LN2 Cooled Apple M4 Chip Surpasses Single-Core Performance of M3 Max and M2 Ultra

[AleksandarK]

According to Geekerwan, Apple's latest M4 silicon has achieved a remarkable milestone by using liquid nitrogen to chill Apple's M4 iPad Pro. This unconventional approach unlocked great single-core performance, surpassing even the M3 Max and M2 Ultra processors in Geekbench v6 benchmark tests. The setup involved cooling the M4 iPad Pro, equipped with a 3+6 core configuration, using a Kingpin Cooling T-Rex Rev 4 CPU LN2 pot filled with liquid nitrogen. This extreme cooling allowed the M4 processor to operate at an astonishing 4.41 GHz during the benchmark run, resulting in a staggering single-core score of 4,001 points. This score represents a 28% increase over the M3 Max found in the 16-inch MacBook Pro and an impressive 44% improvement over the M2 Ultra powering the Mac Studio.

Notably, the M4's single-core performance is capable of reaching scores in the 3,000s. With liquid nitrogen cooling, it suprases the 4,000-point mark, making this achievement all the more remarkable. While the M4's multi-core performance did not match the lofty expectations set by its single-core power, it still managed to achieve a score of 13,595 points, outperforming both the M3 Max and M2 Ultra, which scored 20,957 and 21,330 points, respectively. This was done on the 3+6 core configuration with three P-cores and six E-cores, which is not the top-end M4 configuration. This shows that with adequate cooling, like MacBooks, the upcoming M4 Pro and M4 Max chips could achieve much higher performance than their predecessors.



View at TechPowerUp Main Site | Source

 

[ZoneDymo]

"This shows that with adequate cooling, like MacBooks, the upcoming M4 Pro and M4 Max chips could achieve much higher performance than their predecessors."

But does it really?

 

[Kohl Baas]

This is more nuts than a jar of peanut butter.

 

[Carillon]

That's wild, the newer chip under extreme cooling is faster than the old chips!

 

[Parn]

For a desktop with LN2 maybe... but for a supposed handheld device how impractical that is!

 

[azrael]

For a desktop with LN2 maybe... but for a supposed handheld device how impractical that is!

You're just not thinking far enough outside the box. I'm sure Apple will happily sell you a portable LN2 cooler and the matching iRucksack. Just you wait, it'll sell like hotcakes...

 

[Daven]

“…it still managed to achieve a respectable score of 13,595 points, outperforming both the M3 Max and M2 Ultra, which scored 20,957 and 21,330 points, respectively.”

How does a score of 13,500 outperform scores over 20,000?

 

[_Under2World_]

“…it still managed to achieve a respectable score of 13,595 points, outperforming both the M3 Max and M2 Ultra, which scored 20,957 and 21,330 points, respectively.”

How does a score of 13,500 outperform scores over 20,000?

The other chip have more core so better multi core result. The article was about single core performance and they finished with a small note about multi-core

 

[Daven]

The other chip have more core so better multi core result. The article was about single core performance and they finished with a small note about multi-core

Its a error by the author obviously and the article needs correcting. From the original Tom’s article:

“The M4's multi-core performance didn't impress, however. It was 54% slower than the M3 Max and fell behind the M2 Ultra by up to 57%. Geekerwan submitted various Geekbench 6 entries, with the highest multi-core result of 14,785 points, so the M4 still lies behind Apple's older silicon.”

 

[Vayra86]

So the bottom line here then is that sans LN2 the M4 is actually worse in every way except cost

 

[Darmok N Jalad]

I think some might be missing the point. Just look at it empirically. Normally people do heavy overclocks with LN2, but in this case, no overclocking can be done, so this test is completely lifting the thermal restraints of a passively-cooled system (one that has little cooling capacity to work with). While it's probably not a true representation of what the actively cooled M4 Macs will do, it's also probably not that far off. And of course it's not going to outperform the Max and Ultra in MC, those two chips have considerably more P cores and more memory bandwidth, and this isn't even the best M4 one can get in the iPad. It's a bit of a silly experiment, but we can stick a pin in these results and see how accurate they are when the M4 Macs come onto the scene later this year.

 

[SOAREVERSOR]

I think some might be missing the point. Just look at it empirically. Normally people do heavy overclocks with LN2, but in this case, no overclocking can be done, so this test is completely lifting the thermal restraints of a passively-cooled system (one that has little cooling capacity to work with). While it's probably not a true representation of what the actively cooled M4 Macs will do, it's also probably not that far off. And of course it's not going to outperform the Max and Ultra in MC, those two chips have considerably more P cores and more memory bandwidth, and this isn't even the best M4 one can get in the iPad. It's a bit of a silly experiment, but we can stick a pin in these results and see how accurate they are when the M4 Macs come onto the scene later this year.

Are the M4 Macs going to come with LN2?

 

[Denver]

I don't know if it was intended, but the statement in the title is funny.

Breaking news: "Ryzen 7600 stock is faster than 5950x in ST."

 

[Vayra86]

I think some might be missing the point. Just look at it empirically. Normally people do heavy overclocks with LN2, but in this case, no overclocking can be done, so this test is completely lifting the thermal restraints of a passively-cooled system (one that has little cooling capacity to work with). While it's probably not a true representation of what the actively cooled M4 Macs will do, it's also probably not that far off. And of course it's not going to outperform the Max and Ultra in MC, those two chips have considerably more P cores and more memory bandwidth, and this isn't even the best M4 one can get in the iPad. It's a bit of a silly experiment, but we can stick a pin in these results and see how accurate they are when the M4 Macs come onto the scene later this year.

Euhhhh yeah I suppose? And then you know what exactly, that LN2 is still faster than active cooling in a small form factor.

We could also do that exercise today...

 

[Jism]

I think some might be missing the point. Just look at it empirically. Normally people do heavy overclocks with LN2, but in this case, no overclocking can be done, so this test is completely lifting the thermal restraints of a passively-cooled system (one that has little cooling capacity to work with). While it's probably not a true representation of what the actively cooled M4 Macs will do, it's also probably not that far off. And of course it's not going to outperform the Max and Ultra in MC, those two chips have considerably more P cores and more memory bandwidth, and this isn't even the best M4 one can get in the iPad. It's a bit of a silly experiment, but we can stick a pin in these results and see how accurate they are when the M4 Macs come onto the scene later this year.

Yep. It only shows the unrestraint power of the chip while having zero thermal restrictions.

It's still operating within spec, pre-set voltages and such.

 

[MacZ]

LN2 and Geekbench : doubly useless

 

[Darmok N Jalad]

Are the M4 Macs going to come with LN2?

Of course not, but they will actually have active cooling. The LN2 isn’t overclocking the chip. It’s keeping it from thermally throttling. It’s also not likely capable of drawing excessive power like a desktop SOC, since it’s in an iPad.

Euhhhh yeah I suppose? And then you know what exactly, that LN2 is still faster than active cooling in a small form factor.

We could also do that exercise today...

I suppose if they could take the iPad apart and put a traditional air cooler on it we could find out, but I don’t know how easy that would be. Besides, I didn’t claim it was how a desktop M4 would perform, but it’s probably not that far off either. A desktop M4 will probably be given more freedom with power consumption, not just thermals.

 

[erek]

“…it still managed to achieve a respectable score of 13,595 points, outperforming both the M3 Max and M2 Ultra, which scored 20,957 and 21,330 points, respectively.”

How does a score of 13,500 outperform scores over 20,000?

Article was partially written and interpreted another source using AI

 

[LabRat 891]

Yep. It only shows the unrestraint power of the chip while having zero thermal restrictions.

It's still operating within spec, pre-set voltages and such.

So, there was no 'jailbreaking' and OCing. Just, pure auto-clocking based on thermal/electrical constraints?

Makes sense; I've seen several folks around the web water cool or Peltier cool their phones, and make huge gains (w/o touching a single setting/clock, unrooted, otherwise stock)

AFAIK, any 'automatic dynamic clocking' CPU/APU gets 'free performance' the cooler you can run it...
NtM, Esp. w/ sub-zero cooling, the stock voltage profile (effectively) becomes an overvolt, given the extreme low temps.

 

[Vayra86]

Of course not, but they will actually have active cooling. The LN2 isn’t overclocking the chip. It’s keeping it from thermally throttling. It’s also not likely capable of drawing excessive power like a desktop SOC, since it’s in an iPad.

I suppose if they could take the iPad apart and put a traditional air cooler on it we could find out, but I don’t know how easy that would be. Besides, I didn’t claim it was how a desktop M4 would perform, but it’s probably not that far off either. A desktop M4 will probably be given more freedom with power consumption, not just thermals.

Yeah thanks for letting my penny drop. Bit rusty in there I think. They really did just add LN2 and no extra power did they? That's pretty impressive

Are the M4 Macs going to come with LN2?

Would be a hilarious april fools. Apple sending their M4 customers a small bottle of it.

 

[LabRat 891]

Yeah thanks for letting my penny drop. Bit rusty in there I think. They really did just add LN2 and no extra power did they? That's pretty impressive


Would be a hilarious april fools. Apple sending their M4 customers a small bottle of it.

Compound W Freeze Off® Wart Removal

Compound W Freeze Off® products use the same cryogenic method that doctors use to remove warts in as little as one application.

www.compoundw.com

Just include one of these, ya know for those rare extreme workloads

 

[Darmok N Jalad]

Yeah thanks for letting my penny drop. Bit rusty in there I think. They really did just add LN2 and no extra power did they? That's pretty impressive


Would be a hilarious april fools. Apple sending their M4 customers a small bottle of it.

I don’t see how they could do anything else. To my knowledge, there are no tools to overclock Macs and so it can’t be done. I’ve never heard of someone overclocking one, not even from bootcamp during the x86 days.

One could easily reproduce this test with a modern x86 mobile CPU that has a similar behavior of thermal throttling. I bet the results would be similar, where the chip doesn’t reduce clocks because it’s not hitting a thermal limit. Battery life would obviously suffer, but it’s not really an issue since cooling is the weakest link in this scenario.

FWIW, if I put my passively cooled M2 MBA on an ice pack, I only get about 5% more performance in a CB run. The SOC still pegs 100C, so it’s just not sufficient cooling. LN2 is obviously going to overcome that issue.

 

[LabRat 891]

I don’t see how they could do anything else. To my knowledge, there are no tools to overclock Macs and so it can’t be done. I’ve never heard of someone overclocking one, not even from bootcamp during the x86 days.

One could easily reproduce this test with a modern x86 mobile CPU that has a similar behavior of thermal throttling. I bet the results would be similar, where the chip doesn’t reduce clocks because it’s not hitting a thermal limit. Battery life would obviously suffer, but it’s not really an issue since cooling is the weakest link in this scenario.

FWIW, if I put my passively cooled M2 MBA on an ice pack, I only get about 5% more performance in a CB run. The SOC still pegs 100C, so it’s just not sufficient cooling. LN2 is obviously going to overcome that issue.

Cascade Phase Change cooled Tablet Dock, when?

[I can only imagine the condensation issues...]

 

[Darmok N Jalad]

Yeah, I honestly don’t know what we’d see with modern CPUs on how they’d perform under “stock” settings with thermal constraints removed by LN2. It’s basically in the design of some of these chips to push clocks as long as power and temps allow. Usually when people use LN2, it’s to overclock. I suspect Intel chips would see a decent boost, while AMD not as much? It would make for an interesting experiment.

 

[friocasa]

So they're using liquid nitrogen outside the case, no tweaking of the frequency of voltages

It just shows that in fact when the newer chip it's not burning itself down, can archive better performance than the predecessors in ST, who would have thought...

Probably the MacBook Air will have the same issue if uses passive cooling, and those who want all the performance on the go will have to spend the extra on the MacBook Pro

What a shitty company, always proclaiming to be green and eco friendly while shortening the lifespan and performance of their devices and making repairs and upgrades almost impossible