Strong claim. Last I checked, the world is a very very big place. Nobody can please everyone all the time.
Somehow, I think this claim will not age very well.
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.
ARM Mac Pro - WWDC 2023
- Thread starter Zest28
- Start date
- Sort by reaction score
Somehow, I think this claim will not age very well.
By whom? Themself? Wannabes? "Everyone" loves to talk about iPhones not workstations.
I do, it's true.
I speak for all but you!
But I thought you knew,
that you can do it too!
My educated guesses.
Tim will announce the Mac Pro M2 Ultra 1-die & 2-die at WWDC 2023 3 days from now.
It will likely ship by December like the 2019 & 2013 models.
Starting price will be the same as the 2019 Mac Pro Xeon-W.
If Apple wants to save on industrial design cost they could reuse the same case.
This allows prexisting accessories to be reused.
Tim will announce the Mac Pro M2 Ultra 1-die & 2-die at WWDC 2023 3 days from now.
It will likely ship by December like the 2019 & 2013 models.
Starting price will be the same as the 2019 Mac Pro Xeon-W.
If Apple wants to save on industrial design cost they could reuse the same case.
This allows prexisting accessories to be reused.
Last edited:
I don't follow the logic around those calling for M3 chips at WWDC. Especially not M3 Max/Ultra - we have to see the base M3 variant first. I for one, would prefer Apple slow down on the chip evolutions. Meaningful upgrades >20% are far more worthy than those that are incremental in nature. The M2 really was more of an M1+.
I think the likely candidates for next week are an intro of the MacBook Air 15", refreshed Mac Studio (M2 Max/Ultra) & a preview of what's to come with Mac Pro. My guess would be the first M3 chips you'll see will be in late fall into spring of next year. Likely with the base model MacBook Air & iMac.
I think the likely candidates for next week are an intro of the MacBook Air 15", refreshed Mac Studio (M2 Max/Ultra) & a preview of what's to come with Mac Pro. My guess would be the first M3 chips you'll see will be in late fall into spring of next year. Likely with the base model MacBook Air & iMac.
Well, yes. We’d like to see a new microarchitecture, not an incremental architecture. And Apple needs a new product. It’s not like the competitors are sleeping. The new ARM cortex X4 for example has surpassed Apple design in some metrics.
Apple absolutely need constant releases of updated Macs. A yearly cadence is absolutely good for business.
M2 is not to entice owners of M1 Macs, but for folks still using Intel Macs and for switchers. Similarly M3 is not for M2 Macs owners or even M1s. There should be still a large install base of Intel Macs owners.
I don’t think Apple is adverse to introducing the theoretical M3 “Extreme” for the Mac Pro before introducing the base M3. I’m hoping to be surprised next week..
I might be tempted to switch to M3 from my M1 if there are noteworthy improvements.
Yep, a preview can be better than silence. Quite common in the pro market as well.
Actually, me too ... since I only have a lowly M1 Mac mini ... but there's no justification ... just tech lust ... haha.
All valid points except many of us "like" working inside Apple's OS. The OS itself offers a lot of advantages over Windows just for the sheer fact that it is more or less Unix under the hood.
Apple can not price any sort of new Mac Pro the same as the Xeon model unless they can match GPU performance against the AMD cards. Although once could argue that the Studio Ultra probably *does* have a better GPU than the base GPU in the Mac Pro (definitely better than the base released 580x.) But Apple needs to be striving for RTX4080/AMD 7900 speed GPUs.
Perhaps Apple is going to have "Modules" attached via AMD Infinity Fabric Link which would also mean a GPU could be slotted in. Now that would be bad ass. Yes it would be a little slower in the overall interconnect; but it will still be fast enough. Much faster than a PCIe. The AMD GPU would have to be a custom piece just like the MPX modules currently.
** actually a module system Apple probably wouldn't waste time with AMD parts; but another bad ass idea is a PCIe card that has Apple GPU hardware than can go into the existing Xeon Mac Pro.
Perhaps Apple is going to have "Modules" attached via AMD Infinity Fabric Link which would also mean a GPU could be slotted in. Now that would be bad ass. Yes it would be a little slower in the overall interconnect; but it will still be fast enough. Much faster than a PCIe. The AMD GPU would have to be a custom piece just like the MPX modules currently.
** actually a module system Apple probably wouldn't waste time with AMD parts; but another bad ass idea is a PCIe card that has Apple GPU hardware than can go into the existing Xeon Mac Pro.
Well, that’s easy enough. A moderately higher clocked Ultra should be able to reach 40 TFLOPs of FP32 (that’s real FLOPS, not marketing ones like AMD). That should already make it a decent competitor to AMD, especially with the huge amounts of GPU RAM Apple offers.
Matching Nvidia on raw speed is going to be more difficult of course.
Really? You can't think of any. I certainly can.
Nope, it is the same problem. The exact same problem. Both are antithetical to the Apple Silicon strategy, and require walking back on that strategy to allow them...all at worse performance than what integrated SoC offers. For storage perhaps that matters less. But it's still the same problem.
I don't get why you are claiming AMD was cooking the FP32 numbers. Before the RTX4090 (A6000), Octane was smoking fast on the 2019 Mac Pro with 2 W6800X Duos installed; well it is still smoking fast. That is all FP32.
AMD and nVidia use marketing ploys, so does Apple. Everyone does. I really do not lend a grain of salt to anything manufacturers say about performance. Best to wait for actual reviews from good non-biased sources.
It's more like, no one who needs this is waiting around for Apple to release it. They are all using whatever is available on the market right now, because no other consideration matters other than production and compile times. They also don't care if it requires 1000 watts of power to do it.
We don't "have to" see anything. Apple can launch its own line of Mac Pro chips called X1 or whatever, and it have nothing to do with M3. This linearity people expect is so bizarre. No, the Mac Pro is not beholden to the announcements of M3, M3 Pro, M3 Max, and M3 Ultra, before it can be talked about.
This makes sense if you commit to the view that Apple fully opposes any notion of expandability. I think this would still be a bit premature.
Just for the sake of an argument, let’s assume that Apple does not per se oppose expandability, but doesn’t want to compromise on their architectural vision either. Storage expansion needs PCIe, driver infrastructure, and protocol support. All of this is already supported today. And just 4x PCIe lanes give you the amount of bandwidth that will be more than enough for anything you can do with that kind of machine anyway. If Apple wants, they could easily add a few ports with this kind of utility in mind.
GPUs are different because they need much higher bandwidth to satisfy guarantees offered by the rest of Apple Silicon platform. Not to mention that AND GPUs don’t support some stuff that Apple really wants pro apps to use, plus they need other kind of optimizations. I mean, Nvidia gave their new superchip a 900GB/s CPU/GPU interface. How would that work in a modular setting? It simply wouldn’t.
Two very different problems. One is easily solvable, another not.
Everyone is advertising the best possible (often theoretical) scenario for their GPUs, but AMD might be a bit more disingenuous than others here. For RDNA3, they claimed doubling the FP32 compute throughput per CU. The reality is a bit more interesting. Their ALUs can indeed perform two operations per clock via a new packed instruction, but only under very specific circumstances. That’s why the actual measured performance has been less dramatic.
Except an M2 Max is already much faster than W6800X duo in Blender.
Sure, but it’s not about now. Apple plays the long game.
At what? GPU rendering? Viewport rendering?
Do you have any charts to back that claim up?
Ok I looked it up and that is just the generic Blender median benchmark; I may download it and see what I score since there isn't any system information listed for the 1 entry for a W6800X Duo.
I agree I'd like to see some more substantial micro-architectural improvements in Apple Silicon. Do you have any info on the Cortex X4 and how it has surpassed Apple's chips? I am not that familiar with the Cortex chips. I did read an article over at Arstechnica a couple of days ago:
Arm announces the Cortex X4 for 2024, plus a 14-core M2-fighter
Will anyone build Arm's 14-core mega chip? Will Arm have to do it itself?
One paragraph that caught my eye
"Blame whoever or whatever you want in Arm's path to market, but these launch events have a history of making performance claims that don't align with what actually arrives in consumers' hands. Just look at single core performance, where the company promised the X2 chip in 2021 would be 15 percent faster than the X1, then the next year said the X3 would be 25 percent faster than the X2. You won't see this reflected in any benchmark, and in Geekbench, the X3-packing OnePlus 11 only gets 9 percent higher single-core scores than the X1-equipped Pixel 7a. If Arm's claims lined up with reality, there should be around a 43 percent difference across that two-generation gap. Whether this is Qualcomm's fault or due to insufficient cooling or troubles with the manufacturing process, you can't take any of Arm's performance claims too seriously."
So I don't use Blender, but you can see that only 1 GPU was being used, so having a dual GPU card such as the W6800x actually hurts you in this Blender benchmark since they clock slower. Now perhaps the actual Blender app can use multiple GPUs for rendering in which the score would be better; the consumer 6800XT or 6900XT beats the M2 Max.
But those consumer cards lack what the pro level W6800X has which is lots of VRAM.
Either way you fold this, the fact is that only 1 GPU was being used on the W6800X Duo; and they are clocked slower since there are 2 of them on a single card.
Those RTX4090s just crush it though...lol.
But those consumer cards lack what the pro level W6800X has which is lots of VRAM.
Either way you fold this, the fact is that only 1 GPU was being used on the W6800X Duo; and they are clocked slower since there are 2 of them on a single card.
Those RTX4090s just crush it though...lol.
Attachments
That's what I was thinking -- there are no Apple Silicon devices that are expandable. (other than through TB, USB, and SD Cards.). The integration makes it harder to do and probably not worth it.
There were also no expandable Intel Macs, save for the Pro (and even the Pro wasn’t expandable for some time). But nobody would claim that x86 cannot make expandable computers, right?
M-series chips have PCI-e lanes, just like x86 systems (only fewer of those). Whether to expose these lanes as PCIe slots, external TB or something else is the question of system design.