M1 Rosetta Performance

[Rastafabi]

As Geekbench results are now available for the new M1 chips I was wandering about the Rosetta performance. As currently no results regarding this are available I compared the MacMini Development kit's performance of Rosetta and native benchmarks.
Finding that Rosetta resembles 75% of native single-core performance and 60% of multi-core performance (as the four low powered cores are not taken into account [changed from ADP to M1]) I estimated the M1 Rosetta performance.

Geekbench scores	single-core score​	multi-core score​
M1 Native​	~1700​	~7000​
M1 Rosetta (estimated)	1275​	4200​
M1 Rosetta Benchmark	~1300​	~6000​
Surface Pro X SQ2 (native)	~800​	~3100​
i7-1185G7​	~1600​	~6000​
AMD Ryzen 7 4800U	~1100​	~6800​
​

For the each CPUs datapoint I tended to choose those with high scores levelling out between single and multi-core performance.


It is striking that for instance the M1 MacBook Air is pretty close to the performance of the top of the line 2020 10th Gen Intel i7 MacBook Pro, which retails for double the price.

M1 Graphic Performance (Rosetta)
M1 Rosetta GPU performance impact	~10%

---

11-15: Added actually (very impressive!) Rosetta Benchmark results found by @wyrdness & Surface Pro X results
11-15 (2): Added current Intel and AMD offerings in the 10-28W range.
11-16: Added some Graphics results.

 

[johnhackworth]

I'm not worried about Rosetta 2 performance in terms of pure processor throughput - it's Rosetta 2 + x86 apps + 16Gb RAM which is potentially problematic for me.

I'm looking to replace my 2012 27" iMac, 24Gb RAM with a Mini. The 16Gb memory limit on the M1 Mini is a worry though.

On one hand I'm assuming that ARM-native apps will be more memory efficient than x86 apps - based on what iOS devices can already do with very limited RAM - on the other hand I'm also assuming that x86 apps on Rosetta 2 will require more memory due to the emulation which is required.

 

[FranApple]

I'm not worried about Rosetta 2 performance in terms of pure processor throughput - it's Rosetta 2 + x86 apps + 16Gb RAM which is potentially problematic for me.

I'm looking to replace my 2012 27" iMac, 24Gb RAM with a Mini. The 16Gb memory limit on the M1 Mini is a worry though.

On one hand I'm assuming that ARM-native apps will be more memory efficient than x86 apps - based on what iOS devices can already do with very limited RAM - on the other hand I'm also assuming that x86 apps on Rosetta 2 will require more memory due to the emulation which is required.

Emulation seems to be handled during the installation phase. In other words, wont ongoing emulation be moot? meaning additional RAM won’t be necessary? Just thinking out loud.

 

[Rastafabi]

I think the current 16GB RAM limitation is a business decision in multiple regards. first of the it's far simpler do manufacture eliminating an additional 32GB option. Additionally this keeps people who need it from buying them (obviously) and thus there is a lesser risk of supply shortages. And also those Macs might differentiate themselves better from yet to come M1X (?) models. Also as the ram is apparently a lot faster than traditional DIMM modules compressing the ram is much less a issue and can be used much more aggressively than it is already being done today. I second, that I do not think that Intel applications would consume significantly more ram than native ones.

 

[johnhackworth]

Emulation seems to be handled during the installation phase. In other words, wont ongoing emulation be moot? meaning additional RAM won’t be necessary? Just thinking out loud.

Aha, I suspect you know more about Rosetta 2 than me. Guess it depends on the memory efficiency of the resulting transcoded ARM binary.

Ironically, the PPC to Intel transition was RISC to CISC and the current transition is from CISC to RISC - I accept that that's quite a simplistic view

 

[wyrdness]

I've just spotted what is probably the first Rosetta 2 Geekbench result for M1:

MacBookAir10,1 - Geekbench

Benchmark results for a MacBookAir10,1 with a VirtualApple processor.

browser.geekbench.com

Single core - 1313
Multi core - 5888

That's better than I was expecting. I'd guesstimated around 1200/5000.

If this is correct, then it's faster than the fastest Intel mac for a single core!

For multi-core, it's similar to AMD's Ryzen 7 4800U, but it beats the Ryzen in single core.

 

[ArPe]

I've just spotted what is probably the first Rosetta 2 Geekbench result for M1:

MacBookAir10,1 - Geekbench

Benchmark results for a MacBookAir10,1 with a VirtualApple processor.

browser.geekbench.com

Single core - 1313
Multi core - 5888

That's better than I was expecting. I'd guesstimated around 1200/5000.

If this is correct, then it's faster than the fastest Intel mac for a single core!

For multi-core, it's similar to AMD's Ryzen 7 4800U, but it beats the Ryzen in single core.

If your Intel apps don’t use much CPU power you won’t notice much real world difference unless those apps need a bucket load of RAM and graphics power.

I just hope Rosetta stays around longer this time for people who have x86 games that will never be updated. Last time around Rosetta was dropped in about 2 years.

 

[thenewperson]

I just hope Rosetta stays around longer this time for people who have x86 games that will never be updated. Last time around Rosetta was dropped in about 2 years.

5 iirc, but agreed. This time it's Apple's tech so maybe that'll keep it around longer?

 

[UltimateSyn]

But but but the RAM tops out at 16GB, these aren’t Pro machines!!

 

[CalMin]

I'm looking to replace my 2012 27" iMac, 24Gb RAM with a Mini. The 16Gb memory limit on the M1 Mini is a worry though.

Then, for goodness sakes, wait until gen 2 of Apple Silicon. It seems like you keep your Macs for a long time. Why not sit out the M1 generation and wait until more RAM is available as an option.

 

[UltimateSyn]

Then, for goodness sakes, wait until gen 2 of Apple Silicon. It seems like you keep your Macs for a long time. Why not sit out the M1 generation and wait until more RAM is available as an option.

Won’t need to wait for the second generation, higher-end variants of the Gen 1 M1 (M1X, M1Z, etc.) released between now and the M2 will certainly support more RAM for the machines that.... support more RAM.

OP, if you want 32GB of RAM then don’t buy an entry-level Mac that doesn’t support 32GB of RAM. Not very complicated.

 

[Spindel]

I'm not worried about Rosetta 2 performance in terms of pure processor throughput - it's Rosetta 2 + x86 apps + 16Gb RAM which is potentially problematic for me.

I'm looking to replace my 2012 27" iMac, 24Gb RAM with a Mini. The 16Gb memory limit on the M1 Mini is a worry though.

On one hand I'm assuming that ARM-native apps will be more memory efficient than x86 apps - based on what iOS devices can already do with very limited RAM - on the other hand I'm also assuming that x86 apps on Rosetta 2 will require more memory due to the emulation which is required.

Have you ever used up all of that ram?

I have a 2013 iMac with 24 GB ram and I have used more than 16 GB only a handful of times. And on all of those occations it has been cached webpages in ram and not data I actually need to have in RAM.

 

[KPOM]

It appears that GeekBench “sees” the M1 as an “Intel Family 6 Model 44” chip. Which “generation” is that? My best guess is that Rosetta 2 makes the M1 appear to be a really fast Sandy Bridge processor to x86 software.

 

[Icelus]

It appears that GeekBench “sees” the M1 as an “Intel Family 6 Model 44” chip. Which “generation” is that? My best guess is that Rosetta 2 makes the M1 appear to be a really fast Sandy Bridge processor to x86 software.

Sounds like an Intel Westmere (44 = 2C hexadecimal), a CPU with SSE 4.2 support but w/o AVX support (AVX isn’t supported by Rosetta).

 

[Serban55]

But but but the RAM tops out at 16GB, these aren’t Pro machines!!

Think about what mac pro users says
But but but less than 128 gb ram is not a pro machiene
In these days even if you are on the last place as an athlete you still are considered a pro athlete
So everything is subjective

 

[russell_314]

Think about what mac pro users says
But but but less than 128 gb ram is not a pro machiene
In these days even if you are on the last place as an athlete you still are considered a pro athlete
So everything is subjective

"Pro" is just short for and means "professional" and not "A computer I can impress my friends about the specs". Different professionals need different things. The professional video editor at Disney needs different things than the one writing the script.

 

[Homy]

The single-core is still faster than 10-core i9-10910 3.6 GHz in iMac and the multi-core is on par with 6-core i5-10500 3.1 GHz in iMac 27". I was planning to buy an Intel iMac 3.1 GHz if Apple Silicon didn't impress but this blows my mind. Imagine how fast the iMac M1/2/X/Z will be next year. Now I only hope that the GPU in iMac M can deliver too. Then we could play all our Intel games at least as fast. There will be some graphical bugs for sure but this is very promising and just the beginning.

 

[Rastafabi]

The single-core is still faster than 10-core i9-10910 3.6 GHz in iMac and the multi-core is on par with 6-core i5-10500 3.1 GHz in iMac 27". […] Now I only hope that the GPU in iMac M can deliver too. […]

At least for those short-term, non sustained benchmarks apple has delivered an impressive chip. I'm eager to see "real world" sustained performance and corresponding benchmarks like Cinebench.

Regarding the graphics the M1 apparently excels as well, though the same thoughts as above apply here as well. Whats noteworthy here however is, that according to Geekbench there is a much smaller performance impact for graphics if the binaries are executed via Rosetta.

 

[Botts85]

Has anyone seen a Cinebench R23 result with on an Apple Silicon Mac running the X86 and ARM version of Cinebench?

I'd love to see in Cinebench what the Rosetta 2 impact is.

 

[acidfast7_redux]

Has anyone seen a Cinebench R23 result with on an Apple Silicon Mac running the X86 and ARM version of Cinebench?

I'd love to see in Cinebench what the Rosetta 2 impact is.

yes. a hit of 19-20%

 

[Homy]

League of Legends on MBA
2560x1600 Very High 50-60 fps

Max Tech said that the game was glitchy with dropped frames on a MBP 2018 Radeon Pro 555X.

Rise of the Tomb Raider on Mac Mini
1920x1080 Very High FXAA 39.6 fps

Shadow of the Tomb Raider on MBA
1920x1200 lowest settings 38 fps
1920x1200 highest settings 20 fps
2560x1600 lowest settings 25 fps
2560x1600 highest settings 13 fps

Fortnite on MBP 13"
Upscaled 2560x1600 3D resolution 75% 1920x1200 high settings 40 fps

 

[Botts85]

yes. a hit of 19-20%

I thought this was based on the dev kit, I hadn't seen a production bench showing that.

 

[Homy]

25 games tested. Very impressive! I mean Bioshock 2 Remastered, Borderlands 2-3 and Deus Ex Mankind Divided on ultra settings on iGPU through Rosetta??? WOW!