Comparing The M3 Pro and M3 Max in AI, Gaming, and 3D workloads.

in this video I'm following up on a test from a few days ago where I pitted the top-of-the-line M3 Max against my outgoing M1 Pro to no one's surprise the M3 Mac was much much faster to the point where I personally consider it to be an acceptable machine for AI workloads using stable diffusion and llama based llms the only problem is at 4,000 us before taxes that machine presents two main problems first that's a lot of money and second for AI in 3D work you can build a PC with dedicated GPU that's not just faster but several times faster in other words the Top Line Mac model in my humble opinion is not a very good value proposition and so that's why in today's video I'm going to be bringing us back down to earth and looking at machines that are $900 and $1,600 or less expensive respectively the base M3 Max with 14 core CPU and 30 core GPU and the topspec M3 Pro with a 12 core c CPU and 18 core GPU the two questions we're going to answer are what type of real world performance you'll get from these machines and second do you actually need the topspec M3 Max machine so like the previous video let's start with chat now as noted in the last video chat using llama models is a bit strange right now mostly because the answers generated between the M1 and later models differed in a really significant way namely the M1 generates shorter answers for text based prompts but but then from mril 7B instruct basically a model trained on coding questions it repeatedly generated more practical and usable code samples so for example create a JavaScript program that bounces a ball in an HTML canvas element the M1 created a full HTML page the M3 is just the JavaScript it's really odd now that said the previous conclusions from the last video still stand the M3s start generary response instantaneously whereas the m M1 takes a beat or two despite that as the M3s tend to generate longer responses all machines finished in around the same time okay let's talk image generation so here the newer machines make a much better impression but before I give the exact numbers let me make a quick analogy my M1 Mac is to stable diff Fusion as dialup was to the internet the M3s are more like early Broadband so I think of the staple Fusion as like a search engine you typically don't get the best results straight away so the faster you can query the better your experience so bottom line for 512x 512 images the M1 took 40 seconds the M3 Pro 10 the M3 Max I have here six and the top of the line Max 4 in practice all M3s feel fast enough of course it has to be said my RTX 3080 equip pc 1 second using the custom tensor cores now this disparity is what I mean when I say say the topend model it just doesn't make me feel like it's a great value for far less you could just build a PC and be done with it still apple is making good progress in this area and who knows software tweaks could extract further performance okay let's talk 3D via blender 4 so this is definitely a case where the hardware rate tracing in the M3s is immediately felt this means a good 2x speed up over previous generations right off the top now the kicker here is something I noted in the last video which is the denoiser capabilities and blender are fantastic it does such a good job in fact that in most cases you can have or even quarter your sample count with almost no loss in quality as samples are one of the biggest influences on speed roughly speaking half the samples means half the render time now even despite that the rub here is my M1 was so slow that even when you reduced your sample count to half of what the M3s used it was still four times slower so yeah the hardware rate tracing in the M3 is the real deal so focusing on just the M3s then one interesting thing I found was the ray tracing cor seemed to scale nonlinearly that is the M3 Max with its 30 core GPU render Tugboat in 34 seconds whereas the M3 Pro with 18 cores 56 this pattern repeated in just about every scene I tested where the max was typically two times faster than the pro sometimes more now should quick note the same difference was found with the EV rendering engine and even in more demanding renders neither of the M3s fans became audible in short the M3 is a huge win for Ray tracing and even more so with the max variant so to finish things up I wanted to have some fun in talk gaming now to be clear you should not buy these machines for gaming it just happens to be that the hardware is now good enough in the laptop pH factor that well not only is it possible in some cases actually quite good good for gaming the headline then is this the max in its 14-in form factor pumps out some impressive performance but it comes at the cost of an aggressive fan this machine runs hot and loud now that said if we lower resolutions to say 1920 by 1080 the Maxis fans rarely turn on on the flip side the M3 Pro runs cooler and is a consequence usually much quieter now let's talk specifics starting with Resident Evil village now both machines are capable of locked 60 HZ at non-native resolutions for example 2560 X 1440 but the Max has a much easier time of locking to 60 frames per second and in some cases that magical 120 htz which is the displays native refresh rate now speaking to the thermal challenges noted a moment ago when we set Resident Evil to a 120 HZ refresh rate the max locks to 120 frames per second while drawing in 30 Wat Watts the pro 81 frames per second while drying in 21 Watts now unfortunately that extra 10 watt sends the max into a bit of a head spin and the fans really ramp up however and somewhat surprisingly locking to 60 htz actually flips the results now the max Atri draws less Power by a water cell more importantly at this easier to drive refresh rate both machines locked to 60 frames per second and were effectively silent now I bring up this power thing because if you want to take this machine on say a 3-hour plane flight every bit of efficiency helps if you keep the performance level comparable between the two machines the max actually sips a bit less power but on the other hand if you crank up the settings the max takes in much more so just something to keep in mind okay moving on to balers Gate 3 at high settings using FSR at Native resolution the pro averages 34 frames per second in the max 70 turning down the resolution to 1512 by 982 bumped us up to 50 on the pro and 99 on the max I suppose the key takeaway here is the max once again allows us to get that fantastic clean image that locks to 60 frames per second whereas to lock to 60 on the pro requires some pretty significant image quality sacrifices finally I tested Liza P at medium settins native resolution the pro clocked in at 30 frames per second the max once again doubles that to around 60 so there we have it for AI work the max is faster but not shockingly so though just enough to make repeated image Generations more satisfying experience for 3D honestly I'd say it's not even close the max is literally twice as fast and for gaming both perform quite well but the max more so generally delivering smooth gameplay though at the expense of more fan noise at higher resolutions overall I think both machines are great but if you can stretch for the max I think it's a market improvement over the pro finally to answer the question I posed at the start of this video do you need the top of line Max chip if you only live in a world with Max and your job is 3D I'd say yes rending times are still around 20% faster than the low run Max chips and those savings will pay for themselves for everyone else though I'd say the Bas Max chip is plenty fast and saves you a good bit of money

Channel: Matthew Grdinic

Views: 3,451

Keywords:

Id: AhPGgJdsmKo

Length: 8min 24sec (504 seconds)

Published: Tue Nov 28 2023