Blender on the M3 Max 

I agree. They are always “wrong” and “regret”ing 🤣 There are too much silly reviews in RU-vid

Thank you so much for the kind words! For what it's worth I start off everything I do by being terrified I'm going to waste someones time if I do it wrong - hopefully that comes through by respecting your time and giving you usable data.

I have the same expectation for the m3 ultra being tested in Blender

Not mentioned in the video, but their was a strange sliding scale with performance - at 60 samples and lower the Max was actually *faster* than my 3080 using Optix - greater than 60 and the 3080 gradually becomes faster. As the 4090 is around 2x the performance of the 3080 it's absolutely possible the Ultra chip will indeed match the 4090's performance at lower sample counts. That's not at all what I would have expected a few years back. It also gives me hope that whatever odd performance hit we take at higher sample counts is fixed.

@@matthewgrdinic1238 That is great news! I'm primarily concerned with a new machine being usable, and then hopefully somewhat competitive. It doesn't have to be as fast or faster than a top PC GPU rig, but it also can't lag too badly, etc. It looks like the M3 is finally there. Yeah, hopefully it will get even better in the future, but I was worried with the M1 & M2.

@@matthewgrdinic1238 Could this be due to scene build time being faster on mac?

First thank you for the kind words! On custom fan curves - oddly no, strange considering that's the *first* thing I do to any PC I own (hello getfancontrol.com!) Anyway, your characterization is pretty accurate. Fans are generally off or remain silent for most work but when activated they ramp up quite quickly. As you can hear in the video (10:58, try using headphones), they can also get quite loud when pushed hard. Even then, my only real beef with the factory curves is it feels like they should spin down more quickly.

Are you speculating you can run this on battery and get this level of speed that he is showing off? Or is that a proven fact?

@@citizenclown I have an M1 Pro MacBook and it runs exactly the same on battery as plugged in. So it’s not a speculation. The Apple silicon is real in terms of efficiency. Though, if you need a pure powerhorse for 3D rendering, a proper PC build will still be a better bang for the buck. The thing I don’t like about PC is windows. I personally use Linux on my tour, for the most time. And I have Windows 10, installed to be covered for some apps that are not supported.

@@vladartiomav2473 yah, I'm a Linux enthusiast myself but it's not better than Windows. I use windows because it just works. When I have time to be finicky, I tinker with Linux.

when you render a 30 seconds long animation its a huge difference

@@jbdh6510 completely agreed, but when do you render the final sequence on your laptop anyway, even if it has a 4090.

Same! I have no real work on AE and Blender, but I've used these as hobbies for years since early high school and was looking performance for both AE and Blender together on the new MBP. Hoping the baseline 32 gb ram will be enough for casual hobby editing and uni lol.

Don’t mind. I founded it in the comments below! „So this is the 14/30 Max with 36GB of RAM.“

Edit > Preference > System > Metal

Correct - As I call out in the video optical denoisers are now so good that we're usually wasting time *not* using them. Because of this I run most tests in a manner we as actual users of this machine would, meaning fewer samples and denoising enabled. In some cases the results are not only much faster but of better quality (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-ht_8ISvtIcE.html). Of course this means any comparison on your own hardware should match the videos sample count.

Thanks for the clarification, and sorry, I missed that you mentioned it in the video @@matthewgrdinic1238 I just wanted to make sure as I was rendering the same scenes using the original samples. I will re-run them so I can make a more appropriate comparison.

I wish we have more info about m3max and unreal engine. Im still waiting to buy it and dont onow if its a good idea

@@sahogrygoryan7993 Same, i'm on the edge of getting another laptop but if the M3 Max runs UE5 correctly it would be great

@@canigetuuhhh i’m waiting since 2021)) its getting a shizofrenic idea for me👀remember times when official epic guys wrote that apple silicon will never support rtx features as lumen and nanite and now we have m3) theoretically it should work all properly now, but looks like we just wait for epic to finish their native m3 support to be sure…. I have also read your comment about Metahuman. As i understand “as a UE5 plugin” it should also work fine on M3. But still haven’t figured a lot from real users(( looks like they are rare as hens teeth.

@@sahogrygoryan7993 Oh okay you're very patient lol but yeah you're right, can't find anything except few guys who struggle with ue5 on their m3 macs which kinda scared me but I hope epic will keep supporting m3 as you said, that would be perfect

@@canigetuuhhh yeah… Im in the middle of war in Ukraine, so got used to wait. As I see UE is only becoming easier to use every year, so if u wait some more you dont need to learn things cause they get automated , as it was with lighting and Lumen. Im sure with Apple VR products getting closer we will have better support of UE on macs. There is a growing demand from app devs and game devs. Finaly , looks like with m3 rtx we got everything from hardware side, so the thing remains is epic software. I guess buying m3 for UE is finaly good idea.

Apologies - 14/30, and added to the description :)

How many samples were you using for each of these?

Same settings as in the video@@jeanclaudethedarklord6205

Great question: So this is the 14/30 Max with 36GB of RAM. Of course it all depends on the model, but for reference a 1024x1024 in XL consumes around 20 gigs, leaving a good amount of room. The new SDXL-Turbo models are a bit more RAM hungry, and I am starting to see memory pressure. So perhaps a good compromise position would be go for the top end Max chip with 64 gigs. That comes in at $3,899 whereas 96 in the base chip would be $3,999. Then, take the saved money, sell this machine next year, and get the latest and greatest.

@@matthewgrdinic1238 thanks a mil for the reply and recommendation. 64gb it is!

@@aloto1240im also thinking of the 128 gb. What would you advise? Thanks.

@@matthewgrdinic1238what do you recommend? 30 vs 40 cores unbinned?

@@psysword Hey I decided to just get a an external hard drive - Samsumg. Then I am using Runpod and colab for Stable Diffusion.

I have an RTX 3090. Can you tell me what takes it 20 seconds to do with the 3.3 splash? Render it? What settings do you have it on?

A great question - GPU's on Apple Silicon are packaged on die with the CPU, but so far as I know, present the same as any other external GPU would. One key difference however is this setup allows the GPU and CPU to consume the same memory pool, leading to, in my case, the GPU presenting as having 36 gigs of (V)RAM.

Apologies for not calling this out more clearly - 14 / 30 with 36 Gigs of RAM.

@@matthewgrdinic1238 So, the upgraded 16 / 40 with 48 gigs will be a pretty decent.

Depends on what blender version you use, 3.5 added Metal rendering support and slashes the previous render times by like 4x

Most likely default render resolution on each demo file, so it varies. You can download the files and test 'em "Blender demos"

Maybe, maybe not. But it definitely makes it useless without this information. He does change some settings so it’s not clear if he’s changed anything else.

He doesn't change the resolutions

Good question - aside from samples nothing else was modified from default.

@@matthewgrdinic1238 I see. where can I find the files? Or do you know roughly what resolution they were?

Thanks for the feedback! One of the key points of this video was to note the difference between raw numbers and experience. AI De-noising plus this chip's hardware ray tracing has created an experience that simply didn't exist in this form factor on the Mac. It's not so much about raw numbers it's about what this machine *allows* from a workflow and quality perspective. Same idea but in a slightly different context: I recently picked up a 4090 for my PC (video review incoming!) and my biggest surprise has been how good frame generation is. With the release FSR3 and its hardware independent frame gen though - it's suddenly got me thinking did I really need the 4090? Literally overnight most hardware has the capability of becoming twice as fast - that's just nutty. The larger point is I believe technological advances are expanding what performance means - for rendering that means fewer samples but made up for by de-noise, for gaming lower resolution negated by up-scaling and frame gen.

That's what's so compelling about these tests and what they may mean for an M3 Ultra. At 60 samples (across a variety of scenes), the M3 matches my 3080 (using Optix). As a 4090 is around 2x the performance of a 3080 it's actually possible an Ultra could come close to that monster. Of course by then (late '24) we'll likely have the 50 series but still, quite an achievement, and great news for Mac users.

4090 is twice as powerful as the 3090 and the 3090 is 10-20% more powerful than the 3080.

I don't think the m3 ultra will be close to a 4090. Maybe it will be 50%. Mac is doing great with their m chips, but, they focus more on power consumption and optimizations. That's why they will lag behind Nvidia for the foreseeable future.