Apple should hire back Nuvia employees

[albertserene]

Ever since Apple lost its semiconductor employees to form Nuvia, the development process of Apple silicon has come to the grinding stop. The M2 is only 18% quicker than M1 after almost 2 years gap. The A16 is virtually the same as A15. It seems Apple is unable to further improve upon its ARM architecture. With the lawsuit between Qualcomm and ARM, Apple should take this opportunity to hire back those employees. Or else, it might just have to become a licensee of ARM's reference design.

 

[JPack]

Apple relies on TSMC. If litho stalls, performance stalls. Notice A16 only has 7% more transistors compared to A15. While A15 has 27% more than A14.

 

[albertserene]

Chipset density is only half of the story. The fact is that Apple has not been able to come out with new ARM core design after the Firestorm. That's why there is hardly any performance improvement. The lack of talent to move forward is what I worry about. Intel's Raptor Lake has already exceed the M1 Ultra performance. Apple need to move ahead.

 

[Bodhitree]

I think a lot of talented engineers are tied up making the chipsets for the new Mac Pro. If you look at the rumours that is going to be a beast, and it requires different engineering for the outward facing / upgradeable parts. The effort of making the transition to Apple Silicon across the entire line-up shouldn’t be underestimated.

Thats why you’re temporarily seeing less focus on the A-series cores, and less of a step from M1 to M2. Once the new Mac Pro has finished testing, those engineers will be released back onto other projects and we will see whether Apple has managed to retain the important talent.

 

[leman]

The only lack of change since A14 is in the performance core backend (which is also why CPU IPC didn’t change). A15 and A16 had major changes to the small CPU cores, dramatically improving their performance. They also had big changes to the AMX units, the neural engine and some other accelerator engines. Not to mention extensive changes to the cache subsystems.

As to the lack of radical changes in the main CPU and the GPU… there are basically two possible explanations. One is what OP and some others suggest, that Apple has lost the edge and is stagnating. Another one is that there are no updates because they are working on a big next-gen versions of these processors. Given that we are only two years after A14, that TSMC is experiencing some delays in deploying 3nm I personally tend to interpretation two. I think a new very fast desktop CPU core is incoming, together with a re-engineered GPU that has hardware raytracing suppport.

 

[albertserene]

I thank all's responses. But I tend to think it is due to loss of talent that Apple is stagnating. Otherwise, why will ARM stop licensing to Nuvia and fire a lawsuit against Qualcomm? They are afraid that Qualcomm will stop licensing ARM's reference design if the Nuvia's cores are adapted. That means a big loss of revenue for ARM. It seems the talent that carries Apple way above the competition in the previous years is no longer there. I hope I am wrong because I am a big Apple fan.

 

[playtech1]

I think it's a little early to say if M2 indicates stagnation, but I do see things like chiplets, V-Cache and tiles on the AMD/Intel side and hope that Apple is developing it's own secret sauce to keep up over the long term.

It would be disastrous if Apple Silicon Macs were to bump up against a performance ceiling after only a couple of generations.

 

[leman]

I thank all's responses. But I tend to think it is due to loss of talent that Apple is stagnating. Otherwise, why will ARM stop licensing to Nuvia and fire a lawsuit against Qualcomm? They are afraid that Qualcomm will stop licensing ARM's reference design if the Nuvia's cores are adapted. That means a big loss of revenue for ARM. It seems the talent that carries Apple way above the competition in the previous years is no longer there. I hope I am wrong because I am a big Apple fan.

What does any of this have to do with ARM and their lawsuit?

 

[maflynn]

I'm not a chip designer, nor do I have any technical knowledge regarding chip design, so my opinion and thoughts are based on what little I've read and my personal observation.

Interesting topic and I've struggled to put my thoughts to paper. I don't know if we can say for sure if there's stagnation occurring just yet - its too early. But one thought I had is that if the only performance gains being realized is due to improved fabrication, and there's little to no design changes, then yes that's not a good sign. Basically, if
the only major change between the M1 Pro/Max and M2 Pro/Max is the fabrication, then the case for lack of innovation will be easier to make.

 

[albertserene]

What does any of this have to do with ARM and their lawsuit?

Because even ARM recognize the talents of Nuvia is a major threat to its reference design. The talents were from Apple.

 

[JPack]

Chipset density is only half of the story. The fact is that Apple has not been able to come out with new ARM core design after the Firestorm. That's why there is hardly any performance improvement. The lack of talent to move forward is what I worry about. Intel's Raptor Lake has already exceed the M1 Ultra performance. Apple need to move ahead.

It’s a chicken or the egg situation because without an increased transistor budget, you cannot increase performance. We saw that with Skylake through Comet Lake. New core designs need more transistors.

 

[bruinsrme]

The processor is one thing. What about Ram and the buses. If the other supporting subsystems do not match the speed of any brand processor, would that create a bottleneck thereby limiting the effective performance of newer processors?

 

[JPack]

The processor is one thing. What about Ram and the buses. If the other supporting subsystems do not match the speed of any brand processor, would that create a bottleneck thereby limiting the effective performance of newer processors?

M2 and A16 have 50% more memory bandwidth. The dataset for most benchmarking tools also fits within the cache.

 

[Pressure]

It's weird with all these newly created accounts that suddenly have an opinion on who Apple should hire or whatnot for their SoC development.

18% increase in CPU performance is great. Have in mind that there is a lot more to the A16 Bionic SoC than just processor speed. Everything else got a big increase as well.

No one else offers anything close to that performance in the phone form factor.

 

[Ethosik]

People do realize we had global shutdowns, chip shortage, Russia and other bigger world issues going on right? If it wasn’t for those, I would think M2 would have been better and we would already have the Mac Pro.

 

[maflynn]

It's weird with all these newly created accounts that suddenly have an opinion on who Apple should hire or whatnot for their SoC development.

There's only one new member in this thread and he/she is active, providing their opinion. I think their opinion is just as valuable as anyone elses ¯\_(ツ)_/¯

 

[leman]

Interesting topic and I've struggled to put my thoughts to paper. I don't know if we can say for sure if there's stagnation occurring just yet - its too early. But one thought I had is that if the only performance gains being realized is due to improved fabrication, and there's little to no design changes, then yes that's not a good sign. Basically, if
the only major change between the M1 Pro/Max and M2 Pro/Max is the fabrication, then the case for lack of innovation will be easier to make.

My opinion exactly. The 2023 will be a deciding year. If Apple does not come out with a new, significantly better microarchitecture this year then I’ll agree that they are in trouble.

Because even ARM recognize the talents of Nuvia is a major threat to its reference design. The talents were from Apple.

Do they? Or could it be the case that ARM might believe that architectural license transfer is not legal under these circumstances? In other words, why do you interpret this as ARM being worried about Nuvia designs rather then them simply believing that Qualcomm oves them money?

 

[Ethosik]

My opinion exactly. The 2023 will be a deciding year. If Apple does not come out with a new, significantly better microarchitecture this year then I’ll agree that they are in trouble.

I don’t have high hopes for 2023. I think it will take another year to recover from all the world events. Earliest I expect the Mac Pro now would be December 2023. I do not believe it will be out sooner and IMO December is pushing it.

I’m looking forward to finally replacing my 2010 Mac Pro. But we are approaching a recession so the later the new Mac Pro arrives the happier I’ll be.

 

[GMShadow]

Chipset density is only half of the story. The fact is that Apple has not been able to come out with new ARM core design after the Firestorm. That's why there is hardly any performance improvement. The lack of talent to move forward is what I worry about. Intel's Raptor Lake has already exceed the M1 Ultra performance. Apple need to move ahead.

A16 has revised designs for both performance and efficiency cores. They just couldn't push them super hard due to the node stall.

And as someone else noted, it still got 18% faster, which is more than Intel typically manages even over two generations.

 

[leman]

I don’t have high hopes for 2023.

I do I don’t know about Mac Pro, there are a lot of challenges, but I am half optimistic that we will get a new generation of CPU cores that’s 30% faster than now and a redesigned GPU with hardware raytracing.

 

[Kimmo]

Nuvia, Shmuvia!

When you get a look at my new Apple Silicon Mac Pro, with twice the performance at half the cost, you're going to wonder why you ever doubted me!

 

[theorist9]

It’s a chicken or the egg situation because without an increased transistor budget, you cannot increase performance. We saw that with Skylake through Comet Lake. New core designs need more transistors.

Yes, a larger transistor budget certainly helps. Some have speculated that 3 nm will be needed in order to add hardware RT to AS and still have the chip come in under the reticle limit (I believe that's 26 mm x 33 mm = 858 mm² for current i193 and EUV lithography steppers).

However, performance can be significantly improved without an increase in the transistor budget, by improving the microarchitecture. That was the basis of Intel's tick-tock production strategy: A process shrink (tick), followed by the introduction of a new microarchitecture on that same process (tock).

For example Intel's tock upgrade from Penryn to Nehalem on 45 nm gave a 20%-30% performance boost:

The Nehalem Preview: Intel Does It Again

www.anandtech.com

Indeed, according to Anandtech, Nehalem was able to achieve higher performance than Penryn with fewer transistors:

Intel Developer Forum 2007 - Day 1: More Nehalem, Penryn Announced, and Gelsinger Speaks

www.anandtech.com

 

[albertserene]

It's weird with all these newly created accounts that suddenly have an opinion on who Apple should hire or whatnot for their SoC development.

18% increase in CPU performance is great. Have in mind that there is a lot more to the A16 Bionic SoC than just processor speed. Everything else got a big increase as well.

No one else offers anything close to that performance in the phone form factor.

So what are you saying? Newbies are not entitled to their opinions? Another conspiracy theory? The Mac's competition is with x86 architecture. Raptor Lake already overtake M1 Ultra's performance. On the phone SoC, Snapdragon 8 gen 2 already overtake A16 in Antutu. As I have said, I am a big fan of Apple. It's time to move forward. The world doesn't standstill.

 

[albertserene]

My opinion exactly. The 2023 will be a deciding year. If Apple does not come out with a new, significantly better microarchitecture this year then I’ll agree that they are in trouble.



Do they? Or could it be the case that ARM might believe that architectural license transfer is not legal under these circumstances? In other words, why do you interpret this as ARM being worried about Nuvia designs rather then them simply believing that Qualcomm oves them money?

ARM stopped the Nuvia's license when they learned Qualcomm will take over. Qualcomm does not owe ARM money. ARM does it because they feared Qualcomm will stop using ARM's reference design after acquring Nuvia. It shows ARM sees Nuvia as a serious competition to their reference design. Yeah, we will see if Apple makes big changes in the performance cores in 2023. I am not optimistic about that. It has been 3 years since Firestorm came out and we are just seeing 18% improvement over those 3 years.

 

[theorist9]

On the phone SoC, Snapdragon 8 gen 2 already overtake A16 in Antutu.

...but remains significantly behind it in Geekbench:

Qualcomm Snapdragon 8 Gen 2 Geekbench Benchmark score

See how Qualcomm Snapdragon 8 Gen 2 Geekbench benchmark ❤️ score compares ⚡ with other processors and check out his rank and performance results in tests.

unite4buy.com