Mac Pro Now Starts With 512GB of Storage and Radeon Pro W5500X Graphics

[jeffpeng]

They're doing some weird interposer magic with the M1 Ultra at least for the GPU to make it addressable as a whole, but it would be interesting to see some specific 3d benchmarks that put it over the threshold of 32 or 64gb (per Max die) slightly to see the performance hit on going across the 'UltraFusion' interconnect, which I'm sure is small but probably is measurable.

I haven't seen this yet but hopefully anandtech will do a test like this because it will give us a good idea for what to expect with the M1 Mac Pro (if it exists). If there's a decent amount of latency introduced it might be worth waiting for the M2 version when they refine things a bit. This interconnect technology is from 2017 or so and while it's cutting edge in that they're the first to use it, it isn't cutting edge as far as what TSMC is capable of for future products.

The die stacking and vertical/'3D' interconnects will be very interesting and could truly reduce latency tremendously just due to the very short trace length - see HBM2/3.

Also needing to address the neural engines discretely in the Ultra for ML makes it seem a little bit unfinished/like a compromise product. But the achievement with the GPU really should not be understated, it is huge, especially for a first-gen architecture.

I really want an Icelake (or better) Mac Pro to hold me over for 3-5 years while they sort all this out and get things 100% compatible, but it seems unlikely at this point since the last intel Mac came out in August 2020.

That interposer isn't magic. On a die-level AMD has been doing this since Zen. The really amazing part is just that you can now do it with two seperately manufactured dies at apparently very low latency. But it's no coincidence both are manufacturing at TSMC. Also: the entire NUMA problem isn't as pronounced with GPUs since GPUs hardly ever create thread crosstalk, nor are their threads living long enough to be physically reassigned to another computing unit.

"Within" the M1 Ultra I am confident no meaningful latency exists between the two chips. But that would radically change if another package on another "socket" would be introduced, wired over PCB, probably going through some SerDes implemetation - assuming M1 Ultra even HAS such an interface (of which we have no evidence as of yet).

I really cannot imagine them putting M1 Ultra into the Mac Pro and calling it a day, claiming it's better on the CPU side than the 28 core Xeon. Which it .... technically.... is (but also isn't). They - technically - could, and - technically - wire up PCIe-Cards via internal Thunderbolt, so you would - technically - have expandability, and - technically - there is no hard barrier to running the AMD GPUs with M1 .... but those devices would be severely bandwith starved since one Thunderbolt port delivers basically what's PCIe 3 x4. That's the actual problem: unless M1 Ultra has some interconnect we do not know about M1 Ultra does not have sufficient IO. Not to wire up external PCIe device, not to wire up multiple packages.

Maybe they can actually repurpose the SerDes hardware of the thunderbolt controllers to work in conjunction somehow. And maybe they can be driven much higher than "just" 20 GT/s per lane. Let's just assume they are actually PCIe 5 equivalent, and can drive 32 GT/s. That would mean that an M1 Ultra could - technically - have 32 times 2 times 6 equals 384 GT/s of bandwidth. For comparison: a XEON W has 64 48 PCIe 3 lanes, resulting in a theretical IO bandwidth of (also) 512 384 GT/s. So that doesn't sound totally terrible, and considering the storage is not taking up any of this bandwidth it could actually be somewhat possible, but if we then again consider that you'd have to substract any and all displays from that bandwidth, things don't look as peachy. Not to say that this is a metric ton of IFs and speculation with little to no basis.

But even IF that was actually an accurate depiction of how Apple would integrate the M1 Ultra into the Mac Pro, it would still be quite underwhelming to offer a successor 3 years later that has less not more IO, less CPU cores that at best "trade blows" with the Xeon W when it comes to multi-thread performance, and with significantly less memory - if Apple sticks to the on-package memory, which they kinda have to, since if they wire up normal DDR5 sticks to those memory controllers the GPU performance will tank considerably.

All in all: I just don't see it. I just don't. I don't see Johnny Srouji's pulling an A16 based, N4 fabbed, MCM-style M2 Ultra Extreme Gigazord rabbit out of his pants at WWDC. And that's kinda what they would need to actually make people upgrade to an Apple Silicon Mac Pro.

Note: Edited since the Xeons in the Mac Pro actually have 48 useable lanes, not 64.

 

[novagamer]

That interposer isn't magic. On a die-level AMD has been doing this since Zen. The really amazing part is just that you can now do it with two seperately manufactured dies at apparently very low latency. But it's no coincidence both are manufacturing at TSMC. Also: the entire NUMA problem isn't as pronounced with GPUs since GPUs hardly ever create thread crosstalk, nor are their threads living long enough to be physically reassigned to another computing unit.

"Within" the M1 Ultra I am confident no meaningful latency exists between the two chips. But that would radically change if another package on another "socket" would be introduced, wired over PCB, probably going through some SerDes implemetation - assuming M1 Ultra even HAS such an interface (of which we have no evidence as of yet).

I really cannot imagine them putting M1 Ultra into the Mac Pro and calling it a day, claiming it's better on the CPU side than the 28 core Xeon. Which it .... technically.... is (but also isn't). They - technically - could, and - technically - wire up PCIe-Cards via internal Thunderbolt, so you would - technically - have expandability, and - technically - there is no hard barrier to running the AMD GPUs with M1 .... but those devices would be severely bandwith starved since one Thunderbolt port delivers basically what's PCIe 3 x4. That's the actual problem: unless M1 Ultra has some interconnect we do not know about M1 Ultra does not have sufficient IO. Not to wire up external PCIe device, not to wire up multiple packages.

Maybe they can actually repurpose the SerDes hardware of the thunderbolt controllers to work in conjunction somehow. And maybe they can be driven much higher than "just" 20 GT/s per lane. Let's just assume they are actually PCIe 5 equivalent, and can drive 32 GT/s. That would mean that an M1 Ultra could - technically - have 32 times 2 times 6 equals 384 GT/s of bandwidth. For comparison: a XEON W has 64 48 PCIe 3 lanes, resulting in a theretical IO bandwidth of (also) 512 384 GT/s. So that doesn't sound totally terrible, and considering the storage is not taking up any of this bandwidth it could actually be somewhat possible, but if we then again consider that you'd have to substract any and all displays from that bandwidth, things don't look as peachy. Not to say that this is a metric ton of IFs and speculation with little to no basis.

But even IF that was actually an accurate depiction of how Apple would integrate the M1 Ultra into the Mac Pro, it would still be quite underwhelming to offer a successor 3 years later that has less not more IO, less CPU cores that at best "trade blows" with the Xeon W when it comes to multi-thread performance, and with significantly less memory - if Apple sticks to the on-package memory, which they kinda have to, since if they wire up normal DDR5 sticks to those memory controllers the GPU performance will tank considerably.

All in all: I just don't see it. I just don't. I don't see Johnny Srouji's pulling an A16 based, N4 fabbed, MCM-style M2 Ultra Extreme Gigazord rabbit out of his pants at WWDC. And that's kinda what they would need to actually make people upgrade to an Apple Silicon Mac Pro.

Note: Edited since the Xeons in the Mac Pro actually have 48 useable lanes, not 64.

Yes I agree with you on pretty much all of this, very well put. I didn't consider your point about thread crosstalk for GPU workloads mitigating the multi-die problem.

I am discouraged by the tweaks to the base offering Mac Pro because of this. Speculating more, I'm thinking we might be in a 2010-2012 situation with this 2019 machine where because of parts availability they silently tweak the configuration instead of release something truly updated.

I also don't get how they'd tease an über AS Mac Pro at WWDC when they have professionals buying $8,000 Mac Studios right now. It seems like a difficult story to tell and it would almost be easier for them to set expectations that the M2 Mac Pro will be amazing and is coming a year from now in 2023, and for the time being here's this last Intel update for the professionals that need the GPU power and expandability in the meantime. But it would be a whole lot easier for them to do nothing at all and that's what I'm expecting, but I hope I'm wrong.

I guess there's always the comedy option that Apple will maintain the absolute insane pricing and charge $40,000 for an Apple Silicon Mac Pro that does have a monster M2 this fall. That would be less than ideal to say the least.

With the rapid pace of Apple Silicon development I think we'll probably get updates every 2-ish years, so I expect the AS Mac Pro to fall in a similar price range of $6,000-$15,000 for most configurations and I don't think the yields on a new architecture they haven't even released at the most base level yet can work in that price bracket in 2021, unless they've been building these things for a long time in complete secrecy, which I doubt given the long lead times on the M1 Ultra.

 

[jeffpeng]

I am discouraged by the tweaks to the base offering Mac Pro because of this. Speculating more, I'm thinking we might be in a 2010-2012 situation with this 2019 machine where because of parts availability they silently tweak the configuration instead of release something truly updated.

I guess you might be right on that. I also guess that if apple pulls yet another stunt like this they will become even more scarce in "actually" professional studios. I know quite a few people that have moved to Windows boxes over the last decade in part due to this. Sure, the fact that there was no actual hardware advantage (but for the most part: disadvantage) to running a Mac any more contributed to this probably even more.

I also don't get how they'd tease an über AS Mac Pro at WWDC when they have professionals buying $8,000 Mac Studios right now. It seems like a difficult story to tell and it would almost be easier for them to set expectations that the M2 Mac Pro will be amazing and is coming a year from now in 2023, and for the time being here's this last Intel update for the professionals that need the GPU power and expandability in the meantime. But it would be a whole lot easier for them to do nothing at all and that's what I'm expecting, but I hope I'm wrong.

I had this idea a few weeks ago that "Apple Silicon" coming to the Mac Pro didn't mean that they used M1 Max / Ultra as the CPU - but actually soldered it to an add in board, and use it basically as a traditional GPU. I would make sense, a lot to be honest, since what sets Apple Silicon apart from the Intel/AMD boxes is - aside from power efficiency - M1 Max's and Ultra's impressive array of fixed function silicon dramatically accelerating Mac specific workloads, something Apple offered the hastily clobbered together and now badly aged Afterburner card for.

Also the GPU is no slouch, maybe not a 3090, but competing with AMDs latest offering, and considering those Pro GPUs come in at 5K each - nearly 10k for the duo versions - there exists a world in which you can go and put two M1 Ultra Duo cards into your existing 2019 Mac Pro and, assuming Mac OS can properly coordinate all those separate function blocks on their own dedicated memory, have a truly impressive machine. Even the CPU cores would maybe not got to waste, since one thing especially RISC CPUs aren't totally terrible at is actually ray, or more accurately path, tracing. Not RTX style realtime ray tracing, but in a render program properly supporting this architecture they could probably be employed for this kind of task quite efficiently.

So, if that were true, it would only be the logic conclusion to expect Apple to actually offer an upgraded, intel based Mac Pro, while also giving existing Mac Pro owners the option to at least upgrade their GPUs. Since, here is the thing: M1 Ultra barely beats the Xeon 3275 in mtperf, but not a 3375, and it would seriously get dunked on by a Sapphire Rapids with 48 Golden Cove cores ... which actually might come out just in time to make it into the new and improved Mac Pro.

I guess there's always the comedy option that Apple will maintain the absolute insane pricing and charge $40,000 for an Apple Silicon Mac Pro that does have a monster M2 this fall. That would be less than ideal to say the least.

With the rapid pace of Apple Silicon development I think we'll probably get updates every 2-ish years, so I expect the AS Mac Pro to fall in a similar price range of $6,000-$15,000 for most configurations and I don't think the yields on a new architecture they haven't even released at the most base level yet can work in that price bracket in 2021, unless they've been building these things for a long time in complete secrecy, which I doubt given the long lead times on the M1 Ultra.

2 years, yes. That sounds about right - and reasonable.

As for Apple coming out with the biggest possible M2 first .... nope, not a chance. The smaller the chips, the higher the profit, and also, at least for apple, top-end workstations are the least profitable market they cater to. Not by unit, but by volume. And since I do expect M2 to be based on A16 - not A15 - they will be very, very much interested to first saturate the iPhone market and then the still rather profitable mobile computers market, meaning iPads and MacBook(Air)s with the "small" M2 especially considering their silicon allocation is probably lower. Once they've done that we'll get the the Pro and Max chips for the notebooks again ..... and then 2024, MAYBE end of 2023, we'll see a potential M2 design, probably MCM, that can actually take on intels workstation chips in all metrics, including IO, Memory and raw performance.