
The "workstation" offerings will be a small share of that already small share.
In laptops, for most users, power efficiency will be more important than absolute performance. And yes, I can build a PC more powerful than a Mac Studio for a fraction of the price. And I would, if that was my use case (in fact I make sure most software I use could run on Windows or Linux as well).
But I cannot get a laptop with the same power *and* the same power efficiency and low noise. So for those systems, ARM makes sense.
Now the thing is, by using the same architecture across the board, from Apple Watch through iPhones to displays to Mac Pros, Apple is able to benefit from huge scale effects, both in terms of engineering and also manufacturing.
To have a separate architecture like x86_64 just for that tiny workstation segment simply won't make sense to them.
A same story for others making workstations so not a good excuse.
Although I agree with what you are saying conceptually, your argument fails to make sense.
• It is a false statement that "...Macbook Air/Pro, which get revised every single year." Mac laptops are not revised, they simply get new chips every single year, and a Mac Pro could also easily get new chips every single year.
• Working with tech does not mean we need to buy new innovation every year! A product like Mac Pro in particular is expected to last for years, just like MacBook Pros last me for 5-7 years.
• Mac Pro has been expected to be expandable and failed when it was not expandable. In the past we added HDD, RAM, etc. My personal feeling is that if Apple is to make a really viable MP it will be by somehow addressing the 2026+ era expandability issue.
• IMO Apple still might make a reasonably useful MP even without effectively addressing expandability by making it capable of super power. E.g. like the M3 Studio Ultra on steroids. An M5 Mac Pro Ultra-Ultra ;~) with a TB or more of RAM might be cost effective for a subset of LLM models or other AI needs.
There’s still a VERY prevalent lack of understanding as to how Apple Silicon is different from an Intel or AMD or Nvidia solution. As a result, you’ll still get folks assuming that a discrete solution (over a bus FAR slower than what Apple uses currently) would yield better performance.
Yup. Plus the extra clock cycles required in the AMD, etc. solutions.
257Loner
macrumors 6502a
If Apple can't figure out how to efficiently run Edge AI from their M-series chips, like with the new M5, which has a dedicated neural accelerator in each GPU core, then local processing and all of Apple's chips will be made obsolete as processing is taken over by cloud computing powered by AI.
maflynn
macrumors Broadwell
That's a bold statement, so if local LLMs aren't feasible on Apple Silicon, you're saying that it (apple silicon) will largely fail and be cast aside?
I wonder what the percentage of people buying Macs are doing so for AI purposes - I suspect a tiny percentage, but I have no way to prove that. I also think the AI boom may be facing a reset (bubble bursting as some call it).