How about making a guess as to how many people you think you are discussing? My guess is they sold under 100,000 a year total, and if 1% of those had more than 192GB I would be shocked (that is probably under 4,000 people).
Inflation adjusted the 2019 machine was more expensive with less RAM and SSD.
I know in the studios and visual effects houses with whom I work, there are already more 2023 Mac Pros on order day one than were purchased over the whole life of the 2019 Mac Pro.
In the big picture, there are 8 billion people in the world. The vast majority of those aren’t interested in the Mac at all. Of the few that are interested in a Mac, the vast majority wouldn’t be interested in a Mac Pro.
I hope those studios and visual effects houses know they can't put GPUs in those PCIe slots… I wonder just how useful the machine is to them without multiple GPUs powering simulations, real-time previews, effects compositing, rendering etc. I'd love to see a 2023 M2 Ultra up against four decent GPUs in a Mac Pro 2019.
Wow, thanks for letting me know! None of us knew that Apple has not supported eGPU or non-Apple GPUs on Apple Silicon. First time we have all heard that. I wish we knew that before we bought all the Mac Studios we have been using for the last year.
It will do great as having all that unified memory and a solid Neural Engine will do just what we need it to do. If you can find us a GPU that has as much memory for our models, that will ship in the next week and cost less than a fully configured Mac Pro, please let me know.
With proprietary, I meant that you couldn't use them on a PC.
Exactly, the people that bought macs because needed lots of RAM are now out of the waters.
4 years later, today 64 GB RAM is cheaper than 32 GB RAM of 2019. Same for storage.
I doubt, if they had the intentions to use / support non apple GPU, they'll already announced it.
We don't know how many 192GB+ RAM Mac Pros they sold, you are right are not many, but even if there was just one guy, that poor guy has to move to PC now.
More upgradable, more future proof.
That is most likely because they knew the 2019 Mac Pro was going to be the last Intel Mac, so they waited for longer to get a mac pro with an apple chip.
If it was an insignificant number of people, it is, by definition, insignificant.
More upgradable, clearly not future proof. It had PCIe 3.0 slots, about half the speed of the PCIe 4.0 slots in the current machine. This is the problem with this future proof argument. If performance matters, then these old machines get less and less interesting very quickly. Maybe home users want 4 year old machines, but no studio, visual effects or post production facility keeps machines longer than 3 years unless they are forced to do so (like no replacement has shipped).
Really? They knew that before it was even announced? I wish they had told me! However, given that you are arguing that it is more upgradable and future proof, should they not want to be upgrading them and keeping them around since they were future proof? I am a bit confused.
Still that insignificant amount of people that needs more than 192GB are pissed right now.
I will ignore your sarcasm (it being the lowest form of wit and all that). I’m genuinely intrigued to know what it is you do and how well the current Mac Studios perform. What models are you using? I have a W6800X Duo, and need to know how the M2 Ultra will perform before deciding which upgrade route to take. I could just throw in a second Duo…
Which is insignificant. Apple does not now, nor has it ever made systems that served the needs of all possible users. This has not changed.
Given that the only cards we have in our 2019 systems are video IO, 25Gb/s Ethernet (about to upgrade to 100Gb/s Ethernet), and OWC 8 M.2 storage cards, you have made it clear why we do not think that 4 year old systems are “future proof”.
Again, you seem to want to have it both ways. You argue that these Intel machine are better and more future proof, but than in the same breath say that people did not buy them because they knew they would be obsolete. Which is it?
Serve one for what? Will it run macOS and our rendering pipeline? You have a strange conception of how companies use computers. We buy machines in specific configurations for specific tasks. There is no one who is tasked with trying to find uses for machines that have been obsoleted.
How would have an old, slow, unsupported linux system help us in any way? That is not how real companies work. We do not even sell these systems off for liability reasons, but simply remove the disk drives (where we can or erase them where we cannot) and then scrap them.
No, that is your definition. Again, none of the studios, post facilities or visual effects houses with whom I have worked over the years upgrades machines in place. You are describing how hobbyists use machines, not how professions use them. My definition of future proof is having enough headroom to be able to run any software that needs to be upgraded over the 3 year life of the machine. By that definition, everyone one of these machines is future proof.
Not future proof for your needs, but Apple silicon Mac Pros will have an even lower future proof time, at least the first one. Those computers come with PCIe 4 not 5, and hopefully each PCI express port doesn't shares bandwidth with the rest.
I'm arguing that those intel machines have better RAM capacity.
Obsolete for your needs, but for another person will find those second hand computers a good use.
Not you, but some will reuse that old computer for a second life out of service.
So by your definition, if you need more than 192 GB RAM what does that means?
Again, no. They will do what we need them to do when we buy them and will continue to run the software we need through the three years we expect to keep them. That means they are future proof.
No. I made clear that we do not replace/upgrade hardware in existing machines, so we would not swap cards a year or two from now. The cards we use right now support PCIe 4. In a year or two, Apple may have a new Mac Pro, with PCIe 5 and we may be able to buy cards that take advantage of it. At that point machines we buy will use that, however, like most actual businesses, we would not try to replace cards mid cycle. You model is how hobbyists work, not how real professionals work.
Which matters not at all to me. They have more RAM that I do not need, nor would I buy, but lack Neural Engines and Unified Memory. Those are much more important for me than a spec that is irrelevant to me. Our 2019 Mac Pros were configured with 96GB, as that was all that our testing showed benefited our applications. That we could have spent $50,000 and bought machines with 1.5TB was irrelevant.
Which matters not at all to me, as we do not sell our old machines, we scrap them. Their theoretical use by some future theoretical user that would have no impact on our use.
That is not the common practice for businesses. Again, you are talking about hobbyists, who are not the target market for these machines.
It means that I would not buy these machines. Just as if I needed to run Windows, or have 4 PCIe x16 slots I would not purchase them. However, given that the max RAM one can buy for these machines is 192GB, it seems unlikely that over their three year life span, any of the software we use would require more than that. It is possible that some future version might be able to take advantage of more, but it would not require it. That means that these machines as purchased would continue to meet our needs for their full three year life span.
The thing about this is that businesses don't want computers, they want the thing that computers produce. At my business, I don't need computers, I need CPU and RAM hours. That's what I buy. If they can be provided within my latency constraints, I don't care where they are. There are many of us in the software business who work this way, which is why the cloud providers are able to be successful.
Absolutely true. This is reason that we do no upgrade mid cycle. We purchase machines with support contracts and would not do anything that would void those contracts, nor make them useless (like adding third party RAM - where for a pro it does not void one’s AppleCare+, it needs to be removed for support by Apple, making it impossible to test).
I am guessing that you also need more just sitting in from of a screen typing hours than you need developer container hours, making it make sense to have a reasonable laptop/desktop for your developers, but not a small backend local server farm.
I tried to buy thrust hours for scenes in some of my projects but I could not figure out how to get them delivered inside our renders, so I went back to just simulating them. :-D
Basically correct. We look at Teradici and its equivalents on a regular basis, but so far they do not meet our needs. One day maybe.
Very much this ^. An article I like on the topic (one can ignore past the explanation of the analogy and its origins).
That is a very brutal evisceration of the PCIe layout.
social.treehouse.systems
Going by past history, 2026 is the earliest I would expect to see the next update.