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.
So M1 Max with 32 Core GPU matches RTX 3080 Mobile.
- Thread starter iBug2
- Start date
- Sort by reaction score
I haven't read those rumors that states Apple will make original Apple Silicon specific for pro desktops.
My idea is to reuse the M1 Max 64GB dies.
Jade 2C and 4C are rumored to effectively be doubled and quadrupled M1 Max (Jade C-Die) designs.
That said, the RAM isn’t on the die.
Last edited:
I know there are a bunch of 3D artists here, ZBrush works absolutely incredibly. I have a 75 million poly ecorche which my M1 Air couldn't handle without heavy stuttering which the 32 GB M1 Max is handling with ease. It feels at least as fast if not faster than my desktop AMD 3900X which is really impressive since it's still using rosetta.
So this animator has been doing a series on the M1 and Apple loaned him a 14” M1 Max to play with. Here are their thoughts and some benchmarks:
“The wider takeaway here though is that in order to give the M1 Max some real competition, one has to skip laptop chips entirely and reach for not just high end desktop chips, but for server-class workstation hardware to really beat the M1 Max. For workloads that push the CPU to maximum utilization for sustained periods of time, such as production-quality path traced rendering, the M1 Max represents a fundamental shift in what is possible in a laptop form factor. Something even more exciting to think about is how the M1 Max really is the middle tier Apple Silicon solution; presumably the large iMac and Mac Pro will push things into even more absurd territory.”
Wonder if the agreement with Apple is to avoid comparing with AMD mobile CPU.
"A huge thanks to everyone at Apple that made this post possible!"
No rumors have said that the Mac Pro will have its own Mega Monolithic SoC (which would be hyper expensive & have low yields), all rumors have been towards multiple M1 Max SoCs for the Mac Pro lineup.
So your idea is exactly the same as the rumors.
Possible Mac Pro lineup:
Mac Pro - This may continue for another refresh as an Intel-powered box, or may shift over to M1 Max SoCs.
Mac Pro Cube - "New" product, new design, all Apple silicon. A shoe-in for the M1 Max Duo (Jade 2C), possibly for the M1 Max Quadro (Jade 4C).
iMac Pro - Still a rumored item, thought to be 30" or 32", but more recent rumors might have still at 27", a definite redesign to match the 24" iMac though?
I would think a Mac Pro Cube & a 27" iMac Pro would be perfect for the M1 Max Duo SoC package, reserving the M1 Max Quadro SoC package for the Mac Pro, allowing the excess PCI lanes to be used for actual PCIe (MPX) slots?
Reuse the M1 Max die onto a multi SoC package. I would be surprised if they will make memory user upgradable
You can stop wondering because Anandtech got review units from Apple and compared them to recent AMD-powered laptops.
The catch becomes how to manage movement of data between the dies. Giving each "node" isolated memory does pose issues once you start getting into larger data sets. It also makes I/O more complicated. DMA from Thunderbolt to which die? Then move it to another die's memory? Same thing for SSD access.
This approach poses enough problems, that if Apple's going to go through the trouble of putting 4 dies on a single package, and coordination needs to be solved anyways, they have two options in my mind: Attach the dies to each other using the fabric interconnect and solve the thorny problem of memory/data fragmentation. Or, attach the dies to a small I/O die with the fabric interconnect so that the different dies can see a unified view of memory and I/O.
If we assume the weird die shots of the M1 Max are hiding some circuitry that Apple hasn't yet wanted to reveal, either of these would be possible. Adding an I/O die as the way to solve the I/O and memory coordination issue would not add to the cost appreciably, and keep system management more consistent, and lessen the need to move data between isolated SoC nodes.
My point in mentioning that the RAM isn't on die is that Apple doesn't have to segment RAM to the SoC die, because it's not part of the die. You can still take an M1 Max die and restructure RAM differently as well.
Im glad apple is crushing it because it forces intel, windows, and nvidia to get their act together. However the mac could be as powerful as a 3090 and it still cant play games. Not everyone is a video editor
It's crap though compared to Phoronix and Hardware Unboxed. Prefer real world workloads over specint/fp. And, they're cherry picking two games instead of a whole suite. Plus, Anandtech is/was employed by Apple so there's the bias issue.
Yeah he explicitly said otherwise but sure … mi7chy … sure everyone is else corrupt …
Anand left Anandtech to go to Apple. Always amused by fanboys delusions that if a tech review gives a positive review to a product from a company you dislike or a bad review to a company you like it must be because they’re corrupt. I’ve seen every review outlet under the sun accused of being shills for every company under the sun and only occasionally has it been anything other than projection.
Did you know that SPEC tests are based on "real world workloads"? Have you looked at what's included in these tests?
As for Phoronix, I'm not sure if what they test is more "real world". Last time, they tested algorithms I've never heard about and that were never optimised for ARM, so much so that the X86 version of some algorithm running under rosetta 2 outperformed the ARM native version by a lot.
What does Hardware Unbox test? Games? Placing them before Anandtech is rather absurd, IMO.
We don't need anyone to tell us that macOS is not a gaming platform. That won't show us the capabilities of the hardware.
Nice goalpost shifting BTW. What were you saying about Apple forbidding comparisons with AMD mobile CPUs?
I think the main issue is the lack of hardware accelerated ray tracing on the M1 GPU. Though he said the memory model of the M1 obviously gives the GPU a huge advantage in that aspect. Personally I suspect not enough to overcome the former.
If I remember right, the point about how branchy the code is was to highlight how CPU and GPU renderers are quite different. In his paragraph describing his intention to port his renderer over to GPUs using CUDA and Metal, he said he was going to write a post comparing the two APIs for compute which I will find very interesting as I’m primarily a CUDA guy.
If the M-series is praised in reviews, the reviewer is biased/paid by Apple, if it’s benched very highly, the benchmark is flawed, and only real world tests matter, if it does well in real world tests, then it doesn’t matter because it can’t play games.
It’s a goddamn conspiracy to make Apple look good!
M1 Max GPU beats $6000 Radeon Pro W6900X in benchmark - 9to5Mac
A new benchmark test run with Affinity's tool shows that the M1 Max's GPU beats the $6,000 AMD Radeon Pro W6900X for some tasks.
9to5mac.com
Yeah, Anandtech is the gold standard for in depth reviews of gpus and cpus. Saying phoronix and hardware unboxed are better is the funniest thing ever. Just say you want to dislike it and we can all move on.
specint/fp is like using MS-DOS snippets to benchmark in 2021.
Look how comprehensive and current Phoronix is in comparison:
https://openbenchmarking.org/suites
https://openbenchmarking.org/tests
https://www.spec.org/cpu2017/Docs/overview.html#suites
Anandtech is good enough for first time users but I stopped going there probably over a decade ago. Don't like that they cherry pick their results.
Look how comprehensive and current Phoronix is in comparison:
https://openbenchmarking.org/suites
https://openbenchmarking.org/tests
https://www.spec.org/cpu2017/Docs/overview.html#suites
Anandtech is good enough for first time users but I stopped going there probably over a decade ago. Don't like that they cherry pick their results.
Anandtech provide the most in depth and accurate benchmarking and reviews. Phoronix do not.
I don’t think anandtech does GPU performance reviews, they do talk about architecture though.
Cool. So his render probably would see performance increases from utilizing the hardware bvh units.