I see the RTX 6000 Ada and A6000 in both the hero image and the first products listed. The Quadro 6000's a 24GB 4,608 CUDA-core PCIe 3x16 card[1] and the A6000 is a 10,752 CUDA-core 48GB PCIe 4x16 card.[2]
I know NVIDIA's branding and product line naming sucks, but those clearly aren't the same cards, or even same generation of card.
1: https://www.nvidia.com/content/dam/en-zz/Solutions/design-vi...
2: https://www.nvidia.com/content/dam/en-zz/Solutions/design-vi...
Though there's one detail: Apple probably couldn't have tested against the new "Ada" one because it's not even available yet, so they still tested against the best available workstation GPU, unless I'm missing something.
"Notify Me" button on the page for RTX 6000 Ada says "You will receive an email when the new NVIDIA RTX 6000 Ada Generation graphics card becomes available." I think it's unrealistic to expect companies to benchmark against future products.
So I still don't know why their benchmark is a 2.5-year-old laptop.
I'm also noticing that Apple didn't list the generation of i9 in that marketing disclaimer, so this makes me think they're also possibly comparing against 2- to 3-year-old CPUs.
> Testing was conducted by Apple in November and December 2022 using preproduction 16-inch MacBook Pro systems with Apple M2 Max, 12-core CPU, 38-core GPU, 96GB of RAM, and 8TB SSD, as well as a production Intel Core i9-based PC system with NVIDIA Quadro RTX 6000 graphics with 24GB GDDR6 and the latest version of Windows 11 Pro available at the time of testing, and a production Intel Core i9-based PC system with NVIDIA GeForce RTX 3080 Ti graphics with 16GB GDDR6 and the latest version of Windows 11 Home available at the time of testing.
To be clear, NVIDIA (and Intel) use identical names for their desktop parts and for their laptop versions. There is the 3080 Ti (GA102-225-A1) which has 10240 shaders clocked at 1365 mhz and a 350 watt TDP, and the 3080 Ti (GA103) for laptops which has 7424 shaders clocked at (up to) 1230 mhz and a (up to) 150 watt TDP. The former does about 305 fps in userbench and the latter does 193 fps. (Until it thermal throttles) https://gpu.userbenchmark.com/Nvidia-RTX-3080-Ti/Rating/4115 https://gpu.userbenchmark.com/SpeedTest/1775790/NVIDIA-GeFor... There is an equivalent bait and switch for the Intel i9 desktop vs mobile.
It's unfortunate that Apple isn't making themselves 100% entirely clear on this, but to be fair, neither is the competition.
I'll admit I haven't gone spelunking down the specialist laptop manuafacturer sites, but on the surface it seems to be not an unrealistic claim.
[1] https://store.nvidia.com/en-gb/geforce/store/laptop/?page=1&... [2] https://www.box.co.uk/82TD000WUK-Lenovo-Legion-7-Intel-Core-...
> creators can work on scenes so large that PC laptops can’t even run them
You can't open a 40GB scene entirely in GPU RAM on any single-GPU system, laptop or otherwise, because there aren't any 40GB+ GPUs.
But you can open a 40GB Octane demo scene with out-of-core loading enabled on a Windows laptop with 64GB of RAM. Hell, Otoy has a demo from 2018 of a Windows system loading and editing that "worst-case" 40GB scene entirely out-of-core at 60fps.[1]
So the suggestion is that the M2 is doing it without enabling out-of-core loading because all system RAM is in-core. Which is cool, and something Otoy's CEO was boasting on the M1 MBP's release day two years ago.[2]
So why bother going through the motions of benchmarking anything like this against 2+-year-old systems, just to make a claim the M1 also made, just less precisely?
1: https://www.youtube.com/watch?v=6xE3J56pabk
2: https://twitter.com/JulesUrbach/status/1326922973790367750
The second revision of the A100 has 80 GiB of memory: https://www.nvidia.com/en-us/data-center/a100/
"PCs can't even run this!" ....when using a PC with specs too low to run the specific thing we chose to do.
"Mac's can't even run notepad.exe!" would be fair using Apple's approach to these kind of claims (i.e. choosing to run software with requirements that you know they cannot meet)
If you could run ML models properly on that machine, it would be pretty nice for inference on larger models. Now, of course, Apple being Apple they expect you to hand-code ML algos in Swift (lol. lmao.) but still, 96GB of VRAM is 96 GB of VRAM.
Edit: Just to clarify, I understand that this has 400 gbs of bandwidth, not 3,2 tbs like an Nvidia Accelerator with the same size. But the latter costs tens of thousands and requires, well, a whole datacenter probably. This allows you to run some GPT-X or Diffusion model on your laptop. In theory.
The unified memory architecture is a bliss and they are talking the truth about some popular contemporary workloads being out of the reach of the most PC.
...
> 4. Testing was conducted by Apple in November and December 2022 using preproduction 16-inch MacBook Pro systems with Apple M2 Max, 12-core CPU, 38-core GPU, 96GB of RAM, and 8TB SSD, as well as a production Intel Core i9-based PC system with NVIDIA Quadro RTX 6000 graphics with 24GB GDDR6 and the latest version of Windows 11 Pro available at the time of testing, and a production Intel Core i9-based PC system with NVIDIA GeForce RTX 3080 Ti graphics with 16GB GDDR6 and the latest version of Windows 11 Home available at the time of testing. OTOY Octane X 2022.1 on preproduction 16-inch MacBook Pro systems and OTOY OctaneRender 2022.1 on Windows systems were tested using a scene that requires over 40GB of graphics memory when rendered.
Two things:
- The Quadro RTX 6000 shipped in 2018 and the GeForce RTX 3080 Ti is a 12GB card vs. the 24GB 3090 or 3090 Ti, much less a 4090. I get that it's a marketing eye-roll claim, and it's cool to see a laptop post up against those specs, but why is Apple even bothering to measure performance against 4-year-old or under-specced cards? I wouldn't expect a 40GB OctaneRender scene to run on a 12GB gaming card or 4-year-old Quadro on any system.
- If 60% of VFX workstations are running Linux vs. 11% of macOS,[1] how does the M2 Max MBP stack up against a garden-variety Linux workstation?
1: https://drive.google.com/file/d/15b-4GMTSEE9tyqeQdBfy_LZnxQI...