Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

3D Rendering on Apple Silicon, CPU&GPU

bombardier10 said:
https://browser.geekbench.com/metal-benchmarks 😃
Five years have passed, and Apple's GPU hasn't evolved.
Or maybe it's starting to catch up to the RX 6900 XT )

In professional applications, power consumption isn't an issue.
Suffice it to say that you might have to pay $100 for an hour of cloud computing.
And if you have a powerful graphics card and can do it on your own computer,
just calculate how much you're saving.
Click to expand...
You are both right and wrong there I think. Geekbench is bad for measure usable performance. Just look at any application that tried to harness that theoretical power. Blender is a good example where you can see how that card actually performs through 3 different APIs. looking at AMDs own HIP interface you see perf at about a m4 pro.
That said, I also had hoped that apple had a beefed up GPU variant that competed at least with the current top of the line cards of the same generation. As it is now they are at least 2 gens behind with their best offering. Here's is hoping that the m5 ultra will finally be comparable. tbh, if it had been released right now, it would be a very strong offering don't you think? Sadly there is a risk that we get it late this year or even worse a m4 ultra that will feel very "meh"
 
  • Like
Reactions: novagamer
bombardier10 said:
https://browser.geekbench.com/metal-benchmarks 😃
Five years have passed, and Apple's GPU hasn't evolved.
Or maybe it's starting to catch up to the RX 6900 XT )

In professional applications, power consumption isn't an issue.
Suffice it to say that you might have to pay $100 for an hour of cloud computing.
And if you have a powerful graphics card and can do it on your own computer,
just calculate how much you're saving.
Click to expand...
What use cases come up for you that don't work on an M3 Ultra, M4 Max, or M5 Max?

The advancements in M5 are really good, and I think M7 will be around when we reach full parity with the best consumer PC which is not bad considering where Apple started. I own a 5090 in a PC so I'm not coming from a place of blind bias, either.

Practical use matters more than benchmarks and Apple is closing the gap to a fair degree, especially where it matters at least for what I do. Having both is a nice privilege for now, modulo component availability.

Be specific so we can understand, or else respectfully I think you're wasting your time looking at charts.
 
  • Like
Reactions: Homy
“Five years have passed, and Apple's GPU hasn't evolved.”
For me it doubled in Maya projects, especially those big one at least 3 times from the introductions of ray tracing
And i still have to wait to see the m5 max/ultra
Also some said tha even in gaming from m1 max to m4 max has double or tiple the fps
 
  • Like
Reactions: sputnikBA and Homy
bombardier10 said:
https://browser.geekbench.com/metal-benchmarks 😃
Five years have passed, and Apple's GPU hasn't evolved.
Or maybe it's starting to catch up to the RX 6900 XT )
Click to expand...
M1 is 30,000
M5 is 2.5x that, at 75,000
M3 Ultra is 7.5x that, at 230, 000.

I welcome technically informed discussion in this area, an area I know little about.
But here's the thing - claiming 2.5x or 7.5x counts as "not evolved" is not technically informed...

Do you have useful information to contribute going forward, or is it all going to be this level of delusion? Because if I can't trust you to get basic arithmetic correct, why would I trust anything else you say?
And if I can't trust what you say, then why would I not use macrumors' convenient suite of user controls to simply mute you?
 
  • Like
Reactions: Luftkopf and Homy
MayaUser said:
“Five years have passed, and Apple's GPU hasn't evolved.”
For me it doubled in Maya projects, especially those big one at least 3 times from the introductions of ray tracing
And i still have to wait to see the m5 max/ultra
Also some said tha even in gaming from m1 max to m4 max has double or tiple the fps
Click to expand...

name99 said:
M1 is 30,000
M5 is 2.5x that, at 75,000
M3 Ultra is 7.5x that, at 230, 000.

I welcome technically informed discussion in this area, an area I know little about.
But here's the thing - claiming 2.5x or 7.5x counts as "not evolved" is not technically informed...

Do you have useful information to contribute going forward, or is it all going to be this level of delusion? Because if I can't trust you to get basic arithmetic correct, why would I trust anything else you say?
And if I can't trust what you say, then why would I not use macrumors' convenient suite of user controls to simply mute you?
Click to expand...

MR had even an article about the evolution of M chips last year.
www.macrumors.com

Five Years of Apple Silicon: M1 to M5 Performance Comparison

Today marks the fifth anniversary of the Apple silicon chip that replaced Intel chips in Apple's Mac lineup. The first Apple silicon chip, the M1, was unveiled on November 10, 2020. The M1 debuted in the MacBook Air, Mac mini, and 13-inch MacBook Pro. The M1 chip was impressive when it launched...
www.macrumors.com www.macrumors.com

Here's how the M5 measures up to the ‌M1‌, per Apple's M5 specs:

  • 6× faster CPU/GPU performance
  • 6× faster AI performance
  • 7.7× faster AI video processing
  • 6.8× faster 3D rendering
  • 2.6× faster gaming performance
  • 2.1× faster code compiling
Geekbench comparison scores:
  • ‌M1‌ single-core - 2,320
  • M5 single-core - 4,263
  • ‌M1‌ multi-core - 8,175
  • M5 multi-core - 17,862
  • ‌M1‌ Metal - 33,041
  • M5 Metal - 75,637
 
  • Like
Reactions: sputnikBA and OptimusGrime
thenewperson said:
Did this happen?
Click to expand...
DLSS 5 happened.

this-is-crazy-dlss-5-off-and-on-on-god-of-war-kratos-v0-b35e0t3bsgpg1.jpeg.webp
 
  • Haha
  • Like
Reactions: crazy dave, schnaps, Homy and 5 others
Apologies if answered already, don’t want to read 109 pages…. 😬
Does Octane on M Macs utilise all available memory for the GPU, or is it limited.
Just curious if loading a massive scene which would fail on say a 32GB GPU (in a 7.1) due to lack of vram.

Or am I barking up the wrong tree here.
I haven’t used octane that much.
 
Last edited:
leman said:
Yes, it’s the DGX Station that has been expected for several months now - https://www.nvidia.com/en-us/products/workstations/dgx-station/
Click to expand...

DGX Station is mostly a 'retread' from last year. Press release from March 2025.

https://nvidianews.nvidia.com/news/nvidia-announces-dgx-spark-and-dgx-station-personal-ai-computers

Last year they had pictures of a Nvidia styled box as the enclosure in the background of some of the materials , but more of the photos were of the actual board. For example , again March 2025:

www.servethehome.com

NVIDIA DGX Station GB300 Edition Launched Without a GPU

The NVIDIA DGX Station GB300 edition packs a 72 core Arm CPU, 800Gbps ConnectX-8 networking, a 288GB B300 GPU, and more
www.servethehome.com www.servethehome.com

There are two new changes. First, the 'special sauce' dribbed over it this year is on the right hand side of the page you linked. NemoClaw. Now there is a 'killer app' for it that should goose the number of units sold.

Now have 3rd parties shipping these. So there is a Dell system now.


dgx-station-jensen-huang-andrej-karpathy-nvidia-1920x1080-1.jpg

NVIDIA DGX Station GB300: The Machine on the Desk​


https://blogs.nvidia.com/blog/gtc-2026-news/ [ In the "Machine on the Desk" section ]

Nvidia is somewht making lemonade out of lemons here. They announced pragmatically vaporware last year and are finally shipping this year. Late year there was no timeframe mentioned for this system. So they have wiggle room.

Probably the dual edge sword of reusing the server motherboard basis is that didn't have 'extra' product to put GB 300 into these. So not shipping until a year later. I think the run-rates here are low enough that Nvidia trying to sell these directly is too competition for the system vendors. Nvidia will punt the inventory 'lumps' onto the system vendors.

DGX Spark SoC still hasn't launched on Windows (with its more consumer name). If NemoClaw will all them to sell lots more in the current configuration... they will. Where MediaTek and Nvidia are going there seems to be moving slowly. Nvidia is playing with letting system vendors wrap their own enclosures around DGX Spark also.
( in addition to the Nvidia model being Amazon.com, BH Photo , BestBuy , etc). As long as they sell, they probably are not in a hurry.


leman said:
Essentially a data center board packaged as a desktop machine. Expect these things to cost an arm and a leg. It’s not something available directly to end customer. This is specialized equipment.
Click to expand...

If Dell is selling a variant, they probably won't be that picky who they sell it to. You have to call and there is no quote on page. But if have a 1600W outlet plug, but probably won't sell on a simple credit card payment or anonymous buyers . 🙂

www.dell.com

Dell Pro Max with GB300, Ubuntu and NVIDIA DGX | Dell USA

Shop Dell Pro Max with GB300 and manage, prototype and run AI models locally without using the cloud. View more at Dell.com.
www.dell.com

What Dell probably doesn't want is a bunch of "try it and return it (maybe after doing Youtube video) " or "buy it and toss it onto Ebay" folks. They need to filter out those folks .
 
bombardier10 said:
https://browser.geekbench.com/metal-benchmarks 😃
Five years have passed, and Apple's GPU hasn't evolved.
Or maybe it's starting to catch up to the RX 6900 XT )
Click to expand...

Where is the RX 8900 XT ?
AMD skipped a generation on the biggest GPU to focus on the middle of the line up. Apple is basically doing the same thing. They skipped M4 Ultra to work on more pressing issues. If it is a pissing content of biggest , most expensive GPU possible , Apple isn't playing that 'game'. AMD doesn't even play that game all the time either in consumer space.

If look at GPU peformance of plain M1 , M2 , M3 , M4 , M5 and see no performance increases there isn't much to discuss. You don't want to see 'progress'. Can see also in how the Mn+1 Max gets incrementally closer to the Mn Ultra on each iteration also. There is progress all over that metal benchmarks chart if actually look at it. It is more than which are the 2-3 lines at the top.

bombardier10 said:
In professional applications, power consumption isn't an issue.
Click to expand...

Power probably was an issue why there was no 6900 Duo from Apple. That benchmark chart is interesting :


AMD Radeon HD W6800X 225732
....

AMD Radeon HD W6900X 220137

The W6900X sucks up more power than the W6800X and yet the number is lower.

'Power" is a misdirection. The top two slots are overclocked Windows market 6900's that likely were cheaper than the W6900X. Much of the bristling is really about that; more commodity pricing. Not 'power'.

Apple could have engaged in a 'pissing match' and constructed Ultra that were $4-5K more expensive. Pretty likely 'give me more power consumption' crowd largely would not have bought them, so they did not bother. There is little advantage for Apple to play the "spend any cost to be 'King of the bragging rights Mountain' when they have a whole GPU ecosystem to build.
 
  • Like
Reactions: Homy
Nice!

Has anyone tried MoonRay? I built it for Houdini and thought it was decent, although the way shaders were setup weren't particularly user friendly. Had been hoping to use it as a replacement for KarmaXPU on the Mac, but at the time it didn't fully support GPU (it would use the gpu for intersections etc, but not for the actual shading and rendering).

If anyone else has given it a shot, would be curious to hear their experience, particularly as to what makes it unique.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back