MP All Models M2 Ultra Geekbench Scores

[goMac]

Here's an interesting one in DaVinci Resolve, the M2 Ultra actually beats the RTX 4090 in render tests:

My guess - judging from the test description - is this test is leaning heavily on video decode. Apple Silicon does have faster decode/encode performance than a 4090.

Encode/decode tests are almost always favorable to Apple Silicon across the board. Which is a specific use case Apple has very legitimately optimized for.

Nvidia has also opted not to compete there in favor of generalized GPU performance. So it's a trade off.

 

[MisterAndrew]

It looks like these are the first Geekbench results for the M2 Ultra Mac Pro (the others are Studio).

2794 single core, 21453 multi-core

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

196543 Metal

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

113870 OpenCL

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

 

[Dopemaster]

My guess - judging from the test description - is this test is leaning heavily on video decode. Apple Silicon does have faster decode/encode performance than a 4090.

Encode/decode tests are almost always favorable to Apple Silicon across the board. Which is a specific use case Apple has very legitimately optimized for.

Nvidia has also opted not to compete there in favor of generalized GPU performance. So it's a trade off.

Yep it would undoubtedly be utilising the AS media engine in that scenario. There are other still photography related benchmarks that the Ultra is also coming out ahead of the 4090, again, likely related to the media engine.

For my personal use case, feature documentary production, it’s pretty exciting to see a tiny, travel ready machine outclassing the 4090 in such a crucial and frequent task in my workflow.

Lucky for me I guess. If I were a 3D artist I’d certainly be looking elsewhere…

 

[Dopemaster]

It looks like these are the first Geekbench results for the M2 Ultra Mac Pro (the others are Studio). Side note: You can see that the Mac Pro doesn't actually exist anymore as the model identifier is "Mac14,8".

2794 single core, 21453 multi-core

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

196543 Metal

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

113870 OpenCL

Mac14,8 - Geekbench

Benchmark results for a Mac14,8 with an Apple M2 Ultra processor.

browser.geekbench.com

Interesting it scored lower than the Studios on metal. I notice it’s an entry level RAM config, given that the ram is shared with the GPU, could it be a reason for the lower score?

I notice with the Studios tested that the ones equipped with 128 or 192GB get 220K metal, whereas those with 64GB seem to be closer to 200K On average.

 

[Dopemaster]

Apple M2 Studio Learns Nuke Machine Learning

The new Nuke with Apple Silicon native CopyCat shows why the new Apple M2 Ultra is a great machine learning VFX tool.

www.fxguide.com

Headline: M2 Ultra runs ML in VFX compositing software Nuke more than twice as fast as high-end PCs with current high-end graphics cards.

 

[Matty_TypeR]

My Mac pro 2019 scores on geekbench for comparison. Cant find any Mac Pro 8.1 results in the charts.

 

[Boil]

Cant find any Mac Pro 8.1 results in the charts.

Maybe try looking for Mac14,8

;^p

 

[Matty_TypeR]

Maybe try looking for Mac14,8

;^p

Ahhh now i see them thanks. My 6900xt is not so bad compared in metal. So quite pleased with that. You can see why Apple wont support the 7900xtx though as it would most likely smash the M2 Ultra.

Single core and multicore scores are quite different, but i see no need to upgrade to M2 yet as i need native boot windows to put my 4090 to work.

 

[Boil]

Ahhh now i see them thanks. My 6900xt is not so bad compared in metal. So quite pleased with that. You can see why Apple wont support the 7900xtx though as it would most likely smash the M2 Ultra.

Single core and multicore scores are quite different, but i see no need to upgrade to M2 yet as i need native boot windows to put my 4090 to work.

I am of a mind that the M3-series will bring some much needed improvements to the GPU in Apple silicon; new core design, more cores, hardware ray-tracing, some sort of "M3 Extreme" configuration...?

 

[Matty_TypeR]

I am of a mind that the M3-series will bring some much needed improvements to the GPU in Apple silicon; new core design, more cores, hardware ray-tracing, some sort of "M3 Extreme" configuration...?

The M3 series needs to be extreme in many ways, more Ram, and a GPU to support Ray tracing as Nvidia and AMD do, and any game ports to metal people will want to see RT in them. does Metal even support RT?

The M2 is just a stop gap to say they are out of intel and AMD, but they could have let the buyer's of intel Mac pro 7.1's which were on sale just a few weeks ago have a last AMD update to 7900xtx cards as people spent alot on the 7.1 machines which were toted at the time to be apples road map forward.

The M3 better be something real special and they will need to get some details out soon to stop people going PC workstation as the M2 mac pro is just a studio with gimped PCIe.

 

[dmr727]

does Metal even support RT?

It does, which I why I think people are optimistic that we'll see hardware RT in M3.

 

[Boil]

The M3 series needs to be extreme in many ways, more Ram, and a GPU to support Ray tracing as Nvidia and AMD do, and any game ports to metal people will want to see RT in them. does Metal even support RT?

Yes, Metal supports RT right now...

I feel the delay with 3nm really hurt whatever plans Apple had for high-end ASi headless desktops...

M3 Extreme (or whatever) on a 3nm process, with new core designs, and hardware ray-tracing; that should give a significant "iGPU" performance boost...

Apple could also come out with an ASi GPGPU add-in card (which could also be used in an eGPU enclosure, so more sales to those ASi Mac users without an ASi Mac Pro) to give performance increases in compute/render jobs...

The "iGPU" in the M3 Extreme SoC would give RTX 5090 performance for display output, while the ASi GPGPU add-in cards would give equal (or better with more than one add-in card) performance for compute/render jobs...

 

[ZombiePhysicist]

So the metal scores are indeed slower than a 6900XT and I think slower than the 6800XT. Yeash

 

[Synchro3]

Apple’s M2 Ultra SoC – No faster than last year’s desktop CPUs from AMD and Intel and slower than an NVIDIA RTX 4080.​

Apple’s M2 Ultra SoC - No faster than last year’s desktop CPUs from AMD and Intel and slower than an NVIDIA RTX 4080. | igor´sLAB

The Apple M2 Ultra SoC represents the company's top model. With an impressive number of 134 billion transistors, it offers outstanding performance. The CPU consists of a total of 24 cores…

www.igorslab.de

 

[ZombiePhysicist]

Apple’s M2 Ultra SoC – No faster than last year’s desktop CPUs from AMD and Intel and slower than an NVIDIA RTX 4080.​

Apple’s M2 Ultra SoC - No faster than last year’s desktop CPUs from AMD and Intel and slower than an NVIDIA RTX 4080. | igor´sLAB

The Apple M2 Ultra SoC represents the company's top model. With an impressive number of 134 billion transistors, it offers outstanding performance. The CPU consists of a total of 24 cores…

www.igorslab.de

M2 Ultra is slower than a 4060, and slower than a 6800XT as evinced by real world use by users on this forum. It's bad enough it's that bad on GPU, but what is unforgivable is how bad it is on single/multi-thread compared to a pathetic i9.

It's really a catastrophic fail. The Mac tech press is cheering that the boat was just hit by an iceberg. Bizarre propaganda.

 

[impulse462]

M2 Ultra is slower than a 4060, and slower than a 6800XT as evinced by real world use by users on this forum. It's bad enough it's that bad on GPU, but what is unforgivable is how bad it is on single/multi-thread compared to a pathetic i9.

It's really a catastrophic fail. The Mac tech press is cheering that the boat was just hit by an iceberg. Bizarre propaganda.

But how will they get the next product for free then???

Also yeah, M2 ultra opencl score is still lower than my titan rtx which is now 5 years old at this point.

From a pure engineering perspective, I personally think that score is pretty impressive given the power usage and its all on the same package. But since the 1) apologists always talk about "real world performance" and 2) for a workstation, power consumption doesn't mean much, its definitely lacking.

 

[jeanlain]

Also yeah, M2 ultra opencl score is still lower than my titan rtx which is now 5 years old at this point.

OpenCL really shouldn't be used on the Mac. It runs under a translation layer and no app uses it anyway.
With Metal, the 76-core variant is on par with the best AMD card.

On the 3DMark benchmark, it's on par with the 7900XT and the 3090, at much lower power consumption I suppose.

 

[Matty_TypeR]

OpenCL really shouldn't be used on the Mac. It runs under a translation layer and no app uses it anyway.
With Metal, the 76-core variant is on par with the best AMD card.

On the 3DMark benchmark, it's on par with the 7900XT and the 3090, at much lower power consumption I suppose.

So the M2 ultra is running 3D mark in VM windows? and is on par with a 7900xt and 3090 i would like to see those results if you have a link? the CPU score would be interesting alone.

Lets hope they tried port Royal test, oh they cant no RT on a M2 chip.

 

[thunng8]

So the M2 ultra is running 3D mark in VM windows? and is on par with a 7900xt and 3090 i would like to see those results if you have a link? the CPU score would be interesting alone.

Lets hope they tried port Royal test, oh they cant no RT on a M2 chip.

I believe the mac ultra is running the ipad application to do the comparisons.

There are numerous real world tests that the m2 Ultra is faster than the RTX4090. Depends on what is tested.

e.g. Da Vinci Resolve
Capture One
Adobe Lightroom Classic

This thread is only dealing with synthetic benchmarks like Geekbench. Real world benchmarks can paint a different picture.

 

[impulse462]

I believe the mac ultra is running the ipad application to do the comparisons.

There are numerous real world tests that the m2 Ultra is faster than the RTX4090. Depends on what is tested.

e.g. Da Vinci Resolve
Capture One
Adobe Lightroom Classic

This thread is only dealing with synthetic benchmarks like Geekbench. Real world benchmarks can paint a different picture.

Its faster because of dedicated accelerators that the 4090 doesn't have, to be clear. its not faster at raw compute/FLOPS

 

[thunng8]

Its faster because of dedicated accelerators that the 4090 doesn't have, to be clear. its not faster at raw compute/FLOPS

Nope incorrect. Lightroom classic and capture one are Cpu and gpu intensive. Possibly because image editing take advantage of the unified memory.

Resolve takes advantage of dedicated encoders in combination with gpu. 4090 has dedicated hardware for both as well, so it is a fair comparison.

 

[Matty_TypeR]

Its hard to see how IOS 3d mark can be compared to a 4090 when there are no drivers for the 4090 in IOS but would be interesting to see windows VM scores on timespy EX. As people use so many apps for there work i guess it will depend what suits them as to what they purchase. I would like to see what a 7900xtx would do if supported by apple. but i think that will not happen.

 

[impulse462]

Nope incorrect. Lightroom classic and capture one are Cpu and gpu intensive. Possibly because image editing take advantage of the unified memory.

Resolve takes advantage of dedicated encoders in combination with gpu. 4090 has dedicated hardware for both as well, so it is a fair comparison.

do you have a source for any of this?

 

[thunng8]

do you have a source for any of this?

What do you mean, are you saying Nvidia does not have specialized encoders and decoders?

Of course they do and have for a long time with each new gpu getting better. Here is the support matrix.

Video Encode and Decode GPU Support Matrix

Find the related video encoding and decoding support for all NVIDIA GPU products.

developer.nvidia.com

 

[thunng8]

Its hard to see how IOS 3d mark can be compared to a 4090 when there are no drivers for the 4090 in IOS but would be interesting to see windows VM scores on timespy EX. As people use so many apps for there work i guess it will depend what suits them as to what they purchase. I would like to see what a 7900xtx would do if supported by apple. but i think that will not happen.

the comparison is 3d mark for iOS vs 3d mark for windows using the same graphics test at the same resolution. Of course 4090 cannot run iOS.

iOS version does not have every graphics test just slingshot, wildlife and wildlife extreme.

Also intel macs cannot run iOS/iPad apps, just Apple Silicon macs can(also depends if develop allows this)