People forget that there are plenty of ARM systems that have PCIe slots/lanes. Even my synology NAS with an ARM chip has a PCIe slot. People need to relax. I am pretty sure they will add PCIe slots in the newest Mac Pros. The only thing I am not sure about is the memory situation since Apple Silicon has the RAM baked into the SoC. However there is already DDR5/DDR6 available via Ram Sticks, maybe they'll figure it out.
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.
MP 7,1 Mac Pro 8.1 - WHAT IF...
- Thread starter maikerukun
- Start date
-
- Tags
- 7.1 mac pro next generation
- Sort by reaction score
I never once disagreed with you lol. You came in here tripping cuz I posted an article about the w6800x duos minding our own business and threw a fit about us being satisfied with the 6800 duos lol.Dude, just put down the RedBulls for a while a chill out.
The 4090 is 27% FASTER than a W6800X Duo. Two 4090s would be, according to your logic, 54% faster than your two DUOs.
If you're just talking about Apple's fastest, versus Nvidia's fastest, and comparing the price of entry, then we're talking about $3200 total retail for two 4090s, and $10,000 retail for two DUOs.
The simple fact of the matter is that you're in no way getting better performance for less money on the Apple/AMD side.
As I said, getting a $12k Puget system with 2 4090’s in it. Nobody here dislikes the 4090’s Lolol. You’re the one that needs to calm down.
Also I was comparing against the A6000’s but that doesn’t matter. Sweet dreams ❤️
Apples crack team has struck again! 😂😂😂 But honestly, that’s an awesome theory and it would be so cool if Apple took old footage from a previous keynote and zoomed in on it and revealed it to be the 8.1😊✨What if the answer is hiding in plain sight?
View attachment 2166051
What if those four beautiful, wheeled, unplugged 7,1's on Johny's floor aren't 7,1's at all?
What if they are prototypes of the new, nuclear powered Mac Pro 8,1?
No power cord required and enough carbon dioxide-free juice to effortlessly drive any GPU combination a user can imagine.
And, it will light, heat and cool your house at no additional cost!
My sources tell me that Johny's group has it ready to ship. The only holdup is Apple Legal.
In typical namby-pamby lawyer fashion, they want all sorts of warning labels covering radiation, nuclear waste, no-proliferation, blah-blah-blah. What a bunch of killjoys.
I saw that as well the other day. I hope the GPU solution can at least match what we already have with the 7.1 tho
What if it's just a Studio but with more PCIe lanes for Thunderbolt attached compute modules; CPU and GPU, or both of them integrated in one module, with a price accordingly of course. Compute modules run some iOS version tuned for sharing the module resources efficiently with master. Modules are not able to work independently standalone, they would always need the master they are attached to.
Wouldn't this be very much an Apple-like solution to an already solved problem of PCIe cards in internal slots ?
I'm not saying I'd want this, but I dug my archives of about some 8 years ago, and found it visualized.
Wouldn't this be very much an Apple-like solution to an already solved problem of PCIe cards in internal slots ?
I'm not saying I'd want this, but I dug my archives of about some 8 years ago, and found it visualized.
I think when evaluating these concepts, the question to ask is: what benefit does this provide the user, compared to established approaches? If the answer is nothing, the product has no future.
Apple won't release anything that can't make a compelling case for itself, at least to some subset of users. If it appeals to no-one, people will just buy Studios, or PC worstations.
I guess a workstation that is competitive with the PC equivalent, but can run macOS, is the baseline requirement. If this can be implemented with some kludge of M-series tech, then that could work. What we don't want though, is something that initially achieves parity / superiority, but then falls behind the relentless progress of the PC space. Whatever concept Apple choses, it needs to be based on firm foundations for the future.
Apple won't release anything that can't make a compelling case for itself, at least to some subset of users. If it appeals to no-one, people will just buy Studios, or PC worstations.
I guess a workstation that is competitive with the PC equivalent, but can run macOS, is the baseline requirement. If this can be implemented with some kludge of M-series tech, then that could work. What we don't want though, is something that initially achieves parity / superiority, but then falls behind the relentless progress of the PC space. Whatever concept Apple choses, it needs to be based on firm foundations for the future.
Efficiency is where Apple Silicon can provide benefit to the user. An nVidia DGX A100 uses 6500W - which is not going to fly for a consumer workstation. Even a PC with 2x 4090s would cause power issues for a lot of homes / offices.
Just going off raw wattage numbers Apple could put 10+ M2 Max chips into a Mac Pro with the same PSU as the 7.1, which would be absolutely insane.
I really don’t think we’re going to see things connected by thunderbolt though. Even PCI-e isn’t fast enough for lots of distributed computing workloads which is why you have connections like nvLink and infiniband. A thunderbolt network would be very limited in what tasks would work well on it.
Just going off raw wattage numbers Apple could put 10+ M2 Max chips into a Mac Pro with the same PSU as the 7.1, which would be absolutely insane.
I really don’t think we’re going to see things connected by thunderbolt though. Even PCI-e isn’t fast enough for lots of distributed computing workloads which is why you have connections like nvLink and infiniband. A thunderbolt network would be very limited in what tasks would work well on it.
That's one of the most important technical advances of what Apple has been able to do with their custom Silicon. Keeping power consumption so low while offering so much power. Technically it means they can take the leash off of their chips in a Mac Pro config and let the thing really fly...it may make all the difference in the world.
Word on the street is we will be seeing something this month. Let's see how right and/or wrong we all have collectively been over the past few months
so, if we take the stance that we were apple and we oc knew the perf of the current macpro when it comes to compute. What would be the easiest way forward? That still would not brake the new paradigm of unified mem?
To me it seems that just stopping at multiplying a laptop chip wouldn’t cut it right now or would scale well over time.
As we have seen, not even a theoretical “extreme “ chip would hold a candle to to 2 year old dual duo.
Could it still be the solution choosen? Sure, maybe Apple just sheds the high end temporarily and goes for luxury mid range until they have something that is worthy of succeeding thr current
mp2019.
But that would mean dropping a lot of faith and momentum.
What about a custom highend chiplet design with more omph in the Gpu department?
Sure, if there would be benefits to other markets like a cloud server or more powerful mid range desktops. Or maybe just as an experimental and evaluation platform for features that will be coming to later M* gens.
What about just putting in 3d party accelerators? Basically 7900 duos mpxs? To me this seems like the simplest most straightforward way forward but braking the benefit of the simplicity of the “unified” paradigm.
Damn it, I can’t figure this out!
I just want 20+ Cpu cores and then about 300 gpu cores, 192 Gb unified ram, a ssd raid on pcie5.
I also want to be able to upgrade those gpus once to land on 2x the perf wise when m3 m4 gen is out.
And that new xdr screen that matches the mbp quality.
I also think people should be able to get 40+ cpu cores, 600+ gpu cores, 384 or more Ram etc.
And alos important, a music person should be able to get all the cpu but none of the gpu. As well as a lot of ram for immense audio libraries.
You know what? I actually think apple is smart enough to do give us all what we want and that they will do that. I am more worried about price since I actually compare to what is available on the PC side.
So boring that nobody is dropping any juicy rumors anymore.
To me it seems that just stopping at multiplying a laptop chip wouldn’t cut it right now or would scale well over time.
As we have seen, not even a theoretical “extreme “ chip would hold a candle to to 2 year old dual duo.
Could it still be the solution choosen? Sure, maybe Apple just sheds the high end temporarily and goes for luxury mid range until they have something that is worthy of succeeding thr current
mp2019.
But that would mean dropping a lot of faith and momentum.
What about a custom highend chiplet design with more omph in the Gpu department?
Sure, if there would be benefits to other markets like a cloud server or more powerful mid range desktops. Or maybe just as an experimental and evaluation platform for features that will be coming to later M* gens.
What about just putting in 3d party accelerators? Basically 7900 duos mpxs? To me this seems like the simplest most straightforward way forward but braking the benefit of the simplicity of the “unified” paradigm.
Damn it, I can’t figure this out!
I just want 20+ Cpu cores and then about 300 gpu cores, 192 Gb unified ram, a ssd raid on pcie5.
I also want to be able to upgrade those gpus once to land on 2x the perf wise when m3 m4 gen is out.
And that new xdr screen that matches the mbp quality.
I also think people should be able to get 40+ cpu cores, 600+ gpu cores, 384 or more Ram etc.
And alos important, a music person should be able to get all the cpu but none of the gpu. As well as a lot of ram for immense audio libraries.
You know what? I actually think apple is smart enough to do give us all what we want and that they will do that. I am more worried about price since I actually compare to what is available on the PC side.
So boring that nobody is dropping any juicy rumors anymore.
I've seen a few people bring up this DGX A100 thing, but I'm not getting the connection to a desktop workstation. From a quick look, it seems the DGX is some kind of specialised AI server, with 2x 64 core CPUs, 2TB of RAM, 8x A100 GPUs, 30TB of NVMe SSD etc. etc. No wonder it consumes 6.5KW. No idea what one costs, but I'd imagine it's firmly in 'if you have to ask' territory. I doubt it features in Apple's competitor analysis for the new Mac Pro.
If those are compute modules they will run hot and need adequate cooling otherwise the heat transfers from one module to the next to the next to the next.
If they use some kind of special PCIe connector then that needs to be splash and dust proof, shielded to prevent shocks and provide enough power to each module. The maximum number of modules you can stack is limited by how big the main PSU is and how much power it can transfer upwards.
Lots of things can go wrong.
Acer did this as a stackable PC but to keep heat down they had to use very basic CPU and GPU and each module had its own power connector. Ugly.
The benefit of accepting off-the-shelf GPUs (if presumably with custom firmware), is that it frees Apple from having to build monster GPUs itself. They can ride the coat-tails of the GPU industry.
It does seem that Apple have been shutting the door on that though, in their recent developer documentation. To me, that's the biggest reason Apple are unlikely to suddenly announce support for them in ASi. They could easily have supported them as eGPUs over TB for the past few years, getting everyone up to speed with using them in conjunction with ASi. That they haven't is telling.
Plus, it seems a bit pointless having an energy-sipping architecture, then hooking it up to a 400W GPU.
Yes, it's an intriguing puzzle all right. I can only think that there's a missing piece we're all unaware of, as we've been through all the permutations based on existing products, and none really make any sense. When you start speculating about building computers by connecting multiple M2 Max SoC cards together, you're scraping the barrel.
Seems reasonable.
Doubt it's a high priority for Apple. And if you can only buy them from Apple, they'll cost a fortune anyway.
Mac buyers are loyal and willing to spend a bit extra, but aren't totally price-insensitive.
We'll know at WWDC.
So you can't imagine Apple looking at a 6500W $250k nVidia desktop workstation and thinking "Hey, if we make an Apple Silicon version of this, it would only consume 1400W, only cost $25k to make, and will absolutely decimate every single desktop workstation in its class"?
I mean sure I guess they could just put two laptop SoCs in a big metal box and sell it for $10k, but that sounds pretty boring.
I think Apple will have to do something different for the Mac Pro. Yes it's low volume, and yes Apple like large margins, but perhaps they'll need to suck up the cost a bit for the Mac Pro, just to make it happen. Switching away from x86 has a bunch of benefits; not having off-the-shelf workstation chipsets available isn't one of them.
A reconfigured SoC that's mostly GPU, using the same GPU core design as every other M-series chip, seems pretty plausible. The only issue is how to connect it to a regular Max chip. A Max only has one UF connector, so you'd wind up with something that's got more GPU than an Ultra, but less CPU. Bit of a wash in terms of pecking order.
And there's currently no way to link two Ultras together, since there is no spare UF connection to do so. The only way it could (theoretically) work, at least with the SoC layout we're familiar with, is if a bridge chip sits at the centre of 4 SoC / GPU dies, allowing them all to communicate.
The DGX is highly specialised piece of kit. The Mac Pro needs to be able to sell to a fairly wide range of users to have any chance of being green-lit by Apple.
Also, I don't think there's any reason to believe than an Apple DGX would be any cheaper, or even consume much less power. The 64-core EPYCs only account for 280W each. 2400W (8x300W) is for the GPUs, which far outstrip the performance of anything Apple has built thus far. 2TB of RAM, 30TB of SSDs, 10x 200Gb/s network interfaces and powerful cooling fans will use a fair bit too, though can't quite see how it all adds up to 6500W.
They'll need to do better than that, as the Ultra already achieves most of that for $5K, and the big box won't have upgradeable CPUs, GPUs or RAM in its favour.
Huh? I mean yeah you get a lot of nVidia specialised hardware and software bundled with it, but fundamentally it's just a monster EPYC based PC with 8 fairly normal nVidia GPUs in it. Sure they sell it as an AI server, but you could happily run standard GPU compute stuff on it if you were made of money.
I mean yeah, sure let's stop talking about the DGX A100 and just talk about an EPYC based Windows PC with 2 nVidia workstation cards in it...... then imagine you had 6 more GPUs.
Did you actually read my post, or just lightly skim it...?
The "GPU-heavy" configuration is an option...
Responding to @jmho
Yeah, but it's all that integration and specialised hardware / software, for a niche market, that makes it $250K. It's not just an EPYC workstation with a few GPUs, and if it were, it would be a lot cheaper. I don't know what Boxx charge for an APEX M-Class, but I doubt it's a quarter of a million dollars.
If Apple wanted to have a crack at that specialised AI market, they would also need to put a lot of work in, and likely have to sell it for a similar price.
Yeah, but it's all that integration and specialised hardware / software, for a niche market, that makes it $250K. It's not just an EPYC workstation with a few GPUs, and if it were, it would be a lot cheaper. I don't know what Boxx charge for an APEX M-Class, but I doubt it's a quarter of a million dollars.
If Apple wanted to have a crack at that specialised AI market, they would also need to put a lot of work in, and likely have to sell it for a similar price.