A bunch of new mac studio benchmarks dropped

[dugbug]

Geekbench Search - Geekbench

browser.geekbench.com

BLUF multicore:
M1 Max: 12.6K
M1 Ultra: 24K

looks like one fake one by 'fu apple' someone hurt

and some legit ones:
ystem Mac13,1 Apple M1 Max 3221 MHz (10 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1776
Multi-Core Score 12780

System Mac13,1 Apple M1 Max 3221 MHz (10 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1777
Multi-Core Score 12645

System Mac13,2 Apple M1 Ultra 3221 MHz (20 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1785
Multi-Core Score 23942

System Mac13,2 Apple M1 Ultra 3220 MHz (20 cores)
Uploaded March 9th, 2022
iro
Platform macOS
Single-Core Score 1793
Multi-Core Score 24055

System Mac13,2 Apple M1 Ultra 3220 MHz (20 cores)
Uploaded March 8th, 2022
Platform macOS
Single-Core Score 1793
Multi-Core Score 24055

that head-banging multicore score seems to hold. To get an idea of just how big 24K is on multicore, here are the top CPU scores. The only one that beat the ultra? The $8000 3990X cpu.

AMD Ryzen Threadripper 3990X 2.9 GHz (64 cores)	25157
AMD Ryzen Threadripper 3970X 3.7 GHz (32 cores)	22334
Intel Xeon W-3175X 3.1 GHz (28 cores)	20910

 

[page3]

Why were they dropped?

 

[Bodhitree]

Seems these benchmarks expose some underlying limitations in the interconnects of the non-Apple chips. Its pretty impressive to see the multi-core score for 20 cores come in so close to double the score for 10 cores, that bodes well for future chips.

 

[sfwalter]

Why were they dropped?

Someone has butter fingers

 

[page3]

Someone has butter fingers

I had to go and look the phrase up.

Two countries separated by a common language and all that…

 

[dugbug]

I had to go and look the phrase up.

Two countries separated by a common language and all that…

oh gotcha Just saying they are brand new. There are now 3 Max studios and 4 Ultras.

-d

 

[cmaier]

Seems these benchmarks expose some underlying limitations in the interconnects of the non-Apple chips. Its pretty impressive to see the multi-core score for 20 cores come in so close to double the score for 10 cores, that bodes well for future chips.

The problem with x86 chips is that typically in order to run all of the cores simultaneously, you have to drop the per-core clock rate, due to thermals and IR drop.

 

[Bodhitree]

The problem with x86 chips is that typically in order to run all of the cores simultaneously, you have to drop the per-core clock rate, due to thermals and IR drop.

That’s really interesting. I wonder if it holds for the chips inside the PlayStation 5 and Xbox Series X, where the sustained load would generally be spread across all the cores, including the GPU.

Anyway Apple seems to have a major architectural advantage in this area.

 

[ader42]

but we wants compute metal scoresies

 

[crazy dave]

looks like one fake one by 'fu apple' someone hurt

More fakes now using a variety of … equally “creative” names

 

[Ethosik]

I am definitely getting the M1 Ultra, I am just waiting on benchmarks and real world tests if the higher GPU cores will be better and worth the price difference.

 

[donth8]

but we wants compute metal scoresies

Here you go: https://browser.geekbench.com/search?k=v5_compute&q=M1+ultra&utf8=✓

 

[Appletoni]

Geekbench Search - Geekbench

browser.geekbench.com

BLUF multicore:
M1 Max: 12.6K
M1 Ultra: 24K

looks like one fake one by 'fu apple' someone hurt

and some legit ones:
ystem Mac13,1 Apple M1 Max 3221 MHz (10 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1776
Multi-Core Score 12780

System Mac13,1 Apple M1 Max 3221 MHz (10 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1777
Multi-Core Score 12645

System Mac13,2 Apple M1 Ultra 3221 MHz (20 cores)
Uploaded March 10th, 2022
Platform macOS
Single-Core Score 1785
Multi-Core Score 23942

System Mac13,2 Apple M1 Ultra 3220 MHz (20 cores)
Uploaded March 9th, 2022
iro
Platform macOS
Single-Core Score 1793
Multi-Core Score 24055

System Mac13,2 Apple M1 Ultra 3220 MHz (20 cores)
Uploaded March 8th, 2022
Platform macOS
Single-Core Score 1793
Multi-Core Score 24055

that head-banging multicore score seems to hold. To get an idea of just how big 24K is on multicore, here are the top CPU scores. The only one that beat the ultra? The $8000 3990X cpu.

AMD Ryzen Threadripper 3990X 2.9 GHz (64 cores)	25157
AMD Ryzen Threadripper 3970X 3.7 GHz (32 cores)	22334
Intel Xeon W-3175X 3.1 GHz (28 cores)	20910

The Mac Studio will be used soon but the AMD Ryzen Threadripper 5990X will be also used soon.
I think it is better to compare these two and not using the old 3990X to show how good Apple is compared to a very old CPU xD.

 

[Appletoni]

I am definitely getting the M1 Ultra, I am just waiting on benchmarks and real world tests if the higher GPU cores will be better and worth the price difference.

Don’t look at Apples improved video developer benchmark.
Take a look at other benchmarks like:
KataGo benchmark
LC0 benchmark
Stockfish benchmark - used by more than 200.000.000 people.

 

[jeffpeng]

The Mac Studio will be used soon but the AMD Ryzen Threadripper 5990X will be also used soon.
I think it is better to compare these two and not using the old 3990X to show how good Apple is compared to a very old CPU xD.

5000 Series Threadripper will be PRO exclusive, so you won't see the 64 core variant in a sub 10K workstation. I know this sounds rather confusing, but as far as professional computers go, the Mac Studio is surprisingly affordable. 6 grand for the Ultra machine with 128 gigs is not expensive by any stretch of the imagination if you compare it with competing machines from HP, Dell and Lenovo.

That being said: The 3970X hits 30k on Linux quite easily. Windows is just a dog shoot efficiently allocating threads. 3990X goes past 35K. Since a 3970X scales almost linearly from the 3950X, and the 5950X hits 18K, I guess it's fair to expect a 32 core 5000 series Threadripper to hit 35K+, and the 64 core variant to break 40, maybe better.

But these are systems that suck tremendous power from the wall, come in huge form factors, are certainly not quiet and will come not cheaper than the Mac Studio, even in the 128Gig Ultra configuration. So.... if you don't need expansions or can live with the limitations the Mac Studio has regarding those .... it's almost a bargain.

 

[the8thark]

Someone has butter fingers

They were all thumbs

 

[Appletoni]

5000 Series Threadripper will be PRO exclusive, so you won't see the 64 core variant in a sub 10K workstation. I know this sounds rather confusing, but as far as professional computers go, the Mac Studio is surprisingly affordable. 6 grand for the Ultra machine with 128 gigs is not expensive by any stretch of the imagination if you compare it with competing machines from HP, Dell and Lenovo.

That being said: The 3970X hits 30k on Linux quite easily. Windows is just a dog shoot efficiently allocating threads. 3990X goes past 35K. Since a 3970X scales almost linearly from the 3950X, and the 5950X hits 18K, I guess it's fair to expect a 32 core 5000 series Threadripper to hit 35K+, and the 64 core variant to break 40, maybe better.

But these are systems that suck tremendous power from the wall, come in huge form factors, are certainly not quiet and will come not cheaper than the Mac Studio, even in the 128Gig Ultra configuration. So.... if you don't need expansions or can live with the limitations the Mac Studio has regarding those .... it's almost a bargain.

We would probably buy the MacBook Pro 18-inch with M1 or M2 Ultra chip and all-in configuration.

 

[ader42]

For the Metal scores, M1 Max in a MBP 16” gets about 66k, and the Studio Ultra gets about 103k.
To be honest I was expecting a bigger jump due to more power and bigger cooling in desktop.

Not easy for me to decide between 16” M1 Max MBP with 10/32 cores and Studio Ultra with 20/48 cores.

 

[Appletoni]

For the Metal scores, M1 Max in a MBP 16” gets about 66k, and the Studio Ultra gets about 103k.
To be honest I was expecting a bigger jump due to more power and bigger cooling in desktop.

Not easy for me to decide between 16” M1 Max MBP with 10/32 cores and Studio Ultra with 20/48 cores.

If you want better cooling use a professional cooling paste or pads.
20 cores sounds better to me than 10 cores.
64 cores sounds better to me than 32 cores.
Don’t forget the long term usage if you don’t want to buy every year a new device.
If you need to be very mobile like I do, then go for the MBP 16-inch display with M1 MAX chip with all-in configuration.

 

[crazy dave]

For the Metal scores, M1 Max in a MBP 16” gets about 66k, and the Studio Ultra gets about 103k.
To be honest I was expecting a bigger jump due to more power and bigger cooling in desktop.

Not easy for me to decide between 16” M1 Max MBP with 10/32 cores and Studio Ultra with 20/48 cores.

Doesn’t seem to say if that’s 48 or 64 cores, probably 64. One thing to keep in mind is that the GB5 compute test didn’t scale very well from M1 Pro to Max. According to the Anandtech guys, it’s because the M1 Max AGX takes awhile to spin up its clocks and the individual subtests are over so fast that it never gets up to full speed. Since the Ultra is two M1 Maxes glued together … it makes sense that it would have a similar dynamic.

 

[crazy dave]

That being said: The 3970X hits 30k on Linux quite easily. Windows is just a dog shoot efficiently allocating threads. 3990X goes past 35K.

I think part of it is also GB on Windows rather than just Windows. At least I remember Jeff Pool admitting that high core count GB5 Windows results were suspect and it was something they were working on improving. Though the fact that normal naive scheduling doesn’t work but does on Linux/Mac does kind of imply the Windows (10 and below?) scheduler isn’t very good.

 

[jeanlain]

One M1 Ultra shows up in the GFXBench database.

GFXBench - Unified cross-platform 3D graphics benchmark database

The first unified cross-platform 3D graphics benchmark database for comparing Android, iOS, Windows 8, Windows Phone 8 and Windows RT capable devices based on graphics processing power.

gfxbench.com

Its score is a bit lower than the best competitors, but this is a single test that may not be the 64-core variant. In other subtests it scores less than the M1 Max, which doesn't make sense.

 

[crazy dave]

One M1 Ultra shows up in the GFXBench database.

GFXBench - Unified cross-platform 3D graphics benchmark database

The first unified cross-platform 3D graphics benchmark database for comparing Android, iOS, Windows 8, Windows Phone 8 and Windows RT capable devices based on graphics processing power.

gfxbench.com

Its score is a bit lower than the best competitors, but this is a single test that may not be the 64-core variant. In other subtests it scores less than the M1 Max, which doesn't make sense.

If it's the 48-core variant then indeed it has perfect scaling with the M1 Max in Aztec High offscreen. The other subtests may be topping out and the bigger GPU isn't being saturated and spinning out even slower.

 

[jeanlain]

If it's the 48-core variant then indeed it has perfect scaling with the M1 Max in Aztec High offscreen. The other subtests may be topping out and the bigger GPU isn't being saturated and spinning out even slower.

GFXBench barely uses the CPU, I don't see why it shouldn't saturate the GPU in all offscreen tests.
And there is at least one subtest (Manhattan 1440 offscreen) where the M1 ultra is more than 1.5 times better than the best M1 Max, which suggests that this is the 64-core variant... and that performance scaling is therefore not great (about 1.5x instead of 2x).
We now have two tests of the M1 ultra showing almost identical results (the average being very close to the top result). So this appears to the representative of the what the M1 ultra GPU can do in this benchmark app. It's clearly not faster than the 3090 except in "Manhattan".

 

[senttoschool]

that head-banging multicore score seems to hold. To get an idea of just how big 24K is on multicore, here are the top CPU scores. The only one that beat the ultra? The $8000 3990X cpu.

AMD Ryzen Threadripper 3990X 2.9 GHz (64 cores)	25157
AMD Ryzen Threadripper 3970X 3.7 GHz (32 cores)	22334
Intel Xeon W-3175X 3.1 GHz (28 cores)	20910

Is there an official explanation for why the 64 core variant is only 3,000 points higher than the 32 core variant? Something is not right here.