M3 Max Chip Around as Fast as M2 Ultra in Early Benchmark Results

[danwells]

That should be fine for non-gamers! thanks!

 

[ValO]

That’s like, the least important spec of all.

I reacted to another post, that poster made a big deal of this spec and was inaccurate.

 

[Frantisekj]

Metal scores are now coming up for M3 Max

m3 max - Geekbench 6 GPU Search - Geekbench

browser.geekbench.com

Again just like the OpenCL scores they are the same as the M1 Ultra.

Here is link to comparison

Mac Studio vs MacBook Pro M3 Max - Geekbench

browser.geekbench.com

Some test have 30% difference both +/- so generally it is tie.

 

[Charles50]

Here is link to comparison

Mac Studio vs MacBook Pro M3 Max - Geekbench

browser.geekbench.com

Some test have 30% difference both +/- so generally it is tie.

Well that's an invalid comparison. It puts an unspecified model of the M1 Mac Studio vs. the M3 Max Macbook Pro. Your Metal M3 bench is 155360. My M3 Ultra has a Metal bench of 220558. And this isn't even measuring the SSD speed, the M1 and M2 Ultra both have 800Gb/sec memory bandwidth and the M3 Max only has 150Gb/sec. I'll post a screencap of the Metal scores for both

So NO, the M3 is not even close to the speed of any Mac Studio models.

 

[PowerMike G5]

Nope, not so fast, this is not necessarily the case -- at least for editing. Unless I'm missing something these scores don't account for the full story of how Apple Silicon differs from the old Intel machines. Remember that these chips are modular and contain different subprocessing units for specialized tasks such as media playback. The M2 Ultra chips still have more media processing engines than the M3 Max. This M3 Max may be faster in terms of raw GPU and CPU abilities but what makes Apple Silicon so powerful for editing workflows is the inclusion of those media engines (that's why people can play multiple 8k streams on Apple Silicon when editing).

Maybe the raw performance of the M3 Max's GPU makes up for the lack of extra media engines but it's interesting how these specialized subprocessors (like the NPU or media engines) are becoming more important for workflows than raw horsepower. We will have to wait and see the tests but I can see a world where the extra media engines in a last generation Ultra chip are more valuable for certain workflows than a current gen Max chip.

Yeah the totality of the SOC and how it affects what you plan to do with the computer should drive a lot of your decision making. I noticed this specialization already with even just using the Afterburner in my 7,1 Mac Pro. Combined with the Apple T2 which has hardware encode/decode of H.264 and H.265, it has similar, if an earlier iteration, of the media engine in Apple Silicon.

Editing multi-stream 4K ProRes, my CPU and GPU are almost sitting idle because the hardware acceleration card is doing all the decode. The CPU and GPU become mostly used for real-time effects, speeding encodes, etc.

 

[danwells]

Well that's an invalid comparison. It puts an unspecified model of the M1 Mac Studio vs. the M3 Max Macbook Pro. Your Metal M3 bench is 155360. My M3 Ultra has a Metal bench of 220558. And this isn't even measuring the SSD speed, the M1 and M2 Ultra both have 800Gb/sec memory bandwidth and the M3 Max only has 150Gb/sec. I'll post a screencap of the Metal scores for both

So NO, the M3 is not even close to the speed of any Mac Studio models.

View attachment 2318628

Fully enabled M3 Max has a 400 GB/s memory bandwidth (no, it's not 800, but it's not 150 either - that's the M3 Pro). Nobody's claiming that the GPU speed is as fast as the 76-core M2 Ultra. The claim is that the CPU speed is very similar, and the GPU speed is within about 25%. Losing 25% in GPU and little else for a machine that runs on batteries for many hours at a time is a pretty good trade if you ever work on the road. It also makes a lot of us wonder what the M3 Ultra will be like!

 

[Melbourne Park]

A few areas I disagree with:

1. The whole power consumption is overrated. This depends on the region and needs. In our state, power costs 9-10 cents a KwH. And most companies/labs where these workstations will sit, have lights running near non-stop and other areas where power consumption can be optimized.

2. Noctua / large fans and liquid cooling were becoming mainstream last decade. You cannot buy a 2.5-3k Windows gaming PC without liquid cooling and the large fans make many high end workstations virtually quiet. That and even with the challenges, developers will want their CUDA and OpenCL platforms.
...

Fact is that when people here talk about power consumption, it's most relevant to notebooks, because extra battery capacity is both heavy and costly.

But when wattage is analysed, for places like Asia, Australia, California, Mexico etc., the extra energy cost of cooling should also be included.

Personally, my home office is currently used by my son, it's small, 2 metres x 8 metres ie 6 feet 7 x 23 feet. When his computer is on, the room is hot. He turns the Air-Con on. When his desktop is off, the room is comfortable - the air con is off.

There's huge impact on energy from bad GPU and CPU tech. My son does laser imaging, it all has to be PC based, and most jobs now get rendered in India - labour costs are less there, but also, they use a lot of low cost electricity, due to coal power stations being built. Same with China. Power all adds up if one believes in CO2 contributing to global warming. It's like comparing a '60's Cadillac V8 to today's limited cylinder turbos. Heh Man - look at those cooling fans in that box. Check out the heat sinks and radiators, fab eh? Did you count all those fans on those huge GPUs - what beasts! Those heat sinks on the power supplies are also awesome ...