iPad Pro M4 Geekbench scores wow…

[uller6]

And transistor density has nothing to do with performance? Obviously that's not the only purpose of denser chips: you can make smaller chips, etc. But, as David House said, transistor count may actually drive performance faster than predicted by raw transistors. See from Wikipedia:

The doubling period is often misquoted as 18 months because of a separate prediction by Moore's colleague, Intel executive David House. In 1975, House noted that Moore's revised law of doubling transistor count every 2 years in turn implied that computer chip performance would roughly double every 18 months (with no increase in power consumption).​

​

This was later formalized as Koomey's Law.

You’re forgetting the Dennard scaling limit - which we hit 15 years ago - that leads to modern chips having ~50% dark silicon. More density does not mean faster, since we’re now heavily power consumption (in reality power dissipation) limited. If every transistor operated at once modern chips would melt.

 

[Ghost31]

Disappointing. Was hoping it was gonna be a 10 minute compilation or something.

 

[Boing123]

Hopefully WWDC has some good surprises, and hopefully Apple doesn't artificially nerf the M1/M2 iPads just to upsell..

They have never done this. Come up with facts before making baseless statements like these.

If you are referring to the battery power limit for old batteries: Apple limited the power draw of devices with older batteries, much like electric vehicles limit max performance if a battery is near dead or indeed older. There is nothing nefarious about that: replacing the battery will restore performance or you could choose to just live with it: what good is an iPhone with a 70% capacity and max power if it goes dead in 30m?

For all this complaining ‘modern iPads are OP for multiple generations now’, it is surprising how those same people will moan about supposed ‘slowing down’. 80% of completely overpowered is probably still good enough…

Or put differently, if Apple would have wanted to maliciously kneecap devices to get you to upgrade, they wouldn’t have advertised in their UI.

 

[Boing123]

I'm usually not a naysayer, but Apple did go very hard against Intel for not producing the improvements they wanted and it does seem like Apple isn't providing much better, though bringing it first party has other advantages surely, and third party benchmarks are probably not very high on the list of priorities for the chip team. That said, a 22.83% improvement YOY for single core performance is not in line with Moore's Law, which is ~41%. The fact that they haven't even doubled performance yet on the M line at all is pretty disappointing.

On GeekBench 6, the oldest A series scores are for A8; between the A8 and the A10, the single core performance increased 76% (~430 -> ~760). A11 is 1100, significantly above double.

Funny: everyone, including all the famed YouTube ‘tech reviewers’ who seem to think video editing is the only workload imaginable, claim “the iPads didn’t need more power” and yet at the same time people complain the generational speed gains of the chips aren’t good enough.

Ever stopped to consider Apple might be adding application specific circuitry (e.g. to benefit AI applications) to these chips instead of focusing on probably unnecessary general computing speed increases?

 

[Boing123]

How do you know what tasks they run?

It’s a general fact in computing: most games are e.g single threaded. It is why for the longest time Intel i9 was the God killing of gaming chips even though multi core chipset design AMD CPU’s were vastly superior in other areas.

 

[JustAnExpat]

It’s a general fact in computing: most games are e.g single threaded. It is why for the longest time Intel i9 was the God killing of gaming chips even though multi core chipset design AMD CPU’s were vastly superior in other areas.

How do you know he plays games?

 

[Boing123]

How do you know he plays games?

I do not, that was not the point.

What is your point?

 

[krell100]

After all those years in the computing wasteland Mac users finally get the hardware we always wanted. I get the feeling an M4 Max or Ultra will be humming along in my recording studio before too long..

 

[nathansz]

It’s a general fact in computing: most games are e.g single threaded. It is why for the longest time Intel i9 was the God killing of gaming chips even though multi core chipset design AMD CPU’s were vastly superior in other areas.

Why do you assume they just want to play games?

 

[henrikhelmers]

It’s a general fact in computing: most games are e.g single threaded.

This was true at some point, but now most utilize multiple threads. And writing apps for Apple platforms it is encouraged and easy to move processing off the main (UI) thread.

 

[Boing123]

This was true at some point, but now most utilize multiple threads. And writing apps for Apple platforms it is encouraged and easy to move processing off the main (UI) thread.

Offloading some asset decompression to a small thread vs having a truly multi threaded game loop instead, not leaning heavily one one specific thread … I would bet there is still miles to go before all games are truly multi threaded. It requires knowledge to implement properly.

Should be easy enough to check as well: run any game and track the load on your cores.

.NET gui apps by default run everything on the UI thread as will most other runtimes.

Electron/node based apps, most python (GIL anyone? ), … very likely all still single threaded.

Also, concurrency/async is not the same as multithreaded but is probably more common.

 

[henrikhelmers]

Offloading some asset decompression to a small thread vs having a truly multi threaded game loop instead, not leaning heavily one one specific thread … I would bet there is still miles to go before all games are truly multi threaded. It requires knowledge to implement properly.

Should be easy enough to check as well: run any game and track the load on your cores.

You are assuming too much. Sure, each application will have a main thread, but that does not mean that it is possible (and common) to distribute the workloads. Here's a random sample from a game. While it only has a single subprocess (at this point it time) there are 41 threads, and they are using the full extent of 3.4 CPU cores.

 

[jchurchill]

Anyway, the CPU is wasted in the iPad... like putting a V8 on a lawn mower...

Bad analogy. A lawnmower cuts grass. iPadOS doesn't do anything useful.

 

[EugW]

iPad16,4 - Geekbench

Benchmark results for an iPad16,4 with an ARM processor.

browser.geekbench.com

Given how close the fanless iPad Pro is to 4000 single-core, I wonder if the M4 MacBook Pro will get there.

 

[Iskee]

First binned (9-core) iPad Pro GB6 benchmark has finally posted: https://browser.geekbench.com/v6/cpu/6062510

Single: 3630
Multi: 13060 (+~30% over M2, + ~8-9% over M3)

Single core came out a little on the low end compared to some of the other results we've seen, but we only have this one sample to go off of, so who knows if it's going to be a pattern. Probably will be close to the unbinned version.

I'm not planning to get this iPad because I just bought an M2 last year, but anyone ready to upgrade their iPad shouldn't be concerned about the performance in the binned vs unbinned version, IMO.

 

[opuscroakus]

good bless the Celeron 300A.....overclock that to 450mhz by moving a jumper. 50% MHz gains!

Don’t forget the old AMD Duron and Thunderbird cpu’s that you could unlock and overclock just by “drawing” on top of the chip with a lead pencil to connect the L1 bridges. Those were fun times, back when even small gains seemed monumental.

 

[Confused-User]

Given how close the fanless iPad Pro is to 4000 single-core, I wonder if the M4 MacBook Pro will get there.

Possibly the 16" Max. The rest, probably not, but it doesn't seem impossible.

I think the Studio will score comfortably over 4k.

 

[EugW]

First binned (9-core) iPad Pro GB6 benchmark has finally posted: https://browser.geekbench.com/v6/cpu/6062510

Single: 3630
Multi: 13060 (+~30% over M2, + ~8-9% over M3)

Single core came out a little on the low end compared to some of the other results we've seen, but we only have this one sample to go off of, so who knows if it's going to be a pattern. Probably will be close to the unbinned version.

I'm not planning to get this iPad because I just bought an M2 last year, but anyone ready to upgrade their iPad shouldn't be concerned about the performance in the binned vs unbinned version, IMO.

I compared that M4 9-core bench to M3 in a table.

As you say though, that particular M4 single-core score is lower than average in this one-and-only bench, and the M3 bench I chose is higher than average.

 

[gsal]

First binned (9-core) iPad Pro GB6 benchmark has finally posted: https://browser.geekbench.com/v6/cpu/6062510

Single: 3630
Multi: 13060 (+~30% over M2, + ~8-9% over M3)

Single core came out a little on the low end compared to some of the other results we've seen, but we only have this one sample to go off of, so who knows if it's going to be a pattern. Probably will be close to the unbinned version.

I'm not planning to get this iPad because I just bought an M2 last year, but anyone ready to upgrade their iPad shouldn't be concerned about the performance in the binned vs unbinned version, IMO.

That multi-core score is higher than I anticipated. In that case I will be sticking with my 256GB order. I will happily pocket the £600 difference.

 

[Confused-User]

I compared that M4 9-core bench to M3 in a table.

Nice but the table actually says "M4 10-core" in two places!

 

[Confused-User]

That multi-core score is higher than I anticipated. In that case I will be sticking with my 256GB order. I will happily pocket the £600 difference.

Not to discourage you from your decision - I would probably do the same if I were buying it - but that's only half the picture. GB5 MC or some other bench where subtasks are less interdependent is also of interest.

Actually, only you know what's of interest *to you*. Which MC bench matters to you depends entirely on what loads you'll be throwing at it.

 

[ipaddaro]

Nice but the table actually says "M4 10-core" in two places!

sorry I thought you were referring to GB results

 

[EugW]

Nice but the table actually says "M4 10-core" in two places!

Thanks, I've fixed the column label.

 

[ipaddaro]

That multi-core score is higher than I anticipated. In that case I will be sticking with my 256GB order. I will happily pocket the £600 difference.

also I was thinking it would have been 75% of 10 cores version ‘cause it has 3 out of 4 performance cores… probably also the efficiency cores give their contribution during this tests… don’t know. If confirmed, it would be actually impressive!!!

 

[EugW]

also I was thinking it would have been 75% of 10 cores version ‘cause it has 3 out of 4 performance cores… probably also the efficiency cores give their contribution during this tests… don’t know. If confirmed, it would be actually impressive!!!

If workloads can use all cores effectively, then the efficiency cores contribute a significant boost. And the M4 has 50% more efficiency cores than M3. ie. M4 loses one performance core, but gains two efficiency cores. Benchmarks like this are designed to utilize all cores effectively.

However, these benchmarks may not always translate to real world applications, because some applications may not benefit as much as others from so many extra cores.