Is the Mac ever going to be good for 3D work?
Its been two years already since Apple released the very first Mac with an M1 processor. As expected, the transition to arm was seamless thanks to Apples, extensive experience in this area. Theyve done this way too many times before: theyve switched from 68 000 CPUs to power, PC, Intel and now arm in terms of CPU performance. The M1 chip is very competitive. However, the GPU side of things is still lacking for long time. Apple users, thats, not a surprise Apple – has the tendency of delivering products that could use a little bit more oomph in that department. In this case, though, there was hope that they would manage to finally compete or even get ahead of the competition dont. Get me wrong. The machines are certainly capable, but if youre working with 3D, there are better options available on the PC side. Although Im not looking to switch to a PC anytime, soon, Im interested in seeing the level of performance I can expect from my new Mac and thankfully, a few weeks ago, I had the chance to test 2 different Mac Studio. Setups lets talk about it Music. The tests I ran were focused on 3D related tasks, specifically GPU, rendering and photogrammetry theres, no reason to repeat video editing tests or other similar tasks. Theres, a ton of videos about that already redshift was used for the rendering part and photo catch for the photogrammetry part of that test. Photocatch utilizes Apples, photogrammetry SDK, so things should be as optimized as possible.
As for the hardware used, I tested both the low end M1 Max and the high end fully specced out M1 Ultra when it comes to rendering things seem to scale up well, Im being vague here, because I got some conflicting results and the first one, which was A couple of weeks ago, the M1 Ultra was almost twice as fast as the M1 Pro, but yesterday, when we ran the test again to make some adjustments, the results were out of whack. The M1 Max was faster than the M1 Ultra Im sure its a software issue, but yeah for now well have to extrapolate M1 ultras performance. As for photogram 3, the results were equally unexpected, but at least in this case the results were weird from the very beginning. Before we go into details, I would say if you have a very old Dental Mac, theres, no reason not to upgrade to an M1 Ultra Mac Studio. Both systems are very capable and they manage to beat my iMac Pro. My current setup uses a 10 core 3 gigahertz CPU and a Vega 64 GPU with 16 gigabytes of memory lets start with GPU rendering the test was a 2K still from this cartoony forest scene. It uses a lot of red shaped features, GI of course, depth of field and subsurface scattering, which was key in getting the nice shading in the leaves of the trees. I kept redshifts default sampling threshold of 0.01, which is definitely an overkill for this scene, but I wanted to keep the render going for as long as possible.
In normal circumstances, I would have used a value of 0.1. I did adjust, though the bucket size. It turns out that this simple change can speed up your renders 2X or more. I found that out, as I was going through, the test with Sean astrum Sean is an amazing, C40 artist, and he has this killer set up with two 40 90s, as he was benchmarking the scene. He showed me how big of an impact the bucket size has on rendering. I definitely did not expect this massive difference, but its true once I change the bucket size from 128 to 512. My render times were two and a half times faster. So if you have enough memory, always use 512 as the default size. So how did the iMac Pro do it finish? The render in 4 minutes and 28 seconds, the M1 Max managed to be the iMac Pro, with a render time of 2 minutes and 57 seconds. Thats 51, faster than the iMac Pro its astonishing. Really an entry level setup surpassed what was previously considered a high performing Workstation. The results for the M1 Ultra, though, are somewhat confusing. It finished the render in 4 minutes and 33 seconds, which is almost 50 percent slower than the M1 Max. That, obviously, is wrong and theres something at play here: an issue with a Mac OS update, something going on with redshift Im, not exactly sure what the issue is, but when I first ran the test a few weeks ago, the results were much more predictable back then, The M1 Ultra was twice as fast as the M1 Max, which makes sense, given that it has double the GPU cores.
So if we extrapolate from that initial test, the render with the 512 bucket size should finish in approximately 2 minutes. Lets now check the PC side of things. The first setup has a 3060 TI and the other one is a PC with two 40 90 cards, thats Seans setup. It looks like the M1 Ultra has the equivalent power of a 3060 TI, which is not bad at all, thats, a really good GPU. But when we compare things with a 4090, the M1 results look insanely slow in comparison. This is a screen recording of Seans computer. The footage is not sped up at all its playing back in real time. As you can see, the two 4090s spit out the image in no time at all, and just like that, the render is done. Of course, thats. An unfair comparison were comparing two gpus to one, but even with 1490 enabled the render speed is quite incredible. What we need to keep in mind, though, is that the power draw of the 4090 is on a whole other level. Mac Studio in total draws just a portion of what a 40 90 alone needs and once we add two of these cards, the power consumption exceeds the power of net 1 but 2 max Studios. So I definitely wouldnt expect the Max studio to compete with a power of nvidias 4090. thats, the M1, though, what about the new M2 chips? Apple released the M2 Pro and M2 Max just a couple of days ago and according to Apples claims, the M2 Max is 30 faster than its predecessor.
We dont have any solid benchmarks yet. So this is the point where we enter the wild speculation mode. If the M2 Ultra will follow m1s patterns, then the M2 Ultra should be 60 faster than its predecessor, which will bring it in the performance range of the 40 90.. That would be amazing. That is if my calculations are not completely out of whack, but even if the M2 Ultra doesnt scale that well and we only get closer to 80 or 90 or 49ers capabilities, thats still really good. We will have a machine with great performance, but with much less power draw in general. I like Apples approach on power. Consumption Nvidia recommends an 850 watt power supply to run a 4090 without any issues. So if the M2 Ultra manages to reach the same type of performance but with half as much power, draw itll be a great achievement but yeah for now. If you want the fastest possible rendering machine and connecting as many gpus as possible, PC is the only way to go. A Mac cannot currently compete with nvidias, most powerful GPU Im, hoping that maybe next year, with the M3 Apple, will manage to reach this sweet spot. Amazing GPU performance, but with relatively reasonable power, draw, of course, that remains to be seen. Nothings actually announced. Yet we only hear rumors about the M3 and thats about it. Either way, though, if your Mac User, the M2 lineup, looks good so the moment the M2 Ultra comes out, Im pre ordering whatever Mac comes with it, okay, so thats.
The GPU part lets now check photogrammetry Music. I used 160 pictures to reconstruct this piece of brick. The final mesh is very detailed and its made out of 1.2 million polygons. Initially I was under the impression that Apples, photogram, 3 SDK, was taking advantage of the neural engine course at least thats. The impression I got out of Apples presentations just as a reminder. The M1 Ultra has twice as many neural engine cores is the M1 Max does So in theory, the M1 Ultra should be twice as fast, but unfortunately thats not the case. The iMac Pro finished the Euro Construction in 18 minutes and 45 seconds. As expected, the M1 Max completed the test much faster than that around 60, faster than the iMac Pro, which is an amazing boost in performance. Unfortunately, the M1 Ultra is just 17 seconds faster than the the M1 Max. I was really confused by the results. The M1 Ultra has twice as many CPU cores twice as many GPU cores and twice as many neural engine cores. So no matter the type of resource used to perform the Reconstruction. We should be able to see a much better performance overall, even if the M1 Ultra didnt scale that well, we should see something better than a 17 second Improvement. My guess is that apples for the Grand Prix SDK is at fault. Initially, I thought that photocatch might be the issue, but I tested out another app using Apples, SDK and the results were the same, so theres something going wrong there.
So, if youre doing a lot of photogrammetry work – and you want to get a Mac, getting the maxed out Mac Studio might be a complete waste of money. Im, hoping that Apple will fix this issue and we will get to see some better speed improvements but Im not holding my breath it might take months before we see a fix or maybe even longer, Music. Weird issues aside, it looks like apple is on his way to deliver some really amazing chips, even if the M2 Ultra wont manage to catch up to nvidias offerings, I get the feeling that were awfully close. Will the mythical new Mac Pro get us there thats? What a lot of people are expecting, but if the rumors are true, it looks like the Mac. Pro will only have an M2 Ultra, basically itll, be a Mac Studio with expansion slots, so no fancy M2 extreme or anything of that sort either way. Whatever the case may be, I think we have a bright future ahead. Lets not forget that M3 is just a few months away, supposedly its going to be a three nanometer chip, so we should expect faster performance and improved power efficiency. For now, though, Im going to keep chugging along with my iMac Pro once a new version of Mac, Studio or Mac Pro comes out Ill, be all over it. So whenever that happens, expect an extensive review by me.