M4 Max 32-core 36gb or M3 Max 40-core GPU 64gb

So I bought an M4 Max 36GB and spent a lot of money... But I'm really not blown away with the speed for the price. I've always been a firm believer in getting the best computer you can afford and previously had an M2 Macbook air with 24gb memory, which is honestly a great machine, and while this is a little faster, it's not mind-blowingly so, for literally double the price. People are now telling me that the 36GB memory is what's limiting me. Basically it's rendering Fusion effects in Resolve that's taking the time.

Upgrading to the 64gb M4 Max is out of the question because it's eye-wateringly expensive.

So my options are:

1) I can get an M3 Max 64gb for around the same price, with a 40 core GPU (instead of this 32), does anyone know if that be faster for what I'm trying to do?

2) Return it and wait for the M4 Macbook Air

Macs are great for DR. But not as much for Fusion, due to the extreme prices for RAM.
Fusion is eating RAM for breakfast, if you are doing more than simple titling. As soon as it's running out of RAM, it starts swapping to the far slower SSD, even ruining it in the long run.

If you want to stick to MacOS (which I like for stability), get at least 2 GB of RAM per GPU core.

When you need ram / vram there is no substitute for having more. my main machine is a M1 ultra 128gb. Fusion will use all that RAM some days. The OS is also heavily caching as well. Fusion has gotten a LOT of GPU optimization over the last few releases where it uses the GPU cores more reliably than CPU fallback.

What I can tell you is this, I have 2 Mac studios. the 128gb ultra and the 64gb M1 Max. The ultra machine generally renders 2X faster with big projects than the Max machine. Both GPU cores and RAM matter. Forget the M3 Air for serious Fusion work. Also better to run Fusion standalone for large projects. Then you don't have all the Resolve resources in use.

I'd get a M2 Ultra 128gb 1tb SSD if you need something today. If you got the MacBook Pro which seems to be the only M4 max machine, IMO return it and either get more ram, and a 1TB SSD is fine as long as you keep cache and media on external SSD, or wait until the Studios get a M4 update with presumably Ultra vers.

you also haven't said if the real problem is media storage bandwidth either...

honestly all these Macs are silly fast. you probably don't want to hear about 2000 or so, running a Inferno on a SGI onyx that filled one entire rack, and another rack filled with 36gb SCSI drives. the noise and power was insane and its not even close to what modern laptops can do now.

honestly all these Macs are silly fast. you probably don't want to hear about 2000 or so, running a Inferno on a SGI onyx that filled one entire rack, and another rack filled with 36gb SCSI drives. the noise and power was insane and its not even close to what modern laptops can do now.

100% agree. The machines out in late 2024 are screamingly fast, radioative-hot compared to what we had 6 or 7 years ago. And they're only going to get faster.

I think the one to wait for will be the M5 Mac Studio (or the M5 Mac Pro), which some predict will hit by June 2025 for WWDC. We'll see what happens next year.

I just bought a Mac Studio M2 Ultra 64GB instead of my Macbook Pro M1 MAX.
I think it was a good change. The Studio M2 is absolutely silent, even during long rendering sessions, and the speed is incredible. You have no thermal problems with the Mac Studio systems, then they have a good cooling system, which means no thermal trottle.

Dirk PEL

Thanks for the advice. I’ve ordered the 64gb M3 Max to compare, and will return one of them. I would love a desktop but I travel a lot unfortunately. 128gb would be great but even the refurbs are way too expensive.

Most of my projects are less than 2 minutes in duration but with a lot of effects, so caching is slow. The other time I noticed a slowdown recently was in making a matrix of 9 clips and playing them all back at the same time. Reducing playback resolution to quarter and even turning off effects couldn’t give me smooth playback, I’ll see if the more gpu cores and RAM change that.

All media is on external SSDs.

All media is on external SSDs.

All clips of that matrix on one or distributed?

All clips in the same folder on the same SSD.

Depending on their data rate, that might be a part of the issue.

OK yes good point. I just ran a test bringing all that media onto local storage and I would say that the playback without caching was a little better. At quarter playback resolution I was getting about 17fps of 25.

These are short clips shot in XAVC S 50fps 4K.

My SSD gives me on average about 800MB/s read speed.

if your SSD is 800MB/sec thats your problem. its fine for 2-3 streams, maybe 4 reliably. I'd suggest a TB3+ nVME external case with a 4TB m2 module. I have one ( project on nVME) thats 3500mb/sec, averages 3000mb/sec that handles a couple streams of 4K 400mbit h265. some of it is 60fps @ 24, lots of native 24, filters, grades... in a 8K wide comp

The 2 factors here are storage speed and the hardware decoders for h264/265. Playing 9 streams of 4K @ 50fps is a LOT. The bare drive needs to deliver 9X ( assuming base data rate ) 400mbit = 450MB/sec, and if its higher rate 720mbit = 810MB.sec or more than your storage can never ever deliver that. even at 450MB/sec not all SSD's sustain that rate and may go up and down a bit. Of course the answer is better grade of SSD with more reliable sustained data rates. LGOP vs Iframe is also a factor in how well / fast the streams get decoded. if not iframe might want to consider transcoding a few clips to ProRes to see how that goes. file sizes should be similar and worth a quick test to convert 2-3 streams to ProRes Std 422 as your prores decoders are not bing used, yet work really well.

on my main rig I have a 3500MB/sec 2TB SSD on its own TB3 bus dedicated to cache. its actually a bit too small for my needs. it requires manual clean up once a month as R doesn't clean up its cache as well as it should.

The answer is in part more hardware decoding which means Ultra machine ( pretty sure ) still offers the most of them - 2 complete sets do to the dual chips inside. Go through the CPU specs carefully to see which ones have the most decoder units. the M4's may of boosted that a little but I haven't kept track of that particular detail of how many.

The Apple M4 Max has dual encoders/decoders.
But if you don't desperately need a laptop, I'd wait for new Mac Studio machines if you need such throughput regularly. But if that was for testing purposes only, I'd get a basic MBP for preparations on the road and a new Mac Studio when they arrive.
I can confirm that I get much higher data rates with a TB enclosure (by Acasis) with a NVMe drive inside. Two of those and the source clips distributed to them should help.

Sorry, my last post was misleading.

I moved everything to my internal SSD (5000MB/s) to test that and that's what gave me 17fps out of 25 at quarter resolution. Better, but doesn't fix the problem.

My external NVME gives me 800-900MB/s so it's agreed there could be a bit of bottleneck there.

I bought an M4 Max 36GB... But I'm really not blown away with the speed for the price....previously had an M2 Macbook air with 24gb memory, which is honestly a great machine, and while this is a little faster, it's not mind-blowingly so...with a lot of effects, so caching is slow. The other time I noticed a slowdown recently was in making a matrix of 9 clips and playing them all back at the same time. Reducing playback resolution to quarter and even turning off effects couldn’t give me smooth playback, I’ll see if the more gpu cores and RAM change that...XAVC S 50fps 4K

4k/50 XAVC-S is Sony's version of H.264. That can be 8-bit 4:2:0 up to 10-bit 4:2:2. Which was it?

I have edited hundreds of multicam interviews shot on 4k/29.97 8-bit 4:2:0 XAVC-S. That was impossible without proxies on even an Intel iMac Pro, but on an M1 Max or above, it's OK without proxies or optimized media.

Smoothly editing a nine-camera multicam of 4k/50 XAVC-S would be impossible without proxies on my M1 Ultra Mac Studio, and it has 128GB RAM and a 64-core GPU, two decode engines and four encode engines, plus four ProRes decode/encode engines.

If the XAVC-S is 10-bit OR 4:2:2, decoding cannot be hardware accelerated on any Windows machine using nVidia or AMD GPUs.

The M4 series has about double the per-core CPU performance over the M1 series and is significantly faster than the M3. The M4 series GPUs are improved, but if you're using lots of compute-intensive Resolve effects on high-res material, almost no machine can provide smooth playback at full resolution without caching, proxies, etc.

If you think your M4 Max MBP has a memory constraint during certain tasks, you can check the memory pressure and "swap used" parameters on Activity Monitor's memory tab. However that doesn't necessarily give good insight on CPU/GPU contention to unified main memory.

I did an XCode Instruments trace of a 128GB M1 Ultra GPU bandwidth to main memory during a Resolve Studio 19.1.1 export from a timeline using Gaussian blur, Temporal/Spatial NR at 3 frames, better quality and 70% settings, and also Depth Map. See attached.

It only used an average of 140 GB/sec (read) and 62 GB/sec (write), out of a total 800 GB/sec available between CPU & GPU. Of course the machine had lots of RAM, so maybe that doesn't prove anything since the question is how rapidly does performance degrade in a unified memory system as RAM config decreases.

For those interested, speed comparison of the 2 machines purely for my use case. M3 Max was faster with the extra RAM, and the 2tb storage and €100 discount were the kicker. Returned the M4, although it's a shame cause I loved the display on that thing.

Not convinced by the battery power on the Macbook pro, feels like the Macbook air had a longer battery life. also, a couple of times now the fan has sped up so much that people in the next room ask if everything is ok... The sacrifice for faster processing, I guess.

In response to the above yes it's 10 bit footage.

Just watched this video.

I noticed that at no point did the M4 32 core go into yellow or red memory pressure.

Do you think it would be suitable for 2-3 streams of 4k? I would love to buy the binned chip, but don't want to find that resolve needs more than 36gb ram for 2-3 streams. I do 10-bit 4k talking head interviews.

Do you think it would be suitable for 2-3 streams of 4k? I would love to buy the binned chip, but don't want to find that resolve needs more than 36gb ram for 2-3 streams. I do 10-bit 4k talking head interviews.

50fps or 60fps will be rough on anything. At normal speeds (24fps / 30fps), it should work pretty well.

Do you think it would be suitable for 2-3 streams of 4k? I would love to buy the binned chip, but don't want to find that resolve needs more than 36gb ram for 2-3 streams. I do 10-bit 4k talking head interviews.

50fps or 60fps will be rough on anything. At normal speeds (24fps / 30fps), it should work pretty well.

Thanks for that. And Happy New Year.

So just to be clear: does the 2gb per GPU core apply to all functions within DR, or primarily the heavier ones: Noise Reduction, FX, Fusion, etc.?

Does cutting and colorgrading demand the same ratio (which actually would mean that only two current configurations of the new M4 chip would work: the Pro 48gb/20GPU core, and the far more expensive Max 40 core/ 64 or 96gb ram config. Crazy.

It mainly applies to heavy features, in particular temporal ones, which will load several frames at once.
Regular editing and grading will stay lower, for sure. And then, there’s no hard limit, like with an NVIDIA card in a PC, which will give you a memory warning and stop or even crash right away when it’s out of VRAM.
Apple has given their new machines an excellent memory management. It will normally not crash when RAM gets tight, but optimize its use between CPU and GPU. When that doesn’t suffice, it’s going to swap to the SSD and you’ll just notice some slowing down.
And those 2 GB per GPU core are on the safe side with really heavy stuff. Most people should be fine around 1.5 GB per core.