3D Rendering on Apple Silicon, CPU&GPU

[Herbert123]

Do keep in mind that the only announced product so far is the entry level laptop (albeit in a nice premium chassis with a high-end display). The intended user base are digital nomads, creatives, students, etc… this is not a performance-focused machine. The same chip goes into the MacBook Air and the iPad. The prosumer hardware will be released in a couple of months.

Of course. Looking forward to the M5's inclusion in the Studio later next year.

Yes if M5 Max has the 1.6-1-7 over m4 max boost then its more like 8400-9000, bye bye 5090 Laptop 😉
Hello 5080 desktop in a laptop with loads of Vram and close in memory bandwidth too.

Only if Apple releases a MacBook Pro with an M5 40 cores, or so. Which I doubt will happen. We should be comparing laptops with laptops and desktops with desktop models.

Obviously it will happen for a Mac Studio next year - as a desktop model still way behind the desktop 5090 in terms of expected render performance. But it's still fast. Almost as fast as a 4090. And with the benefit of 64GB or more RAM.

Win some, lose some.

 

[OptimusGrime]

Yes if M5 Max has the 1.6-1-7 over M4 max boost then its more like 8400-9000, byby 5090 Laptop 😉
Hello 5080 desktop in a laptop with loads of Vram and close in memory bandwidth too.
Memory bandwidth should be close to 700 GB/s

True. Although they would need equivalent increases in memory speed etc. I wonder how feasible that is?

Afaik the fastest LPDDR5 ram is 10700. Would they use that? HBM? LPDDR6?

 

[OptimusGrime]

Of course. Looking forward to the M5's inclusion in the Studio later next year.


Only if Apple releases a MacBook Pro with an M5 40 cores, or so. Which I doubt will happen. We should be comparing laptops with laptops and desktops with desktop models.

Obviously it will happen for a Mac Studio next year - as a desktop model still way behind the desktop 5090 in terms of expected render performance. But it's still fast. Almost as fast as a 4090. And with the benefit of 64GB or more RAM.

Win some, lose some.

?

The M4 Max has 40 cores currently. Unless I am misunderstanding you?

 

[richinaus]

Would the M5 studio Ultra be getting close to a 5090 then?

 

[l0stl0rd]

Not even to think if the AMD chiplet rumor is true and the will do something along those lines:

Then we are talking close to 5090 desktop contender.

And even if the just to the typical ultra and scaling is perfect or pretty good 16000 in the benchmark should not be an issue.

At the moment that would be like spot Nr2 in the ranking which I never believed I would see from Apple, slighlier under the RTX ADA 6000. If the manage that it really might be worth thinking about an Ultra 😉

Not to forget 5.5 USD for the 96 GB M5 Ultra would be cheap for an RTX ADA 6000 class GPU...

 

[l0stl0rd]

Of course. Looking forward to the M5's inclusion in the Studio later next year.


Only if Apple releases a MacBook Pro with an M5 40 cores, or so. Which I doubt will happen. We should be comparing laptops with laptops and desktops with desktop models.

Obviously it will happen for a Mac Studio next year - as a desktop model still way behind the desktop 5090 in terms of expected render performance. But it's still fast. Almost as fast as a 4090. And with the benefit of 64GB or more RAM.

Win some, lose some.

And why not the M4 Max has 40 cores....

 

[Herbert123]

?

The M4 Max has 40 cores currently. Unless I am misunderstanding you?

Oh sorry - I meant that the M4 Max 40 cores reaches 5275 points in the Cycles benchmark versus the 5090's 14905 points. So 1.7 times the M4 Max performance would still be unable to reach the 5090 in this particular rendering benchmark.

As far as I am aware Apple will not be releasing a 40 core M5 for their laptops, correct? Only for their desktops.

But it's all speculation at this point.

 

[leman]

As far as I am aware Apple will not be releasing a 40 core M5 for their laptops, correct? Only for their desktops.

M3 Max and M4 Max have 40 GPU cores and ship in a laptop. There is no performance difference between Apple Silicon laptops and desktops (cooling notwithstanding), they use the same chips. The only exception is the Ultra.

As to how M5 Max will look like, who knows. If it ships with 40 cores and the rendering performance improvement is the same as what Apple claims for M4, then we would be looking at ~ 9000 points in Blender benchmark, which is the ballpark of desktop 5080. I don't think it will be that good, although it's not impossible — Blender in particular seems to respond very well to Apple's Dynamic Caching tech (which makes sense), so improvements in this area plus faster RT plus all the other GPU advancements (and there are plenty!) could work out well for complex async kernels.

 

[Pet3rK]

Oh sorry - I meant that the M4 Max 40 cores reaches 5275 points in the Cycles benchmark versus the 5090's 14905 points. So 1.7 times the M4 Max performance would still be unable to reach the 5090 in this particular rendering benchmark.

As far as I am aware Apple will not be releasing a 40 core M5 for their laptops, correct? Only for their desktops.

But it's all speculation at this point.

 

[OptimusGrime]

Oh sorry - I meant that the M4 Max 40 cores reaches 5275 points in the Cycles benchmark versus the 5090's 14905 points. So 1.7 times the M4 Max performance would still be unable to reach the 5090 in this particular rendering benchmark.

As far as I am aware Apple will not be releasing a 40 core M5 for their laptops, correct? Only for their desktops.

But it's all speculation at this point.

Right but why would we compare the Max to the desktop 5090 and not the laptop 5090? I think comparing the Ultra to the desktop 5090 is completely fair btw.

 

[Herbert123]

View attachment 2568301

Missed that. My bad.

 

[terminator-jq]

I remember seeing some credible rumors a while back that Apple will reserve a larger part of the M5 Pro / Max for GPU upgrades. I wonder if we might actually see an increase in GPU cores to something like 48 on non-binned M5 Max configurations.

Although I wonder if they will save core count increases for the move to 2nm with M6.

 

[leman]

I remember seeing some credible rumors a while back that Apple will reserve a larger part of the M5 Pro / Max for GPU upgrades. I wonder if we might actually see an increase in GPU cores to something like 48 on non-binned M5 Max configurations.

Although I wonder if they will save core count increases for the move to 2nm with M6.

There is a remote possibility that they are introducing die stacking, which could potentially increase the effective logic area. We do know that they have been working on it for many years. But whether this technology will arrive with M5 Pro/Max and which form it might take is anyone's guess.

 

[name99]

Of course. Looking forward to the M5's inclusion in the Studio later next year.


Only if Apple releases a MacBook Pro with an M5 40 cores, or so. Which I doubt will happen. We should be comparing laptops with laptops and desktops with desktop models.

Why not? They sell a 40 GPU core M4 Max MacBook Pro today.

 

[skippermonkey]

I use an M3 Ultra with Blender Cycles GPU rendering. I'm just going to assume an M5 Ultra will be fast enough to warrant an upgrade.

 

[innerproduct]

also m3 max user here. And from long time experience I say that in order to get a workflow change, performance has to jump about 4x. M1max to m3max almost did that for blender. Clearly a m5max at best will be 2x but it he m5 ultra might reach at least 3x. So it that is what i am waiting for as well.

 

[Pressure]

also m3 max user here. And from long time experience I say that in order to get a workflow change, performance has to jump about 4x. M1max to m3max almost did that for blender. Clearly a m5max at best will be 2x but it he m5 ultra might reach at least 3x. So it that is what i am waiting for as well.

Are you really saying that a 50% reduction in render times isn't worth it for you? 😅

 

[innerproduct]

Are you really saying that a 50% reduction in render times isn't worth it for you? 😅

No, that was at least not what I wanted to convey. What I meant was that in order for my workflow to change I need a big change, otherwise it will just be the "same old" but a little faster. 4X seems to be about the improvement needed for things to matter and be really noticeable. Of course 50% improvement or 2x improvement is also good but it seldom enables anything new; it is just convenient. And I will not buy a new device for small changes.

I mainly relate to the interactive experience here. For rendering a sequence, it seems almost idiotic to use a mac at all if you work professionally. Need fast GPU rendering locally? get few gamers pcs with dual "whatever is the most bang-for-the-buck nvidia". Need CPU rendering? Threadrippers, epics and xeon blades.... Do not need on prem? well there are many beautiful rendering farms with great prices and interfaces to your dcc. Gridmarkets for example.
That is just my experience.

 

[leman]

M5 scoring ~ 1700 in Blender benchmark is absolutely wild IMO. This is still a sub 20W (closer to 10W) GPU with merely 1280 shading units that we are talking about. Just to give some perspective, that's roughly 70% higher performance than Nvidia's Blackwell per Ghz and SM.

Apple M5 (GPU - 10 cores) - Blender Open Data Benchmark

Blender benchmark for Apple M5 (GPU - 10 cores)

opendata.blender.org

 

[dmccloud]

There is a remote possibility that they are introducing die stacking, which could potentially increase the effective logic area. We do know that they have been working on it for many years. But whether this technology will arrive with M5 Pro/Max and which form it might take is anyone's guess.

There has been some discussion that die stacking might be why M5 Pro and Max were not announced this week.

 

[Boil]

Die stacking...?

Hello, M5 Extreme in the all-new Mac Pro Cube...!

;^p

 

[crazy dave]

M5 scoring ~ 1700 in Blender benchmark is absolutely wild IMO. This is still a sub 20W (closer to 10W) GPU with merely 1280 shading units that we are talking about. Just to give some perspective, that's roughly 70% higher performance than Nvidia's Blackwell per Ghz and SM.

Apple M5 (GPU - 10 cores) - Blender Open Data Benchmark

Blender benchmark for Apple M5 (GPU - 10 cores)

opendata.blender.org

The other neat thing is that 16-core M5 Pro should get very close to the RTX 5050 while the 20-core should get very close to the mobile 5060. That's much better than previous generations where it was the 20-core M4 Pro that lined up against the mobile 4050. Obviously we'll have to see for how these new GPUs do in gaming (and other rendering engines), but that's already a substantial improvement relative to Nvidia's offerings from where Apple was before.

 

[deconstruct60]

Die stacking...?

Hello, M5 Extreme in the all-new Mac Pro Cube...!

;^p

Yeah a joke, but more pervasive , verticle 3D stacking is pointing in the opposite direction.

If Apple can put the 'power' of an Ultra into a Max then that is more likely that Ultra will be the top end of the line up. As long as Apple primarily wants to put "ludcrious power' into the MBP 16" as the objective, then that expensive chip needs to be offset with higher volume. Ultra's soaking up those maxes will be the likely path. If have stacked the I/O die uner the CPU/NPU/GPU die then already '2 high'. Going Ultra is a '3 'high' bridge; which is even more expensive.

The Studio isn't going anywhere , anytime soon. Pragmatically already have a 'cube'.

As for mostly horizontal die stacking.... Apple has already done 3 generations of that with the Ultra. Yes, there wil iterative refinements, but there is no 'revolution' change there.

 

[Homy]

The base M5 is already 60% faster than M4 in the same Blender GPU test. If M5 Max 40c performs as well it would score around 8443. That would make it faster than 5090 M, desktop 4080 and 4080 Super. If it will have more GPU cores, like 48c it would score around 10 000. That would make it faster than desktop 5080 and close to 4090. M5 Ultra would score over 12 000 which would make it as fast as 4090 D or 5090 D.

 

[innerproduct]

Yes. But then they will release it only in the mbp in spring, and put it in the studio first in 2027 if they continue the pattern they have had for some time. so once you actually have a m5 ultra on your desktop, competition is on next gen nvidia. And also the m6 max almost beats the m5ultra. Especially since there will be a node switch. With this candence, i am amazed anyone buys the desktops or ultras at all