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.
[CPU only] Apple M1/2(Max/Ultra) (TSMC 5nm) vs AMD Zen 4 (TSMC 5nm) - Technical Analysis
- Thread starter exoticSpice
- Start date
- Sort by reaction score
If there are no rumors about Apple and Nvidia's future GPU, what makes you think that the M3 SoCs GPU could match Nvidia's RTX 5000 series in performance and features?
What workloads are you talking about? 3D rendering?
AMD Radeon PRO W7800 & W7900 Workstation Performance Review
AMD's Radeon PRO series of workstation GPUs has just been augmented with the first models infused with the company's RDNA3 architecture. We're taking a look at the high-end 32GB Radeon PRO W7800, as well as the even higher-end 48GB Radeon PRO W7900, across a wide-range of production workloads.
techgage.com
We have to contextualize the situation. The niche where the Mac Pro is inserted IS precisely targeted at users who DO care about flexibility and upgradeability. So much so that it has expansion slots, but it is LESS flexible than its last iteration.
No matter how you slice it, it doesn't make sense to pay an extra what, $3,000 to pretty much only be able to add specialized sound cards to your system?
If you don't care about flexibility, you might as well go for a Mac Studio, a Macbook, or a low-cost solution if you only use a PC for light documents.
In practice, that's the niche of the niche. You would need to have someone who specifically relies on unified memory. On the PC side, you can get a Geforce 3060 for around $250-270. Stack two of these and you have 24 RAM.
Is it ideal? Of course not. But it's a far more customizable / flexible solution than what Apple offers, with the added bonus that it'll be reasonably faster everywhere else, including at machine learning and raytracing, which you just mentioned. And all for around $500. It's a steal if you play the cards right.
Yeah, maybe I was a bit careless about the performance bit, that will be hard to do simply because about the available die area. But I was thinking more about features and capabilities. I am fairly optimistic that Apple's upcoming raytracing technology is going to be as good as Nvidia's (maybe even better). Also, I am certain that their next-gen GPU will fill in missing gaps in regards to GPU atomic operations etc. And finally, I believe we are about to see a really big push in ML inference for Apple, which will result in dramatically improved performance.
What workloads are you talking about? 3D rendering?
AMD Radeon PRO W7800 & W7900 Workstation Performance Review
AMD's Radeon PRO series of workstation GPUs has just been augmented with the first models infused with the company's RDNA3 architecture. We're taking a look at the high-end 32GB Radeon PRO W7800, as well as the even higher-end 48GB Radeon PRO W7900, across a wide-range of production workloads.
Pretty much anything really. Rendering, compute, you name it. I had a look at the benchmarks, and AMD leads in entirely meaningless categories like viewport performance (like who cares if your viewport renders at 120fps, that's not going to impact your work in the slightest).
Well, are you talking about upgradeability or are you talking about specialisation? Those are different things. I can certainly imagine that there are users out there that need specialised equipment or large amounts of disk storage — Mac Pro works well for them. Upgradeability is something else.
Absolutely, and that will cover the needs of the vast majority of users. But that won't give you specialisation.
Not necessarily unified memory, just large enough problems. Multiple 3060 will work if you can neatly divide the problem into smaller parts that can be solved independently. Not all problems are like that though. Rendering is a great example. While you can pre-process the scene and only send each GPU the data it needs, that pre-processing step itself is going to be fairly expensive and will mostly defeat the purpose.
Possibly. But at what costs?
I just built a custom PC, which is comparable with the M2 ultra CPU-wise (7950X) and beats it by a landmile GPU-wise (4090) for roughly half the cost.
A future ASi variant that can compete with the 4090 (or, in the future, a 5090?) almost certainly would go up in price even more.
So maybe Apple can compete on sheer compute power - for what? 3x the costs? Would not be worth it for me, honestly. It would still be an island solution, still no CUDA, still non-upgradeable
Same as today I suppose?
Why would you think so? Cost of Apple product didn't change much since 2006 despite inflation. Cost of third-party GPUs went up tremendously though. In fact, I would be more worried about the price of the 5090... if it is indeed on 3nm, and if 3nm is indeed as expensive as they say, you might be looking at $2000 or more for the GPU.
No, no, that's not correct at all.
Quote: "It was deeply customizable, it was wicked fast, it came with 1GB of RAM but could address 32GB of RAM, and Schiller announced that it was shipping. On August 7, 2006, you could buy one for $2,499 — $3,571 in today's money."
Source: https://appleinsider.com/articles/2...d-and-still-the-best-mac-ever-made----for-now
The MP of today starts at 6999 - a 100 % increase. And its disputable if the new MP is comparable
You are right, sorry, I was mentally stuck on laptops.
Comparisons are a bit difficult, because the tier ranking have completely changed. For example the 2019 model started with the Xeon W-3223. That was the lowest tier available Xeon-W model. Contemporary Sapphire Rapids alternative would probably be Xeon W-2435 or maybe W-2445 (they are also priced similarly to the W-3223). But the M2 Ultra is equivalent of the upper-mid range W-3465X, which is more or less the successor of the 2019 24-core W-3265M, and that was the $12000 Mac Pro option. Overall, I'd say that the M2 Ultra Mac Pro is much better value for money than the 2019 Mac Pro — but this comes at the expense of configurability options. You get the equivalent of a $3k Xeon and a $3k workstation GPU, but that's it, you can't get anything better.
Just a note, the 3090ti MSRP was $1999, heck the Titan cards (that the 90-series is derived from) were $2400/2500 cards. So yeah nvidia does have room to up the price if they choose to.
I think you are too optimistic. Apple has yet to add some IP, such as the AV1 de/encoder and ray tracing hardware. Doesn't the Nvidia GPU offer more performance per surface area than the Apple GPU?
P.S. I find it funny that a thread about CPU comparisons ended up in a GPU comparison.
Last edited:
You said it perfectly. It's an ISLAND solution. Apple has isolated themselves, really. That's worse than the Apple Tax they charge.
I wonder why. I doubt they intended the ASi machines to be like this. Also doubt that they underestimated the required engineering effort. One must be true though
IMHO they really went greedy / overconfident in how much their solution would scale. E.g, they thought they would maintain their performance advantage, that it would be larger, and that they would be able to scale it up well.
But it's not what happened. As I also mentioned around 2 years ago, Apple has placed a gamble. Many people scoffed at the my back then that AMD and Intel wouldn't stay still. Well, that's exactly what happened, and even more so from AMD's part.
If Apple loses the performance advantage, then they have really nothing to show. Even if AMD and Intel had worse performance, the legacy compatibility and extra flexibility already heavily weights in their favor. But if they have the same performance / efficiency or BETTER, then you're paying a premium for... what exactly, if Apple is not investing in polished and cutting edge software anymore?
Supposed Apple released a Mac tomorrow that's almost as powerful in compute compared to a 4090 (on paper) with 192GB of Unified Memory, would you be in the target market for such a solution?
Looks like most folks don't think the world is a very big place with many different needs. People's computing needs are always what they "know".
Oh well.
I purchased the 1,1 immediately after its announcement.
It's been up-graded to dual x5365's, and a Radeon 5770
Haven't powered-on this Big Beast in a handful of years, but I can safely assume that the new (new (new)) MP would laughably spank it.
It's interesting that you criticise island solutions and then mention CUDA, a proprietary API that's locked to a single vendor. In a way Apple's current insular policy is a consequence of Nvidia's business strategy. Back in the day Apple was heavily invested into the open GPU computing, only to see their efforts be sabotaged by Nvidia who wanted to build their empire. Well, Nvidia succeeded , and Apple decided to do their own thing.
If Appel can deliver the performance and the features, interest will come. I certainly see a future where Nvidia dominates cloud and cluster GPU computing and Apple dominates mobile (+partially desktop).
They are still essentially on their first generation hardware and even M2 is not much more than a replicated iPhone core (with a new fabric). I think your conclusions are being a bit premature.
Yes, if by not staying still you mean compensating the lack of architectural progress by packing more processing cores and cranking up the power consumption. Because this is such a scalable strategy... Less than a decade ago 65watts was the TDP of an enthusiast-class desktop processor. Today this is an enthusiast class laptop processor.
AMD made some great progress though and I am looking forward to Zen5. They managed to gain parity with Intel in some key metrics and overtake them on energy efficiency. Still, Zen4 is not yet on M1 level and AMDs GPU tech is focused on cost reduction more than anything, so I wouldn’t really overemphasize their progress.
Last edited:
Per surface area? If I remember correctly, they are roughly the same, normalized for clock frequency (I think we had an extensive discussion about this?). The big advantage for Nvidia is that they can cram the due full of GPU compute, while Apple can dedicate Mac 40% of due area to it.
This will all come to next-gen manufacturing and packaging technologies. With the wafer costs rising as they are multi-die technologies will be key. I am cautiously optimistic about Apples prospects here since they have been investing more R&D in this area than other companies.
Well, looks like in these forum threads, such criticism only applies to Apple. All others will get a free pass.
You know as they say, winners are not judged. Nvidia definitely emerged a winner from the GPU infrastructure wars - they were more focused, more consistent, more driven while delivering steady technological innovation (and a healthy dose of market manipulation didn’t hurt either)
It’s definitely too early to declare Apple a winner or a loser. Next five years will show. I just want to remind everyone that not that long time ago Nvidia GPUs were vastly inferior to ATI and Intel lagged behind AMD.
I'd prefer a GPGPU solution that's not vendor/OS locked. Can you name any that are still relevant?
Since OpenCL is dead for practical purposes CUDA is the obvious choice. In comparison to Metal, using the given metaphor, CUDA is not an island but the sea.
Yes, I don't like the situation, but it is what it is