> It's worth pointing out how extremely far ahead Apple seems to be in terms of CPU power and efficiency.
Honestly, I'm not convinced they're THAT far ahead. They don't have a lot of the legacy baggage Intel has to contend with, and they're the only company making high end ARM chips (besides Amazon and a few other weird server implementations), but being able to match big core i7s in some benchmarks single threaded is to a large extent something that Intel's own low power chips can also do, at least burstily.
There are a lot of challenges to big many-cored chips beyond single-core performance, and we really don't know where they are with that yet, as there are no publicly-available examples of Apple desktop chips.
> and they're the only company making high end ARM chips
That's, if you'll pardon the pun, an Apples to Oranges comparison. Apple isn't making "high end ARM chips" either if your comparator is powerful servers. You need to look at what Apple is doing within their power envelope and compare that to what everyone else is doing within _their_ power envelopes. The A13 Bionic is an uncooled 6W TDP chip blowing past 95W base TDP chips that require hefty active cooling.
To add to that, if you look up the benchmarks of the i7 8500Y, which is intel’s top shipping 5W offering, you see it is vastly slower than the A13. A third slower in single core, less than half the performance in multicore.
This whole thread reminds me of how passionate the power pc enthousiasts were defending it as superior, right before apple switched to intel and doubled mac performance overnight.
> This whole thread reminds me of how passionate the power pc enthousiasts were defending it as superior, right before apple switched to intel and doubled mac performance overnight.
To be fair, the PowerPCs _were_ measurably better and faster when each one was released. Apple just couldn't get a G5 CPU that would fit into a laptop, and IBM was an unreliable partner with a slow release cycle, so by the time the transition happened they had fallen behind.
The PowerPC G3 and G4 were great chips. The G3 was much better than competing x86 chips, and G4 beat contemporary ones consistently too.
The G5 wasn't great though. When Steve Jobs announced it, Apple already showed it only trading blows with the then-current Pentium 4 (a Pentium 4 - they sucked!). And a few months after the first Power Mac G5s were launched, they were already resoundly beaten by the new Athlon 64s [1].
Add to that the G5 was basically a POWER4 server chip, and IBM was only building server chips in the future, Apple basically had no choice. Nothing really to do with PowerPC vs x86, but more to do with what kind of processors their suppliers were willing to build.
That was probably the last PPC they did and was probably ready when Apple made the announcement. It just wasn't worth for IBM to invest in workstation-class PowerPC chips with laptop power envelopes. The only other use for PowerPCs, from their PoV, was their own workstations and those could use the higher end POWER chips.
IBM continued making PowerPC G3 derivatives for a while for Nintendo. Nintendo ended Wii U production in January 2017, which would I guess would mark the end of IBM's production of PowerPC as well.
The Wii U for all its flaws is probably the most "practical" Power-based machine you can get nowadays, given relative power, availability, size and price. A 1.2GHz triple core PowerPC G3 would probably still eek out Raspberry Pi 3 like performance. Shame the Linux port to it never really got off the ground (also partially due to IBM's hackjob of an SMP implementation for the G3).
>There are a lot of challenges to big many-cored chips beyond single-core performance, and we really don't know where they are with that yet, as there are no publicly-available examples of Apple desktop chips.
The A12X from 2018's iPad Pro is the sort of chip I'd expect to see in an ARM laptop, and its multicore scores are similar to the top end of 2018 Macbook Pros.
2020's iPad revision didn't get much in the way of processor improvements (just one more GPU core), so the new 16" MBP has pulled ahead with an 8-core i9, but when we get a new iPad based on an A13X or A14X I expect it to be back in that range again.
And these are in thin fanless tablets. With a proper cooling system, there's got to be some extra juice to be squeezed out of them.
I wonder if anyone's ever done A-series chip "overclocking", or at least manual overvolting with active cooling. It'd be cool to see what kind of performance increase might be available.
> Being able to match big core i7s in some benchmarks single threaded is to a large extent something that Intel's own low power chips can also do, at least burstily.
You're not wrong. In general, it's true that a microbenchmark amplifies the Apple A13's strengths due to power limits. The assumption I make is that microbenchmarks indicate the true peak performance of Apple's architecture, and as power limits become a smaller constraint when Apple uses its chips in laptops and desktops they will make available that performance in a more sustained way.
But even low power Intels don't compare that favourably. Intel's new i7 10510U delivers very nice single core performance [1]. But it's worth noting that 1) that still does not quite match the A13's burst performance 2) that chip is still rated for a power profile much larger than the A13's and 3) as always in these discussions - Intel's "TDP" is a marketing term not a power limit. At high turbos the chip is permitted to consume quite a bit more power than the 15W it's rated for.
This particular Intel chip boosts to 4.90GHz. For Apple chips, even stuff like clockspeeds are a matter of conjecture, but Wikichip without a source claims that the A13 tops out at 2.65GHz [2] which if true indicates a lot more thermal and frequency headroom in bigger form factors.
I just benchmarked my MacBook Pro+Safari in Jetstream 2.0 [3] - not a microbenchmark - and it scored nearly 145 compared to the nearly 130 the iPhone 11 scores [4]. That's with a "45W TDP" Core i7 8850H topping out at 4.3GHz. It's hard to benchmark iPhones well, but all evidence points to the fact that they are actually really fast.
* 6 year old Macbook Pro i7-4980HQ, Windows 10 - 102
* 5 year old Macbook Pro i7-5557U, OS X - 100
* Threadripper 2990WX desktop - 99
So, uh, I might have some questions about this benchmark's general validity now?! - though maybe it is some evidence in favour of my vague feeling that the Threadripper sometimes doesn't feel as fast as it seems like it ought to feel.
The Jetstream benchmark is made by the WebKit team. Its scores will vary per browser, so you have to compare browser to browser. I compared an iPhone 11 running Safari to a MacBook Pro running Safari. Apples to apples.
Moreover, while the Threadripper 2990WX is a really awesome processor, single core benchmarks (I think Jetstream is mostly limited to a single thread) aren't particularly its strength. Over multiple runs it should beat your Macbook Pro, but not by a huge amount. If not, take a look at how you're cooling that beast :)
I updated a six core i7 3930k with a sixteen core 1950x threadripper on one of my boxes. Biggest reason for the update was the disk IO (m2 drives) and number of cores and I'd had the 3930k since the launch day at Microcenter. On Windows, single threaded at stock speeds, the cores were comparable from a 6-8 year gap between the two CPUs. For hosting virtual machines... the speed did not matter as much as having an abundance of physical cores. Still - I was not expecting the 'core speed' as reported by a video game to be as close as it was.
The Zen2 (3900x) core speed on the other workstation reported as almost twice as fast, with 12 cores. Really wish that TR4 board supported the 39xx threadripper series.
Zen 2 was a big uplift in single-threaded performance.
Ryzen and Threadripper 1000- and 2000-series, and Ryzen Mobile < 4000-series are all on Zen or Zen+ architecture.
The current gen Ryzen and Threadripper 3000-series and the Ryzen Mobile 4000-series are the ones running on Zen 2. This is where AMD is competitive with Intel on single-threaded workloads, largely across the board.
It's the opposite. The latest generation of ThreadRipper is not only ahead in multicore peformance it is also comparable to the highest possible single core performance from the regular Ryzen Lineup.
>besides Amazon and a few other weird server implementations
One of those 'few other' was Scaleway, but they recently buckled up and ended their ARM server lineup abruptly[1]. They were running Marvell ThunderX SoCs(Upto 64 cores, 128GB RAM).
So, Amazon might soon takeover ARM server market i.e at-least till Tim Cook does a Satya Nadella and brings in Apple IaaS with ARM CPUs.
> and they're the only company making high end ARM chips
I'm typing this with a Surface Pro X running a ARM64 CPU called SQ1 which a customization of a Snapdragon 8cx. It is quite high end and is not made by Apple. It might not be the amazing custom CPUs they have on iOS devices but it is still a pretty good CPU.
I mean, the Surface Pro X chip is fine, but it would be hard to call it high end; it significantly lags the usual intel chips used in tablets and small laptops on performance, especially single core. The newest Apple ones are competitive with those or beat them.
When your cheapest smartphone is faster than the fastest competitor smartphone, I’d say you are far ahead. But your point on how well that will apply in desktop PCs is certainly valid.
Honestly, I'm not convinced they're THAT far ahead. They don't have a lot of the legacy baggage Intel has to contend with, and they're the only company making high end ARM chips (besides Amazon and a few other weird server implementations), but being able to match big core i7s in some benchmarks single threaded is to a large extent something that Intel's own low power chips can also do, at least burstily.
There are a lot of challenges to big many-cored chips beyond single-core performance, and we really don't know where they are with that yet, as there are no publicly-available examples of Apple desktop chips.