Davinci on Apple M1 Max vs 5950X 3090 - 2x faster !!!

Here's a test many have been waiting for.

It'd be good if there were such tests with Fusion.

In the video, he compares both Davinci Resolve and Premier Pro on a fully-spec'd Apple M1 Max, versus a Ryzen 5950X with Nvidia 3090 with 128GB RAM.

Here's another similar test:

Curso.png (596.32 KiB) Viewed 21229 times

Just mind blowing, I love how Apple is pusing the limit of the industry, and makes my 5 months new PC sounds like a joke.

That's pretty impressive.

I'd like to see a test with some temporal noise reduction on.

Glad I didn't jump on a PC, I was very close to doing it before the GPU prices skyrocketed.
I'll check NR later today, my new shiny MBP is on the way.

But we don't know the details of what could be a bottleneck in that video. He doesn't show the h.264 encoding settings on the PC. Is he using the nvme listed in the description as a data drive or was the footage stored on something else? He's using an AMD CPU, we don't know if an Intel CPU would have performed better for that particular task since v17.4 included faster hardware acceleration on Intel.

Heck, we don't even know the version of Resolve he's using or if it's even the Studio version.

The MacBooks are impressive, but I don't think h.264 encoding is a test that says everything I would want to know.

That's impressive.
I love the performance of my basic M1 laptop, even without the 'Pro' and 'Max' go faster stripes ... and my 5950x desktop is more than adequate for my workflow for the time being ... but still, I'm very much looking forward to seeing how the land lies when Apple start shipping full blown desktop class machines with this tech.

Hi.

The Nvidia RTX 3090 is more than one year old and is produced on Samsungs 8nm process node, which actually is a refined 10nm process node. The new M1 max is produced on TSMC's 5nm process node. So there is two full process node difference in the node used to produce the M1 max and the RTX 3090.

The Ryzen 9 5950X was launched November 5, 2020.
And AMD will host its Accelerated Data Center Premiere on November 8, 2021 at 11 a.m. ET, showcasing the company’s upcoming innovations with EPYC processors. The new Milan-X chips will feature 3D V-Cache stack technology similar to the ones that AMD has promised for its Zen 3 based Ryzen CPUs coming to the AM4 socket in Q1 2022.

So CPU wise, is he comparing a one year old Ryzen 9 5950X with the new M1 max, and do it just 4 months before AMD's next Ryzen CPU coming in Q1 2022.

So what will you call a YouTube comparing old nVidea and AMD product, just months before next versions, with the all new Apple M1 max?

Regards Carsten.

Notice how he didn't used Nvidia hardware acceleration ON for the PC.

He basically tested Hardware on the M1 vs Software on the PC.

But we don't know the details of what could be a bottleneck in that video. He doesn't show the h.264 encoding settings on the PC. .

The guy doesn't seem technical at all, and also the results make no sense. It could be as simple as Mac version is using GPU decode and encode, and the PC version is no GPU decode(free version) and CPU encode. The default on M1 version looks to be Hardware encode but not on PC version. Not that he's deliberately trying to convey a false message, he just doesn't know any better

H.264 might be faster on the M1 even if he is using hardware acceleration on the PC as the dedicated chip for decoding/encoding on the M1 is so much newer, but that's only a small piece of the chain and to an extent it's separate from CPU/GPU performance.

H.264 might be faster on the M1 even if he is using hardware acceleration on the PC as the dedicated chip for decoding/encoding on the M1 is so much newer, but that's only a small piece of the chain and to an extent it's separate from CPU/GPU performance.

With my quick maths (and prematurely closing the calculator app) I believe the M1 (resolve) encoded at 136fps at 1080p. I know resolve (windows) can do over 300fps using Nvenc (2000/3000 series) so the windows slow down is not due to M1 having a faster encoder. If we believe this guy's figures, the m1 has a better GPU and CPU then the 16 core Ryzen and 3090GPU. It's not believable (in my opinion)

Not apples to apples at least. He's wearing a baseball cap sitting at his computer, I don't trust his technical aptitude

If we believe this guy's figures, the m1 has a better GPU and CPU then the 16 core Ryzen and 3090GPU. It's not believable (in my opinion) He's wearing a baseball cap sitting at his computer, I don't trust his technical aptitude

Every is commenting about the baseball cap guy - but the 2nd video from the Japanese guy also shows a similar advantage of M1 Max over the same PC setup. So it's two bench tests that have similar indications.

The Japanese reviewer is using 8K XAVC footage from a Sony A1 camera. Is there hardware acceleration for decoding that on his PC in Resolve? I assume so, otherwise it would probably be much slower. But so far all the comparisons raises more questions.

I'm getting tired of these videos
Being a "MacBook Fanboy" (will be ordering 14" M1Max mixed out except the SSD) I see it very easy to "doctor" such a "comparison". Just throw in ProRes RAW (in Premiere Pro), do not apply any NR, sit back and enjoy how much faster the M1Max will be

The Japanese reviewer is using 8K XAVC footage from a Sony A1 camera. Is there hardware acceleration for decoding that on his PC in Resolve? I assume so, otherwise it would probably be much slower. But so far all the comparisons raises more questions.

I think it would be HEVC 4.2.0. Resolve studio can hardware decode it but I don't remember if it defaults to hardware. I think so.

But IIRC that's also low datarate 8K . Going by memory the highest data rate is 1/3 of a Canon R5 which is also 4.2.2 .

Hi.

Please click on 'YouTube' in the lower right corner.
Stop the video, and then click on 'SHOW MORE'.

Quote: 'This is a very unscientific real world test based on the way I work in Premier Pro and
Davinci Resolve with Canon C70, Blackmagic raw and ProRes files.'

I expect the M1 max have hardware acceleration of ProRes files, and the RTX 3090 don't.

And he write that he use a Canon C70:

Quote:
'Recording Modes XF-AVC 4:2:2 10-Bit:
4096 x 2160p at 23.98/24/25/29.97/50/60 fps (24 to 410 Mb/s)
4096 x 2160i at 50/59.94 fps (24 to 410 Mb/s)
3840 x 2160p at 23.98/24/25/29.97/50/60 fps (24 to 410 Mb/s)
3840 x 2160i at 50/59.94 fps (24 to 410 Mb/s)
2048 x 1080p at 23.98/24/25/29.97/50/60 fps (24 to 410 Mb/s)
2048 x 1080i at 50/59.94 fps (24 to 410 Mb/s)
1920 x 1080p at 23.98/24/25/29.97/50/60 fps (24 to 410 Mb/s)
1920 x 1080i at 50/59.94 fps (24 to 410 Mb/s)
1280 x 720p at 23.98/24/25/29.97/30/50/59.94 fps (24 to 410 Mb/s)
1280 x 720i at 50/59.94 fps (24 to 410 Mb/s)
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:2 10-Bit:
4096 x 2160p at 119.88/120 fps
3840 x 2160p at 119.88/120 fps
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:2 8-Bit:
2048 x 1080p at 179.82/180 fps
1920 x 1080p at 179.82/180 fps
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:0 8-Bit:
2048 x 1080p at 23.98/24/25/29.97/50/59.94 fps (24 to 35 Mb/s)
1920 x 1080p at 23.98/24/25/29.97/50/59.94 fps (24 to 35 Mb/s)
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:2 10-Bit:
4096 x 2160p at 23.98/24/25/29.97/50/59.94 fps (135 to 225 Mb/s)
3840 x 2160p at 50/59.94 fps (135 to 225 Mb/s)
2840 x 1080p at 23.98/24/25/29.97/50/59.94 fps (50 Mb/s)
1920 x 1080p at 23.98/24/25/29.97/50/59.94 fps (50 Mb/s)
1280 x 720p at 50/59.94 fps (12 Mb/s)
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:0 10-Bit:
4096 x 2160p at 23.98/24/25/29.97/50/59.94 fps (100 to 170 Mb/s)
3840 x 2160p at 23.98/24/25/29.97/50/59.94 fps (100 to 170 Mb/s)
2048 x 1080p at 23.98/24/25/29.97/50/59.94 fps (35 Mb/s)
1920 x 1080p at 23.98/24/25/29.97/50/59.94 fps (35 Mb/s)
1280 x 720p at 50/59.94 fps (9 Mb/s)
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:2 10-Bit:
4096 x 2160p at 119.88/120 fps
3840 x 2160p at 119.88/120 fps
2048 x 1080p at 179.82/180 fps
1920 x 1080p at 179.82/180 fps
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:0 8-Bit:
4096 x 2160p at 23.98/24/25/29.97/50/59.94 fps (150 Mb/s)
3840 x 2160p at 23.98/24/25/29.97/50/59.94 fps (150 Mb/s)
2048 x 1080p at 23.98/24/25/29.97/50/59.94 fps (35 Mb/s)
1920 x 1080p at 23.98/24/25/29.97/50/59.94 fps (35 Mb/s)
1280 x 720p at 50/59.94 fps (8 Mb/s)
H.264/H.265/MP4/XF-AVC/XF-AVC Proxy 4:2:2 8-Bit:
4096 x 2160p at 119.88/120 fps
3840 x 2160p at 119.88/120 fps
2048 x 1080p at 179.82/180 fps
1920 x 1080p at 179.82/180 fps'

From: https://www.bhphotovideo.com/c/product/ ... html/specs

I don't know which color subsampling or bitwith he use. But if he use 4:2:2 10-Bit have the RTX 3090 no hardware acceleration.

Regards Carsten.

It would be nice to see a comparison test of something straightforward. Like h264 files from a GH5 or BRAW and maybe a DJI drone. Things that the NVIDIA has decode and encode for in hardware. Then it would make sense to identify that if you have file type XXX then the M1 is faster. If you have file type YYY the PC is faster ( and a lot cheaper ). For instance with my GH5 files my system is fine with a 1080Ti.

I will wait for tests from know sources.

These tests are pretty much BS. I'm testing temporal NR right now in a 12K timeline with 12K BRAW Q0 on a M1 Pro. It works, but it's not really fast, I'd expect more from a 3090.
Looks like more than 20 minutes for 20 seconds of video, set to a radius of one and "Better". I'll post some numbers later, got other stuff to do now.

I agree that the "exporting" test is silly for DR. It accounts for such a minor share of a typical workflow. And once you start it, it is largely unattended.

I am sure the M1 will kill a lot of previous Intel/AMD/NVidia configs. But the export test is not going to get us wild here. Noise Reduction, transcode of various formats and amount of varying nodes are the bread-and-butter of your typical DR workflow.

I'm calling BS on this one. Either there's something wrong with his settings, or the software is poorly optimized

There's no way a 65 watt laptop should come close to outperforming a 650 watt desktop. Apple has a monopoly on certain things, but the laws of physics isn't one of them.

Blackmagic needs to explain what's going on here.

I wonder if Puget Systems will do a comparison test (because they normally compare PC gear). If not, wait for someone reputable before drawing any conclusions.

Blackmagic needs to explain what's going on here.

The first test is inconclusive. Like Carsten said, if that's C70 footage with 4:2:2 subsampling that PC can't use hardware accelerated decoding.

What's needed is more transparent tests.

I wonder if Puget Systems will do a comparison test (because they normally compare PC gear). If not, wait for someone reputable before drawing any conclusions.

Puget mostly tests stuff they sell. The idea being to help you pick the best Puget system. They don't even test laptops since they don't sell them.

OK, rendering the same process as above on a Mac mini M1 with 16 GB RAM takes 1 hour 50!

This is obviously caused by limited memory, CPU and GPU are not that much weaker. But the machine is swapping like mad. It's still remarkable that it runs stable and doesn't get hot at all. Trying the same NR in such huge resolution on a machine with Intel and a discrete GPU with only 8 GB VRAM results in a message about insufficient memory. Unified memory solves this at the cost of speed, but you'd ruin your SSD pretty fast.

Setting the timeline to UHD results in 4 minutes 14 seconds on the Mac mini.
The MBP M1 Pro does it in 17 seconds and is not challenged at all. I render into ProRes, so the hardware encoder helps.

Puget Systems benchmarks are downloadable:

https://www.pugetsystems.com/labs/artic ... olve-1523/

For that matter, that ancient coffee cup test which has been circulating for years would be more useful than h264/5 encoding....

If there is a new Mac Mini Pro I will be interested in buying compared to upgrading my 1080Ti . If I could even get a GPU now!! The combination of PC and M1 Mini Pro would be good. Not interested in laptop.

But the screen is excellent.

German magazine Slashcam has made a more realistic test with DVR:
https://www.slashcam.de/artikel/Test/Ap ... lles-.html

But the screen is excellent.

It looks excellent but another screen I do not need on my desktop. I have no need for a travelling unit. A Mini would be fine and just fit in with the other equipment.

German magazine Slashcam has made a more realistic test with DVR:
https://www.slashcam.de/artikel/Test/Ap ... lles-.html

Interesting. So they're likening it to a 2080Ti:

"All in all, however, the MacBook M1 Max can be attested to delivering an average performance similar to that of a desktop RTX 2080 Ti under Resolve. And that is quite a number. And even more: With its 11 GB of memory, the RTX2080 Ti repeatedly reports full GPU memory during 8K editing, while there were never such problems on the MacBook Pro M1 Max (with 64 GB of RAM, mind you).

And last but not least, the Resolve timeline on the MacBook M1 Max felt absolutely smooth even in 8K, which could not always be said of the RTX2080 Ti"

The difference in spatial NR is quite significant (1.5 FPS for the 64GB Max vs 7.5 for the 2080Ti), but in other tests it's very close to the 2080Ti.

Pretty impressive.

The difference in spatial NR is quite significant (1.5 FPS for the 64GB Max vs 7.5 for the 2080Ti), but in other tests it's very close to the 2080Ti.

Pretty impressive.

Wondering out loud here, but what do you suppose could account for the huge difference in spatial NR between a Max and some other PC with an RTX 2080Ti?

I guess it's hard to know without understanding how spatial NR is implemented (and perhaps it has not been optimized for the M1 Max at this point)?

Very interesting.

Can some person share a particle project in Fusion, several different effects (scale, transitions, titles), so anyone can import it and test on each setup? Real test, not encode test.

Sent from my CPH2025 using Tapatalk

Make one and put it on a share service.

The difference in spatial NR is quite significant (1.5 FPS for the 64GB Max vs 7.5 for the 2080Ti), but in other tests it's very close to the 2080Ti.

Pretty impressive.

Wondering out loud here, but what do you suppose could account for the huge difference in spatial NR between a Max and some other PC with an RTX 2080Ti?

I guess it's hard to know without understanding how spatial NR is implemented (and perhaps it has not been optimized for the M1 Max at this point)?

It is known that the M1 has the equivalent of an ASIC that does hardware Noise reduction (common ASIC is NVENC/NVDEC in Nvidia GPU's for encoding/decoding certain h.264/h.265). It is possible certain Noise reduction elements can be assisted in hardware while other types can not.

I agree that the "exporting" test is silly for DR. It accounts for such a minor share of a typical workflow. And once you start it, it is largely unattended.

I am sure the M1 will kill a lot of previous Intel/AMD/NVidia configs. But the export test is not going to get us wild here. Noise Reduction, transcode of various formats and amount of varying nodes are the bread-and-butter of your typical DR workflow.

I totally agree - it's almost irrelevant to me. The only exports I do are for producer playouts or studio roll in VTs and the bulk of these are handled by the post department. Most every NLE can handle this now much faster than real time, Avid's UME for example. Rendering speed is what I'd be more interested in, a big part of my time.

I'm very interested in the 64GB M1 Max 16" MacBook Pro but until and if Avid works with it I have to stick to HP - not Apple's fault, of course. But I would like to see more professional real world tests and less fanboy hyperbole. I never understand why consumers take on the job for free of marketing for uber rich conglomerates, though I understand brand ownership is like the protectiveness we feel towards our children - if someone says 'Little Tommy is thick and slow' we leap to their defence because he's our Little Tommy But nonetheless here Intel does not seem to have much of a defence in this seeming gamechanger and if I can see with my own eyes, real world desktop performance in this laptop, PCvsMac, Windows/Mac OS goes out of the window for me, after many years on PC, if you'll excuse the pun!

Wondering out loud here, but what do you suppose could account for the huge difference in spatial NR between a Max and some other PC with an RTX 2080Ti?

I guess it's hard to know without understanding how spatial NR is implemented (and perhaps it has not been optimized for the M1 Max at this point)?

Yeah I'm not sure. This explanation seems reasonable:

It is known that the M1 has the equivalent of an ASIC that does hardware Noise reduction (common ASIC is NVENC/NVDEC in Nvidia GPU's for encoding/decoding certain h.264/h.265). It is possible certain Noise reduction elements can be assisted in hardware while other types can not.

And yes, optimisation could definitely be a factor. The systems are brand new, it's quite possible that BMD will find additional ways to optimise for it in future releases. Like the "3x faster performance!" claimed in 17.3 for M1 systems, based on some M1-specific optimisations. I wouldn't be at all surprised if future updates had further optimisations, perhaps specific to noise reduction and other GPU effects, which closed that large Spatial gap, and maybe made it even faster in the other areas where it's currently matching the 2080Ti.

I wonder how Neat Video will compare. They have an M1 version, but the latest new release was in August so presumably they haven't done any M1 Max/Pro optimisations as yet.

Laptops have thermal throttling. How about long renders with noise reduction and OFX plugins?
At what point does speed go down due to thermal throttling?
Would a MacMini be a better idea because of better cooling?
In any case its super impressive that a tiny macmini box or laptop now has workstation performance.

The 14" blows its fans full power on long renders, but without throttling.
The 16" is hardly getting warm.

But I wouldn't call this workstation performance before having tested all the details.

Just seen an Octane X 3D rendering score for the 64GB Max on OTOY's forums.

Here's the graph of previous results, to which I've added in the Max and NVidia 3080Ti:



So it's definitely impressive for a laptop, and a good step up from the M1 GPU (over twice as fast as the M1 in this test), but it's not workstation-beating as yet.

Of course, if a 65W laptop is at this level, who knows what a full size iMac Pro or Mac Pro is going to be able to do..

Here's the graph of previous results, to which I've added in the Max and NVidia 3080Ti:

I don't feel good about paying M1 Max 64GB prices to get the performance of a desktop 5700XT.

A few YouTubers have promised to test Fusion performance on the M1 Pro/Max, so I'll wait for their results.

Meanwhile, for more complex Fusion, I'm coping with the low GPU performance of the M1 by using Saver/Loader nodes, so I might wait to see what the next year's M2 generation offers.

I have MacBook Pro m1 14’’ base model and it’s twice as fast in export times than my desktop with intel i9 11900 , 32gb ram, nvidia rtx 3060 12gb vram.

It’s also quite faster in fusion , about 1.5 times.

Both have nvme ssd 1tb.

I am quite impressed from this little monster!

Timeline was 4k , with 8k and 4k clips with mixed codecs like canon raw, Sony mp4 10bit and 8 bit, Fuji 4k 60p, braw etc







Sent from my iPad using Tapatalk Pro

Some people really want these Macs to perform poorly, it seems.

Not at all. But sometimes these early tests are performed by people who haven’t thought it through and they’re not testing what they think they’re testing. In the end it becomes misleading.

They’re fantastic laptops that are good at many things, just not best at everything as they’re sometimes portrayed.

For me it's fake test. We don't have any info about render settings. As a default Resolve sets NATIVE CPU rendering. We don't have any info about that he switched renderer to NVENC.
Second - he almost don't use any CUDA processing. What we see here is decoder/encoder benchmark. Very lame. RTX 3090 would hit M1 GPU with 3D processing. 4x higher performance.
We need objective benchmark, project. We all could benchmark ONE and the same project on our machines and share results. Then it makes sense.

Man on the video seems to be technical lamer, unfortunately.

Just analyze that two benchmarks from Premiere. There is no way to say that M1 is 2x faster than RTX

PC RTX3090 https://www.pugetsystems.com/benchmarks ... p?id=63335
M1 max https://www.pugetsystems.com/benchmarks ... p?id=63388

In test from video for sure is something wrong.

RTX3090 is twice as fast as M1 graphics... Which is also very troubling for the Nvidia - their flagship GPU being just 2x faster than a small SoC.
I am being a little bit evil.

Sent from my Mi 9T using Tapatalk

And you can buy the whole laptop for the current price of an available RTX3090 card.

I'm glad people still get themselves into situations where macbooks outperform high end windows pc's, somehow. It keeps the market healthy and keeps the boys separated from men who do know how to handle a windows/linux pc.

I just got myself a MacBook Pro with the M1 Pro 10 core. I'm a bit disappointed to say that film grain seems to really kill performance. Otherwise it's mostly as fast as my desktop with 2080 Ti which draws power like it can't wait to kill the planet through climate change.