Apple's M1 Max Benchmarked in Adobe Premiere Pro

[Lightwreather]

I did this because i was bored. Just saying.

Summary

The very first benchmark results of Apple's M1 Max system-on-chip in a professional application have arrived.

 

Quotes

Quote

On Thursday an M1 Max-based MacBook Pro user posted performance numbers obtained in Adobe Premiere Pro to the PugetBench database. Benchmark results were obtained using PugetBench for Premiere Pro 0.95.1 and Premiere Pro 15.4.1, so it is possible to compare them to results obtained on other high-performance systems and find out how the new MBP stacks up against machines based on x86 CPUs and discrete GPUs.

Overall

Apple's MacBook Pro 16 based on the M1 Max SoC achieved a 1168 standard overall score and 1000 extended overall score in PugetBench for Premiere Pro 0.95.1. Significantly higher when compared to high-end laptops. Furthermore, Apple's scores are close to those of advanced desktops. 

Live Playback

Apple's M1 Max clearly has an excellent media playback engine that outperforms not only standalone mobile GPUs, but even Nvidia's top-of-the-range GeForce RTX 3090.

Export

The new MacBook Pro 16 clearly outperforms its predecessor in Export workloads, but is somewhat behind other high-end laptops and is drastically behind advanced desktops.  

It is necessary to point out that PugetBench for Premiere Pro's Export workloads include heavy CPU effects and heavy GPU effects sequences, so overall score is somewhat mixed. It still makes a lot of sense because real-world workflows may require different types of effects.  

GPU Score

The new M1 Max SoC can also compete very well against standalone mobile GPUs, namely the GeForce RTX 3060 and RTX 3080 (which seems strangely slow in this benchmark), in Premier Pro while consuming much less power.

 

My thoughts

So, this pretty much confirms what we know about the new M1 Max chips...., They're "Fast" (Apple used that word multiple times during the presentation, look it up). But yea, It's pretty good for a mobile device, able to come close with some "advanced desktops" even. Now, the source article for this post, is a little clickbaity, it compares the GPU performace to Desktop's as well, which personally I think is slightly unfair, but eh. However, one should ALWAYS take these pre-release benchmarks with a grain of salt (or maybe be even a tablespoon's worth or maybe a sea's worth, who knows) and wait for independant reviews to come out. Altho, I for one am really excited ( I feel very tempted to get a 14" (Must resist))

Sources

Tom's hardware

PugetBench

[SignatureSigner]

Seems like apple it breathing new life into the macs. Magsafe, good preformance, decent keyboard, great trackpad, whats next up-gradable ram?

ha ha lol never -apple

[leadeater]

The difference between Live Playback and Export is a little disappointing, wonder if that has to do with long term power limits or total package power limiting the combined CPU and GPU performance in a full workload utilization of both.

 

M1 Max

Quote

4K H.264 150Mbps 8bit	MultiCam - Live Playback (Full Res)	59.64	FPS
4K H.264 150Mbps 8bit	Standard - Live Playback (Full Res)	59.94	FPS
4K H.264 150Mbps 8bit	2x Forward - Live Playback (Full Res)	119.88	FPS
4K H.264 150Mbps 8bit	4x Forward - Live Playback (Full Res)	233.22	FPS
4K H.264 150Mbps 8bit	Standard - Export (H.264 40Mbps UHD)	90.40	FPS
4K H.264 150Mbps 8bit	Standard - Export (ProRes 422HQ UHD)	44.10	FPS
4K H.264 150Mbps 8bit	Live Playback Score	503.50	Score
4K H.264 150Mbps 8bit	Export Score	68.60	Score
4K ProRes 422	MultiCam - Live Playback (Full Res)	53.70	FPS
4K ProRes 422	Standard - Live Playback (Full Res)	59.94	FPS
4K ProRes 422	2x Forward - Live Playback (Full Res)	119.88	FPS
4K ProRes 422	4x Forward - Live Playback (Full Res)	216.18	FPS
4K ProRes 422	Standard - Export (H.264 40Mbps UHD)	90.92	FPS
4K ProRes 422	Standard - Export (ProRes 422HQ UHD)	49.73	FPS
4K ProRes 422	Live Playback Score	119.50	Score
4K ProRes 422	Export Score	69.60	Score
4K RED	MultiCam - Live Playback (Full Res)	3.90	FPS
4K RED	Standard - Live Playback (Full Res)	39.26	FPS
4K RED	2x Forward - Live Playback (Full Res)	14.34	FPS
4K RED	4x Forward - Live Playback (Full Res)	15.88	FPS
4K RED	Standard - Export (H.264 40Mbps UHD)	37.74	FPS
4K RED	Standard - Export (ProRes 422HQ UHD)	29.31	FPS
4K RED	Live Playback Score	36.00	Score
4K RED	Export Score	47.70	Score
4K Heavy GPU Effects	Advanced - Effects (ProRes 422HQ UHD)	33.84	FPS
4K Heavy GPU Effects	Extreme - Effects (ProRes 422HQ UHD)	6.51	FPS
4K Heavy GPU Effects	Effects Score	41.90	Score
4K Heavy CPU Effects	Advanced - Effects (ProRes 422HQ UHD)	70.66	FPS
4K Heavy CPU Effects	Extreme - Effects (ProRes 422HQ UHD)	14.33	FPS
4K Heavy CPU Effects	Effects Score	95.30	Score
8K H.265 100Mbps	Standard - Live Playback (Half Res)	59.94	FPS
8K H.265 100Mbps	2x Forward - Live Playback (Half Res)	8.94	FPS
8K H.265 100Mbps	4x Forward - Live Playback (Half Res)	5.54	FPS
8K H.265 100Mbps	Standard - Export (H.264 40Mbps UHD)	57.61	FPS
8K H.265 100Mbps	Standard - Export (ProRes 422HQ UHD)	38.38	FPS
8K H.265 100Mbps	Live Playback Score	67.60	Score
8K H.265 100Mbps	Export Score	113.80	Score
8K RED	Standard - Live Playback (Half Res)	19.88	FPS
8K RED	2x Forward - Live Playback (Half Res)	21.28	FPS
8K RED	4x Forward - Live Playback (Half Res)	17.78	FPS
8K RED	Standard - Export (H.264 40Mbps UHD)	19.84	FPS
8K RED	Standard - Export (ProRes 422HQ UHD)	18.80	FPS
8K RED	Live Playback Score	78.50	Score
8K RED	Export Score	52.10	Score

 

5900X + RTX 3090

Quote

4K H.264 150Mbps 8bit	MultiCam - Live Playback (Full Res)	20.38	FPS
4K H.264 150Mbps 8bit	Standard - Live Playback (Full Res)	59.94	FPS
4K H.264 150Mbps 8bit	2x Forward - Live Playback (Full Res)	119.48	FPS
4K H.264 150Mbps 8bit	4x Forward - Live Playback (Full Res)	59.19	FPS
4K H.264 150Mbps 8bit	Standard - Export (H.264 40Mbps UHD)	110.64	FPS
4K H.264 150Mbps 8bit	Standard - Export (ProRes 422HQ UHD)	75.36	FPS
4K H.264 150Mbps 8bit	Live Playback Score	173.40	Score
4K H.264 150Mbps 8bit	Export Score	95.80	Score
4K ProRes 422	MultiCam - Live Playback (Full Res)	41.96	FPS
4K ProRes 422	Standard - Live Playback (Full Res)	59.94	FPS
4K ProRes 422	2x Forward - Live Playback (Full Res)	109.89	FPS
4K ProRes 422	4x Forward - Live Playback (Full Res)	151.00	FPS
4K ProRes 422	Standard - Export (H.264 40Mbps UHD)	133.51	FPS
4K ProRes 422	Standard - Export (ProRes 422HQ UHD)	70.36	FPS
4K ProRes 422	Live Playback Score	99.60	Score
4K ProRes 422	Export Score	100.30	Score
4K RED	MultiCam - Live Playback (Full Res)	24.33	FPS
4K RED	Standard - Live Playback (Full Res)	59.94	FPS
4K RED	2x Forward - Live Playback (Full Res)	88.91	FPS
4K RED	4x Forward - Live Playback (Full Res)	47.85	FPS
4K RED	Standard - Export (H.264 40Mbps UHD)	87.22	FPS
4K RED	Standard - Export (ProRes 422HQ UHD)	52.77	FPS
4K RED	Live Playback Score	109.50	Score
4K RED	Export Score	96.60	Score
4K Heavy GPU Effects	Advanced - Effects (ProRes 422HQ UHD)	77.00	FPS
4K Heavy GPU Effects	Extreme - Effects (ProRes 422HQ UHD)	36.83	FPS
4K Heavy GPU Effects	Effects Score	144.40	Score
4K Heavy CPU Effects	Advanced - Effects (ProRes 422HQ UHD)	63.05	FPS
4K Heavy CPU Effects	Extreme - Effects (ProRes 422HQ UHD)	17.35	FPS
4K Heavy CPU Effects	Effects Score	97.80	Score
8K H.265 100Mbps	Standard - Live Playback (Half Res)	59.94	FPS
8K H.265 100Mbps	2x Forward - Live Playback (Half Res)	16.48	FPS
8K H.265 100Mbps	4x Forward - Live Playback (Half Res)	17.68	FPS
8K H.265 100Mbps	Standard - Export (H.264 40Mbps UHD)	43.42	FPS
8K H.265 100Mbps	Standard - Export (ProRes 422HQ UHD)	32.55	FPS
8K H.265 100Mbps	Live Playback Score	122.60	Score
8K H.265 100Mbps	Export Score	90.20	Score
8K RED	Standard - Live Playback (Half Res)	45.55	FPS
8K RED	2x Forward - Live Playback (Half Res)	43.11	FPS
8K RED	4x Forward - Live Playback (Half Res)	19.33	FPS
8K RED	Standard - Export (H.264 40Mbps UHD)	42.34	FPS
8K RED	Standard - Export (ProRes 422HQ UHD)	30.53	FPS
8K RED	Live Playback Score	117.40	Score
8K RED	Export Score	96.10	Score

 

5900X + RTX 3080

Quote

4K H.264 150Mbps 8bit	MultiCam - Live Playback (Full Res)	16.93	FPS
4K H.264 150Mbps 8bit	Standard - Live Playback (Full Res)	59.94	FPS
4K H.264 150Mbps 8bit	2x Forward - Live Playback (Full Res)	119.48	FPS
4K H.264 150Mbps 8bit	4x Forward - Live Playback (Full Res)	52.15	FPS
4K H.264 150Mbps 8bit	Standard - Export (H.264 40Mbps UHD)	110.90	FPS
4K H.264 150Mbps 8bit	Standard - Export (ProRes 422HQ UHD)	74.40	FPS
4K H.264 150Mbps 8bit	Live Playback Score	157.90	Score
4K H.264 150Mbps 8bit	Export Score	95.40	Score
4K ProRes 422	MultiCam - Live Playback (Full Res)	42.26	FPS
4K ProRes 422	Standard - Live Playback (Full Res)	59.94	FPS
4K ProRes 422	2x Forward - Live Playback (Full Res)	109.59	FPS
4K ProRes 422	4x Forward - Live Playback (Full Res)	140.96	FPS
4K ProRes 422	Standard - Export (H.264 40Mbps UHD)	125.80	FPS
4K ProRes 422	Standard - Export (ProRes 422HQ UHD)	68.67	FPS
4K ProRes 422	Live Playback Score	98.00	Score
4K ProRes 422	Export Score	96.20	Score
4K RED	MultiCam - Live Playback (Full Res)	24.03	FPS
4K RED	Standard - Live Playback (Full Res)	59.94	FPS
4K RED	2x Forward - Live Playback (Full Res)	88.01	FPS
4K RED	4x Forward - Live Playback (Full Res)	49.00	FPS
4K RED	Standard - Export (H.264 40Mbps UHD)	86.99	FPS
4K RED	Standard - Export (ProRes 422HQ UHD)	51.98	FPS
4K RED	Live Playback Score	109.80	Score
4K RED	Export Score	95.80	Score
4K Heavy GPU Effects	Advanced - Effects (ProRes 422HQ UHD)	61.65	FPS
4K Heavy GPU Effects	Extreme - Effects (ProRes 422HQ UHD)	22.08	FPS
4K Heavy GPU Effects	Effects Score	99.10	Score
4K Heavy CPU Effects	Advanced - Effects (ProRes 422HQ UHD)	62.33	FPS
4K Heavy CPU Effects	Extreme - Effects (ProRes 422HQ UHD)	17.55	FPS
4K Heavy CPU Effects	Effects Score	97.80	Score
8K H.265 100Mbps	Standard - Live Playback (Half Res)	59.94	FPS
8K H.265 100Mbps	2x Forward - Live Playback (Half Res)	16.38	FPS
8K H.265 100Mbps	4x Forward - Live Playback (Half Res)	8.39	FPS
8K H.265 100Mbps	Standard - Export (H.264 40Mbps UHD)	43.92	FPS
8K H.265 100Mbps	Standard - Export (ProRes 422HQ UHD)	34.23	FPS
8K H.265 100Mbps	Live Playback Score	90.00	Score
8K H.265 100Mbps	Export Score	92.80	Score
8K RED	Standard - Live Playback (Half Res)	45.30	FPS
8K RED	2x Forward - Live Playback (Half Res)	43.71	FPS
8K RED	4x Forward - Live Playback (Half Res)	21.68	FPS
8K RED	Standard - Export (H.264 40Mbps UHD)	42.53	FPS
8K RED	Standard - Export (ProRes 422HQ UHD)	29.93	FPS
8K RED	Live Playback Score	124.10	Score
8K RED	Export Score	95.30	Score

 

Bolded largest differences between the M1 Max and 5900X + RTX 3090 for those that want to see exactly where the differences are present.

 

Edit:

TL;DR With a M1 Pro/M1 Max + Premiere do not use RED, you'll regret it A LOT.

[Kronoton]

23 minutes ago, SignatureSigner said:

whats next up-gradable ram?

While you were joking, feel free to take your time to calculate how many SO-DIMMs (channels) you'd need to match the transfer speeds Apple gave for these.

[Kisai]

24 minutes ago, J-from-Nucleon said:

Now, the source article for this post, is a little clickbaity, it compares the GPU performace to Desktop's as well, which personally I think is slightly unfair, but eh.

The problem is that this benchmark is kinda vague. Like note how the Geforce 3090 is behind the 3080 Ti. If anything the benchmark favored the system with larger amounts of memory and barely weighed in on the GPU because the bottleneck was very likely the system memory.

 

As I mentioned before in another thread, the memory bandwidth of the 3060 is slightly under the M1 Max. So those scores being similar to the 3060 seem pretty much where it should be.

[leadeater]

4 minutes ago, Kronoton said:

While you were joking, feel free to take your time to calculate how many SO-DIMMs (channels) you'd need to match the transfer speeds Apple gave for these.

4 Channel DDR5-6400 for the M1 Pro and 8 Channel DDR5-6400 for M1 Max, neither of which is unreasonable for a Mac Pro, much the case in a laptop lol.

 

It's more for LPDDR5 because each channel is half the width compared to DDR5 so each memory module has 4 16bit channels. Total bit width between DDR5 SODIMM and LPDDR5 would be the same, it's just that while you could get 4 SODIMMs in a laptop it's not the nicest thing to do. You could even do 8 but again physical space and having to use both sides of the PCB would bring in some additional mounting challenges to the chassis.

 

Very much a case of a capability not well suited to a portable device, I see 4 SODIMMs as the most practical possible but not desirable.

 

Replaceable LPDDR5 modules would be awesome though, I'm sure that would actually be possible if enough effort were put in to making it a possibility.

[leadeater]

15 minutes ago, Kisai said:

The problem is that this benchmark is kinda vague. Like note how the Geforce 3090 is behind the 3080 Ti. If anything the benchmark favored the system with larger amounts of memory and barely weighed in on the GPU because the bottleneck was very likely the system memory.

 

As I mentioned before in another thread, the memory bandwidth of the 3060 is slightly under the M1 Max. So those scores being similar to the 3060 seem pretty much where it should be.

You can use my posted results if you prefer, they are from systems with 128GB system memory and same memory speed, and same Windows build as well as Nvidia driver version.

[Kronoton]

9 minutes ago, leadeater said:

4 Channel DDR5-6400 for the M1 Pro and 8 Channel DDR5-6400 for M1 Max, neither of which is unreasonable for a Mac Pro, much the case in a laptop lol.

 

 

Well this is a laptop, and I wouldn't be surprised if the M1Über in the 2022 MacPro would double that once again...

 

Routing these signals in a laptop would be a pain, longer lines with more connections would require higher latency and voltage impacting both real life performance and battery live.

 

All for something most Mac user never did even back when it was a question of 1 small Phillips screwdriver and 5 minutes of time.

[leadeater]

6 minutes ago, Kronoton said:

Routing these signals in a laptop would be a pain, longer lines with more connections would require higher latency and voltage impacting both real life performance and battery live.

The latency benefit is way over stated, power sure but that actually has more to do with how SODIMMs are made and the number of memory packages on them not the distance and number of traces.

 

Just remember to sanity check things like this, to have a significant power increase due to traces and distance this would be power loss over the traces and turned in to heat, if you know anything about copper then you'll know it's resistance is very, very low so power loss or dissipation in to heat would also be very low.

 

LPDDR5, and LPDDR in general, main benefit is the layering you can do to create a module of the capacity and bandwidth you desire while doing so in a way that does use less power and a huge amount of less space. The limiting factor however is the total capacity of a module and implementation is lower than that of SODIMM. Not that a Mac Pro would use SODIMM however.

[Kisai]

4 minutes ago, leadeater said:

You can use my posted results if you prefer, they are from systems with 128GB system memory and same memory speed, and same Windows build as well as Nvidia driver version.

If I were to guess, I'm going to say that Premire Pro has not been optimized for the M1 or M1's hardware decoder/encoders, as I know off hand the Geforce parts have "locked" encoders, where the quadro's do not. GP104/GP102 (Pascal) had 2 or 3 NVEnc's on the high end parts, where as TU (Turing) and GA (Ampere) only have 1. On PLEX benchmarks the x80 parts (Pascal and Turing) can only encode 6 4k streams at a time, or 19-24 1080p streams. Keep in mind that each encoder requires explicit amounts of video memory, transcoding 4K requires 1.3GB of VRAM where 1080p to 1080 is 320MB.

 

At any rate, I'd like to see a benchmark with DaVinci Resolve Studio before I'd put too much stock in these scores. DaVinci does support M1 as of 17.1, and as of 17.3 actually has been better optimized for it.

[AlTech]

39 minutes ago, leadeater said:

The difference between Live Playback and Export is a little disappointing, wonder if that has to do with long term power limits or total package power limiting the combined CPU and GPU performance in a full workload utilization of both.

I think it was Luke Miani or someone else on YouTube in a video I watched who said that the M1 Pro and Max would struggle with sustained long term performance because of throttling.

 

Despite the efficiency of Apple Silicon, if Apple continues to use bad cooling then it's still possible their chips will throttle during sustained workloads.

[Kronoton]

5 minutes ago, leadeater said:

if you know anything about copper then you'll know it's resistance is very,

 

At these frequencies copper resistance is a secondary concern at best. Signal integrity is the name of game, the PCB acting like as a capacity, other stuff radioing and so.

Also while the copper trace might be perfect that SODIMM socket will never be. Hence you need to run the signal at higher voltage compare to 100% defined connection on the SoC.  

[leadeater]

5 minutes ago, Kisai said:

If I were to guess, I'm going to say that Premire Pro has not been optimized for the M1 or M1's hardware decoder/encoders, as I know off hand the Geforce parts have "locked" encoders, where the quadro's do not. GP104/GP102 (Pascal) had 2 or 3 NVEnc's on the high end parts, where as TU (Turing) and GA (Ampere) only have 1. On PLEX benchmarks the x80 parts (Pascal and Turing) can only encode 6 4k streams at a time, or 19-24 1080p streams. Keep in mind that each encoder requires explicit amounts of video memory, transcoding 4K requires 1.3GB of VRAM where 1080p to 1080 is 320MB.

 

At any rate, I'd like to see a benchmark with DaVinci Resolve Studio before I'd put too much stock in these scores. DaVinci does support M1 as of 17.1, and as of 17.3 actually has been better optimized for it.

There are sub tests where the M1 Max is way, way faster than the RTX 3090. Looking at the results I get the impression that it's a limitation of the total bitrate required, as well as the encoding used and decoder capabilities.

 

Apples architecture seems very weak in RED and H.265 but very strong in H.264 and ProRes, which does make sense.

 

So I think it's less about optimization and more about workflow and knowing what to use with your hardware to get the best out of it i.e. not using RED with Apple.

[leadeater]

5 minutes ago, Kronoton said:

At these frequencies copper resistance is a secondary concern at best. Signal integrity is the name of game, the PCB acting like as a capacity, other stuff radioing and so.

And if that really were a problem that actually impacted latency then it would be an observable issue on say an EPYC server with 24 DIMMs per socket. 

 

8 minutes ago, Kronoton said:

Hence you need to run the signal at higher voltage compare to 100% defined connection on the SoC.  

LPDDR5 0.5V/1.05V/1.8V, DDR5 1.1V. Not seeing a great difference here. What's being used here is most likely 1.05V or it could even be 1.8V, I don't think so but that is in the LPDDR5 spec and Samsung's packages are much larger (layers wise) than typical LPDDR5, at least the ones being used in the Apple laptops. Quite unlikely to be 0.5V.

 

Similar situation with DDR4 generation as well. What you're saying was much more the case in DDR3 era.

[Kronoton]

Even if it was just 1.05 vs 1.1V, that is quite a lot when you are trying to build a thin laptop with 20h battery.

AFAIK noone but Apple uses these chips atm so there really is no way knowing what the are doing and what special sauce they may have been going on with these.

 

 

No I don't know how bad it all is but even a 5% reduction in power or 5% boost in performance does count when you are trying to build something that needs to have a certain *wow* factor over the competition.

 

Apple put all they're eggs in 1 basket with the ARM/AS switch and being "just a bit faster than Intel" would have been seen as failure.

[igormp]

1 hour ago, leadeater said:

The difference between Live Playback and Export is a little disappointing, wonder if that has to do with long term power limits or total package power limiting the combined CPU and GPU performance in a full workload utilization of both.

I believe that the playbacks are making use of the dedicated decoders found on the GPUs of each device, and Apple's decoder should be way better, but that implies that it's not testing the actual GPU/CPU cores, thus the export results would make sense since those would actually be using the CPU for the job (I guess?)

[leadeater]

31 minutes ago, Kronoton said:

Even if it was just 1.05 vs 1.1V, that is quite a lot when you are trying to build a thin laptop with 20h battery.

Total worst case if the ram was always at 100% power state would be 19.1 hours instead of 20 hours, that's also assuming the 20 hours was also with the ram at 100% and neither of these would be true.

 

While the difference is going to be quite small if you only purely look at the voltage DIMMs themselves put more load on the IMC and use more current. It wouldn't affect performance however, you can control allowed power to sections of the SoC and cores themselves and increasing peak allowed power from say 35W to 36W (or less tbh) isn't really a big deal. I wouldn't bother brining in performance in to this, that's far to tricky of a thing to try and generalize.

 

So like I agree there is a power difference and it's going to be measurable but it's also not very pronounced and latency, that I do not agree with that all. Like I said originally, that's way overstated benefit.

 

Total physical area and bandwidth per module are the two primary reasons Apple went with LPDDR4X and LPDDR5, soldered memory was never a problem for them on the Intel based products and since they were limited in memory bit width going with DDR4 instead meant you could offer both higher capacities and also at cheaper price.

 

31 minutes ago, Kronoton said:

AFAIK noone but Apple uses these chips atm so there really is no way knowing what the are doing and what special sauce they may have been going on with these.

They are literally just the next generation of the older LPDDR4X modules used in the M1. There's no special sauce going on, other than Apple having exclusive access to supply, or at least near exclusive. I don't know of anyone else using them, or even wanting them since they are hell of expensive.

[igormp]

23 minutes ago, Kronoton said:

AFAIK noone but Apple uses these chips atm so there really is no way knowing what the are doing and what special sauce they may have been going on with these.

 

1 minute ago, leadeater said:

They are literally just the next generation of the older LPDDR4X modules used in the M1. There's no special sauce going on, other than Apple having exclusive access to supply, or at least near exclusive. I don't know of anyone else using them, or even wanting them since they are hell of expensive.

 

I wondered the same thing in another thread and found the answer, I'll just quote myself here:

2 hours ago, igormp said:

I don't see how that has anything to do with my comment, but they're apparently using regular LPDDR5:

 While it's impressive that they managed to get it and be the first with an lpddr5 product (feel free to correct me here, I'm not aware of any other product using those), all of the other companies should be able to do the same if they got their heads out of their asses, because using a shiny toy shouldn't be anything to give praise to a company for (instead, the one that made the toy should get the kudos), and now we're back at my initial point that Apple is getting ahead by simply buying the latest and greatest while other companies lag behind, for god-knows-why reason.

 

Nvm about being the first, other devices have it since last year. Apparently we don't see it in other laptops because the memory controller found in current CPUs don't support it, shame on them.

 

[leadeater]

5 minutes ago, igormp said:

I believe that the playbacks are making use of the dedicated decoders found on the GPUs of each device, and Apple's decoder should be way better, but that implies that it's not testing the actual GPU/CPU cores, thus the export results would make sense since those would actually be using the CPU for the job (I guess?)

Check out the more detail results I posted. Evidence does look like hardware acceleration differences just based on the differences in encoding format used. However to me it appears they are still being able to be used when exporting and it feels like power limits to me.

[leadeater]

1 minute ago, igormp said:

I wondered the same thing in another thread and found the answer, I'll just quote myself here:

Yea I've already seen it, had mathed out what each module was before that anyway. It's pretty easy to figure out just by looking at Samsung's LPDDR5 spec. Original M1 and it's LPDDR4X were also just an high end as these LPDDR5 modules are.

[igormp]

Just now, leadeater said:

However to me it appears they are still being able to be used when exporting and it feels like power limits to me.

I don't think so, otherwise I believe that the ProRes results should be way higher since they spoke a lot about it.

[igormp]

1 minute ago, leadeater said:

Yea I've already seen it, had mathed out what each module was before that anyway. It's pretty easy to figure out just by looking at Samsung's LPDDR5 spec. Original M1 and it's LPDDR4X were also just an high end as these LPDDR5 modules are.

Oh, I was just referring to the exclusivity part, there are many phones using LPDDR5 already, so there's no magic going on for apple other than having a memory controller that supports it, unlike other x86 chips.

[leadeater]

1 minute ago, igormp said:

I don't think so, otherwise I believe that the ProRes results should be way higher since they spoke a lot about it.

How so? The ProRes performance is already higher than the RTX 3090, just not in export which is why I think it's power limited when exporting.

[igormp]

3 minutes ago, leadeater said:

How so? The ProRes performance is already higher than the RTX 3090, just not in export which is why I think it's power limited when exporting.

Yeah, I was referring to the export performance, sorry for not making it clear.

Didn't they mention also having a ProRes encoder? If so, it shouldn't be power limited since it's an asic dedicated to such media stuff.

Another thing might be that there's no support for the dedicated decoder yet, so it's being done through software.

 

Also, I believe they're not using NVENC for those comparisons (it'd make no sense IMO due to quality and size), so the exports are done by the CPU in both systems, the GPU has almost no use in there (unless you're using effects, which the 5900x+3090 did manage to get ahead), so the comparison is actually between the M1's CPU and the 5900X.

The decode part does make sense to be done through the decoders when available (not the case for the 3090 with ProRes since NVDEC has no support for it).

[leadeater]

2 minutes ago, igormp said:

Didn't they mention also having a ProRes encoder? If so, it shouldn't be power limited since it's an asic dedicated to such media stuff.

They did but I'd have to re-watch to catch exactly what they said to know what it means, both in the overview of the SoC and when the woman was talking about ProRes streams vs 28 core Xeon with Afterburner card.

 

What's annoying about this is we only know about it from benchmark database search, what I really want to see is Final Cut Pro tests because then I know it's 100% using the M1 Max properly.