News Liquid nitrogen-cooled M4 iPad Pro flaunts remarkable single-core performance gains — M4 outperforms M3 Max and M2 Ultra

Toggle sidebar Toggle sidebar
It's on the very most advanced node in the world at this time, so how is it surprising that it beats Intel and AMD on single-core performance? Also, kind of moot having a chip like this in an iPad; honestly, the M1 runs cool compared to its successors and wouldn't perform noticeably slower for most users -- even "Pro" users.
 
  • Like
Reactions: artk2219
It's important to emphasize that Geekerwan's M4 iPad Pro wasn't utilizing the top 4+6 configuration. It'll be interesting to see how big of a difference the extra performance core can make. However, we wouldn't hold our breath because Apple's iPadOS is still holding back the silicon, so we likely won't fully grasp the M4's performance until it makes its way into a MacBook Pro or Mac Studio with proper cooling and macOS.
Click to expand...
This seems like a silly statement.

He's cooling it with liquid nitrogen, I don't think the fans in a MacBook Pro or Mac Mini/Studio are going to top that.

Clearly what will top that is more cores.
 
  • Like
Reactions: artk2219
The multi core score isn't all that surprising. The M4 used was the version with only 3 P core variant, so it's starting off in a defect to begin with. The multi-core score is about that of the 4 P core version of the M4, not surprising considering the single core score was only about 17.6% higher than the standard M4 single core score . It's also not surprising it loses in multi core score vs the 6 P core M3 Pro or the 12 P core M3 Max.
 
  • Like
Reactions: Stately, bit_user and artk2219
hotaru251 said:
i mean this is a pretty "meh" article topic.

apples always known to ignore cooling (and thus hamper own performance) as to them its fashion before function.
Click to expand...
Thats not the point. The cooling only gives it a few percentage points improvement. The point is that a tablet is outclassing cutting edge desktop CPUs.

"But its an expensive tablet" This variant of the chip starts at $1,000, while the desktop i9 14th gen that it handily beats sells for almost $700.
 
  • Like
Reactions: Stately
kealii123 said:
The point is that a tablet is outclassing cutting edge desktop CPUs.
Click to expand...
again not a shock.
Modern ARM CPU beating x86 CPU that have chosen to support stuff from 30yrs ago?

AMD & others plan to have ARM cpu by like 2025 or 2026 and then we'll see how it compares.


and thats ignoring fact Desktop have basically given up high singlecore advancement as its main focus for multithread. (and thats why near everything anymore can use multiple cores unlike back in the old days where it was hard to find stuff that did as everything was single threaded back then)

Is the M4 chip impressive? yes.
is it shocking? not really. (again ARM and RISCV will replace x86 and few ppl doubt that)
 
I really wish more people would use SPEC2017, rather (or in addition to) GeekBench. GB is closed-source and we really don't know how well-optimized it is for different platforms, making the business of using it to compare between different ISAs somewhat fraught.

What's interesting about the LN2 experiment is that it gives a clue into how well the M4 will perform in a chassis with better cooling, such as laptops and especially desktops. I doubt you really need LN2 for the cores to near their max clocks. Rather, the main impediment should be the tablet chassis.

funguseater said:
This just in! Overclocked CPU outperforms Stock CPU!!! Halt the Press'
Click to expand...
Nowhere in the article does it say it's overclocked! In order to accomplish that, you'd probably have to hack iOS, which I doubt he did.

I think all this exercise accomplishes is simply letting the CPU reach & sustain its normal max boost clocks.

As a matter of fact, the article states the benchmark ran at 4.41 GHz, which indeed is what a little "web research" shows is the CPU's normal peak clock speed!
 
Last edited:
  • Like
Reactions: The Historical Fidelity, Stately, slightnitpick and 2 others
hotaru251 said:
Is the M4 chip impressive? yes.
is it shocking? not really. (again ARM and RISCV will replace x86 and few ppl doubt that)
Click to expand...
I'd say it's pretty shocking, for a single-generation improvement of Apple's own cores.

Also, maybe try comparing it to the fastest non-Apple ARM core out there? If you exclude the unlaunched Qualcomm Oryon cores, I think you'll find an even bigger gap.

So, this isn't just a story of ISAs. Apparently, whatever cores they have in the M4 (or the P-cores, at least) is really a big leap, for a company that already lead the pack!
 
  • Like
Reactions: Stately and JamesJones44
DS426 said:
It's on the very most advanced node in the world at this time, so how is it surprising that it beats Intel and AMD on single-core performance?
Click to expand...
While the node advantage is real, it's certainly not everything. TSMC claims their N3B node is only 10% to 15% faster than N5 at iso-power:

wikichip_tsmc_logic_node_q2_2022-2.png

Source: https://fuse.wikichip.org/news/7375/tsmc-n3-and-challenges-ahead/
​

...and Zen 4 is made on TSMC N5.
 
  • Like
Reactions: The Historical Fidelity, Stately and JamesJones44
I would rather see real world performance than to rely on a single benchmark. It is true that the chip is faster, but I doubt many will think it is significantly faster in their day to day usage.
 
funguseater said:
This just in! Overclocked CPU outperforms Stock CPU!!! Halt the Press'
Click to expand...
You can't overclock this CPU, all they did was eliminate thermal throttling. Someone did a similar water block replacement on a M1 Ultra Mac Studio. Bench scores were the same, (the heatsink of the ultra has 1kg of copper in it) Anything beyond the cooling needed to prevent throttling wont make it perform any better.
 
  • Like
Reactions: Stately, bit_user and slightnitpick
kealii123 said:
Thats not the point. The cooling only gives it a few percentage points improvement. The point is that a tablet is outclassing cutting edge desktop CPUs.

"But its an expensive tablet" This variant of the chip starts at $1,000, while the desktop i9 14th gen that it handily beats sells for almost $700.
Click to expand...
It "handily" beats 1/24th of the 14900k, since we are talking about a single threaded benchmark.

In the MT part it scored 13k. My 12900k scores 50% higher....
 
TheHerald said:
It "handily" beats 1/24th of the 14900k, since we are talking about a single threaded benchmark.

In the MT part it scored 13k. My 12900k scores 50% higher....
Click to expand...
50% higher than a fanless tablet, in the lower core count (budget) variant against a desktop CPU, and you're congratulating yourself?
 
  • Like
Reactions: bit_user
kealii123 said:
50% higher than a fanless tablet, in the lower core count (budget) variant against a desktop CPU, and you're congratulating yourself?
Click to expand...
I'm not congratulating anything, you said the M4 handily beats a 14900k when that can't be any further from the truth.

Being fanless or a tablet is irrelevant. I already posted how much power the 12900k draws to complete the GB6 run, 19.1 watts on average. The fans were off so I was running fanless too.

The m4 (that tablet chip) has 2.5 times the transistors of the 12900k. It should be actually decimating it, yet the 12900k is 50% faster.

Mind you even the MT portion of Geekbench 6 doesn't really scale with cores, that's why the power draw was so low to begin with. In a proper MT workload I don't think the base M4 gets remotely close
 
TheHerald said:
The m4 (that tablet chip) has 2.5 times the transistors of the 12900k. It should be actually decimating it, yet the 12900k is 50% faster.
Click to expand...
This is a ridiculous point. The M-series iGPUs are bigger/better and the M4 also has a dedicated NPU. Plus, it's a full SoC, which your i9-12900K is not.

You should compare transistor count in the cores, if you're comparing anything of the sort.
 
bit_user said:
This is a ridiculous point. The M-series iGPUs are bigger/better and the M4 also has a dedicated NPU. Plus, it's a full SoC, which your i9-12900K is not.

You should compare transistor count in the cores, if you're comparing anything of the sort.
Click to expand...
Until we have a die shot of the m4 we can't do that though. Still, regardless of all that, saying that the m4 easily beats the 14900k when it gets actually blasted by a 3 year old 12900k is just ridiculous. If it's losing in a workload like GB6 which doesn't come nowhere near stressing the chip (hence the 19.1 w power draw recorded on my 12900k) what's gonna happen in a fully MT workload?
 
bit_user said:
You should compare transistor count in the cores, if you're comparing anything of the sort.
Click to expand...
Putting things into perspective, I just disabled 4P and 4 ecores, so I'm running 4+4, with the ST boost completely disabled.

So a 3 year old Intel 4+4 chip, passively cooled matches the m4 under LN2 in MT performance in GB6. 😱😱

https://browser.geekbench.com/v6/cpu/6156635

But of course the m4 smashes the 14900k, easily, lol.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom