DuhSesame

tht said:

maciekskontakt said:

So what Apple is going to offer in those Mac Pro's? 8 core and 32 GB RAM maximum? "You do not need more"? Stop harassing those who use all that power and give them what they ask. I have used 64 core Linux based servers circa 2010 in business and 16 core was my personal for development in bank. Now some content creators may need more regardless how good M2 might be. I write it from Linux desktop with 8 cores and 64GB RAM that also run Windows 11 Pro in VirtualBox whenever needed (I wish Apple would allow VM's outside Apple hardware - I would pay for OS itself).

People understand concept of modern CPU and lower power consumption, but limiting hardware just because "you do not need it" is an arrogance when it comes to power workstations and servers and power users.

For the Apple Silicon Mac Pro, it will likely be, for lack of the actual branding right now, an M1 Max, M1 Max Duo, and M1 Max Quad, with possible maximum RAM to be 64, 128 and 256 GB respectively. Possibly 128, 256 and 512 if they get enough 128 gigabit density LPDDR5 RAM. IOW, take the M1 Max, put two of them in package for the M1 Max Duo, and 4 of them in a package for an M1 Max Quad, along with their in-package memory channels and memory.

The M1 Max has 8 performance cores, 2 efficiency cores and 32 graphics cores (plus media ASICs). 8+2+32 in shorthand. 2 of them means 16+4+64. 4 of them means 32+8+128. 128 graphics cores will be about 40 TFLOPs of single precision compute, about what a Nvidia 3090 yields. 32 performance CPU cores will be anything from a 20 to 64 core x86 chip depending on type of operation. They may have a variants using M1 Pro chips too.

Apple Silicon CPUs are about 70% faster per cycle than x86 cores (3.2 GHz ~= 5 GHz Intel). Since most of the high core count x86 CPUs clock their CPUs cores at 2 to 3 GHz depending on core count for sustained loads, Apple's 32 p-core CPU could performe like a 64 core x86 CPU. It will all depend on app design with all these core counts. A lot of apps are using GPU compute or ML/inference algorithms, making it even more dependent on app design. Memory capacity is up in the air a bit. The above capacities are based on the M1 Max and doing a 2x and 4x. They can stack RAM 2 to 4 layers high and double and quadruple up from there.

Yes, I agree with you on the last comment. Apple needs to stuff as many chips as possible into a 1500 W workstation. Don't leave that off the table.

blastdoor said:

DuhSesame said:

Marvin said:

blastdoor said:

9secondkox2 said:

A quad m1 Max based on current architecture would terminate the discussion. 

A Mac Pro with that or more would be legendary. 

Key word is “would.”

if the quad max takes until the end of 2022 then add “have.”

AMD is supposed to have 5nm zen 4 out later this year. For high end systems, at least 64 cores. 

Apple will have the advantage of unified memory, and maybe that will give them an edge for some workloads. But I really hope they pick up the pace of apple silicon rollout + update on the Mac. 

A quad M1 Max would be expected to be close to a Threadripper 3995WX:

https://www.newegg.com/amd-ryzen-threadripper-pro-3995wx/p/N82E16819113675
https://www.amazon.com/AMD-Ryzen-Threadripper-PRO-3995WX/dp/B08V5HPXVY/

That's $7-8k just for the CPU. The GPU performance would be about 30% higher than a $3k Nvidia 3090, e.g an Nvidia 3090ti:

https://www.amazon.com/NVIDIA-RTX-3090-Founders-Graphics/dp/B08HR6ZBYJ/

In late 2022, AMD is saying they can double the performance with Zen 4 on 5nm and Nvidia will likely have a 1.5-2x 4090. But, it will still cost around $10k for those parts. Apple can sell a quad M1 Max for $3k. Even if Zen 4 and 4090 is 1.5-2x the performance due to higher power usage, an M1 Max Quad would be really competitive.

Apple currently sells Mac Pro hardware that is way behind Threadripper chips, they don't need to compete with them on raw performance but the special hardware they have can offer huge improvements. Someone here compared their PC with 3090 against M1 Max and the Max chip beat it for render/export times in some cases:



Multiply that by 4 and for people in those workflows, a 2x improvement on the PC side is still half.

Something I just realized recently.  Intel has been testing their 56-core Xeon Ws for quite a while, I don’t see the M1 Max will ever match that.  If there’s a Mac Pro based on the M1, it’ll likely to be a lower-end to an Intel Mac Pro, at least until M3 released.

Well.... if you mean Ice Lake, then I think it's conceivable that a 40 core M1-based system could beat it (maybe). 

But Intel doesn't represent the best of what x86 has to offer. AMD's Threadripper is the current x86 king of the workstation market. If Intel is ever able to get Sapphire Rapids out the door, then maybe Intel can recapture the x86 lead, but that's both a big 'if' and a big 'maybe.' 

I'd love it if Apple kicked x86 to the curb completely, but only if they realize the full potential of Apple Silicon. 

But if Apple decides it's not worth it to completely replace x86 and wants to retain an x86 Mac Pro for the tippy top of the lineup, then I hope they move to use the best x86 solution available, rather than relying exclusively on Intel. 

9secondkox2 said:

GeorgeBMac said:

felix01 said:

Gotta pay 'em to keep 'em. Competitors are hungry for top talent. 

As Steve pointed out:  "A Players" don't chase money.  They chase challenges and an environment that supports their ability to create great things.  As with America's great industrialists:  the money was just a way of keeping score.

Apple just poured a LOT of resources into a CPU for the low-profit Mac where it will take years if not decades to recover their investment while the product matures and incorporates itself fully into Apple's ecosystem.   So, most likely Apple is wisely pulling back from that high cost push -- and the A Players are headed out to where the action is.

The core of that development is paid for by iPhone on day one. 

The customizations aren’t going to take long to recoup. Inside one year. 

The foundational milestones, like die to die interconnects and the all inclusive fabric will amortize over a couple years max and be around for a very long time. Apple has factored this into max pricing, particularly with custom builds. Apple doesn’t like to carry costs. That’s part of why there is an “Apple tax.” 

Well worth it. 

The sad part of these developers leaving is hat they probably weren’t happy at Apple. Be great to figure out why. But if it has to do with the “woke” philosophies, there isn’t much to do there. That’s Apple snd most engineers simply don’t buy into that. 

Then again, maybe Srouji is a hard boss. He seems pretty strict (not wasting a single transistor) and that’s why Apple silicon is so good. 

JustSomeGuy1 said:

DuhSesame said:

JustSomeGuy1 said:

To answer one of the questions posted here: Thunderbolt (TB3, specifically) is not tied to a version of PCIe. It is *roughly* equivalent to a PCIe3 x4 link, in bandwidth - 40gbps vs 32gbps. But you can use it with any PCIe.

And further about the Mac: If you're sticking with Intel chips, you don't have any option for PCIe >3 yet. Though that will change in the next few months with Sapphire Rapids, and if Apple really does release another Intel Mac Pro, it will likely use that, bringing it up to PCIe5. Of course, they could have used AMD, and indeed I think they absolutely should have. Using Zen2, they'd have had PCIe4. And they'd still have that now, with Zen 3... which still puts them only one generation behind.

There are plenty of problems with the Mac Pro, but this PCIe nonsense is a meaningless sideshow.

Right, can you list all the problems then?  Zen 2 aren't that reliable back then.

That's not true. Zen 2 was great, and it basically rewrote the rules for desktop and server. There were minor USB issues that affected some users on the desktop; that's about it, and very easily worked around, if Apple cared, given the 128 lanes of PCIe4 coming from each chip (64 each usable in two-chip systems, still 128 total).

And no, I won't make a list, it's been done to death here in the past including by me. However in brief - the Mac Pro was competitive and a class leader for Intel systems, but grossly uncompetitive against AMD systems. The SSD setup is some serious engineering but brings relatively few benefits compared to the repairability drawbacks. The lack of nVIDIA drivers, while possibly not Apple's fault, still holds them back significantly in some applications.

But the biggest problem with the Mac Pro is new since then... which is that it's not new since then. There have been no revisions. That's an appalling error, which they should know better than committing AGAIN after the last decades-long (seemingly, it was actually six years - which is ridiculous anyway) period of stagnation. If SuperMicro (for example) can bring out two dozen new motherboards every year or two for Intel processors, Apple can certainly bring out ONE new motherboard for their Mac Pro. And yet...

it's my sense that, due to stagnation in the server chip business through 2021 on the Intel side, Apple hasn't yet totally lost all relevance, but they're hanging on by a thread. The M2-based Pro presumably due this year (or perhaps still M1-based, if they're using a four-way symmetrical mirror of the M1 Max) will prevent that, but they will still lose a big chunk of what's left of their market without an x64 system, as much of the Pro software won't be available for the Mx chips yet, and many people will avoid Rosetta despite its overall good performance. So it's still reasonably likely that they will come out with a Sapphire Rapids based Intel Mac Pro this year, along with whatever Mx-based Pro (or iMac Pro) they ship. (And that would be a predictable shame, as an AMD Zen3 system would still be better.) We'll see soon enough.