Apple announces opening trio of Apple Silicon Macs

[DanceswithUnix]

AMD Renoir is 9.8 billlion transistors and is an SOC.

Exactly. Now there are a whole ton of things we just don't know yet, but 60% more transistors for a chip with the extra camera processing and tensor units seems a bit high but not unreasonable to me. A14 is supposed to be 11.8B transistors, it might be more interesting to try and work forwards from that figure.

I don't know why you are talking like single threaded is easy; my old fx 8350 was pretty good in multi threaded tests it is the single threaded stuff that requires the high issue width, bit reorder buffers, spot on prediction etc. Bulldozer's clock speed certainly didn't save it there.

I totally agree that AMD & Intel have nothing to worry about from this though.

[CAT-THE-FIFTH]

Exactly. Now there are a whole ton of things we just don't know yet, but 60% more transistors for a chip with the extra camera processing and tensor units seems a bit high but not unreasonable to me. A14 is supposed to be 11.8B transistors, it might be more interesting to try and work forwards from that figure.

I don't know why you are talking like single threaded is easy; my old fx 8350 was pretty good in multi threaded tests it is the single threaded stuff that requires the high issue width, bit reorder buffers, spot on prediction etc. Bulldozer's clock speed certainly didn't save it there.

I totally agree that AMD & Intel have nothing to worry about from this though.

Apple A14 is only a dual core,with 4 small low power cores. Benchmarks are testing the performance of the two performance cores. Dual core SOC is 11.8 billion transistors,and quad core SOC is 16 billion transistors.

I am not saying its easy,but again Apple is throwing a ton of transistors at the problem on a cutting edge node. Renoir is double the performance core count than the M1(8 cores with HT). So 3X the amount of threads presented to an OS. So if they are going for large cores,and a limited number(which means memory bandwidth is less of a problem),then I would be shocked if they didn't make a CPU which performs well in single core benchmarks,especially with a closed software ecosystem.

OTH,AT trying to make it sound like AMD and Intel are DOOOOOOOOOOOOOOOOOOOMED(tm) is frankly going a bit OTT. I am just getting fedup with everything Apple being overhyped to the nTH degree,and basically many of the tech press implying everyone else should just not bother,with virtually most Apple iProducts launched.They never cover all the issues and problems Apple products seem to have too. It's like some sort of tech religion now.

In the scheme of things both Mac and iPad sales are down. Macs are well under 10% of the entire laptop/desktop market,so if anything Qualcomm,Mediatek,Hisilicon,etc are actually having more impact on the market. Apple is basically merging its iPads,Macs and iPhones into one line to save money.

Also funny the moment AMD actually beats Intel in all metrics,the tech press have to cling onto Apple or some other company!

[DanceswithUnix]

I am not saying its easy,but again Apple is throwing a ton of transistors at the problem on a cutting edge node.

But AFAICS that might have worked back in the early Pentium days where you could just throw transistors at things like the primitive floating point units to make them faster, but modern chips have already had that done.

So more transistors can easily buy you more cache, and easily buy you more core count. What it doesn't easily buy you is single thread performance, but I believe Apple have been working towards this for years.

[CAT-THE-FIFTH]

But AFAICS that might have worked back in the early Pentium days where you could just throw transistors at things like the primitive floating point units to make them faster, but modern chips have already had that done.

So more transistors can easily buy you more cache, and easily buy you more core count. What it doesn't easily buy you is single thread performance, but I believe Apple have been working towards this for years.

Yes but it doesn't change the fact they are relying on jumping onto new nodes quickly and throwing tons of transistors on large cores,caches,etc. This is how Apple is going wider and wider and wider.

This does buy you single core performance if not you have to limit transistor count due to power considerations. Then tons of transistors on custom accelleration blocks on top of this.

See the few times Apple couldn't jump quickly and they ended up stagnating. Imagine if Intel had functioning nodes according to their roadmaps,they would have been on much higher performance cores by now.

Another issue is that AMD and Intel seem to have far more experience in packaging technologies,which Apple lacks. Apple is still making very dense monolithic dies,and it makes me wonder how much redundancy is in these SOCs??

Is their 8 and 16 core derivatives going to be chiplets?? Do they have a low power connection fabric?? The tech press has none of these answers,but seem to hype still.

This is why Apple is so desperate to jump onto new nodes as quickly as possible. You wait and see what happens if Samsung and TSMC have problems,companies like AMD and Intel are much more used to optimising on existing nodes. This is the same problem ATI had - they used to jump to new nodes much quicker than Nvidia,but Nvidia tended to get much more out of older nodes than ATI historically. When TSMC had problems at 28NM its when ATI/AMD fell behind massively.

As usual the tech press hype everything Apple does and AT literally implying AMD/Intel are doomed in their latest article(they can't catch up). All based on synthetic tests using different OSes,using one core. Literally just dual and quad cores on the lowest power consumption nodes money can buy. Like I said if they can't get decent performance out of this,then something has gone wrong.

Edit!!

Also if process nodes had no bearing on performance,it makes me wonder whether you could get Zen3 to work on 32NM,and how performance and power would actually look.

[3dcandy]

Has nobody noticed the massive problem with this as well. DRAM is integrated on the SoC so these new Apple laptops are limited to 16 gig of memory or 8 if you buy it like that. You can't ever upgrade as the memory is integrated in the SoC

So your memory fails - it's bin the laptop time because Apple now pairs other chips to the motherboard meaning you cannot repair it. Even Apple don't repair them nowadays they bin them. iphone 12 is the same, no spares if anything fails so if you break your screen they currently bin the lot

[CAT-THE-FIFTH]

OK,I looked at some measurements for the cores.

From TH:

https://www.tomshardware.com/uk/news...ionic-revealed

the dual-core FireStorm complex with a big L2 cache is around 9.1 mm2

So a 2C/2T FireStorm core on 5NM is 9.1MM2 it appears. TSMC 5NM is apparently upto 80% denser than TSMC 7NM:
https://www.kitguru.net/components/c...ensity-vs-7nm/

So despite the "large 4MB L2 cache" the process node is much more denser than 7NM. The A14 is 11.8 billion transistors on TSMC 5NM and is around 88MM2. Renoir is 9.8 billion transistors and is 156MM2.

From Reddit:
https://www.reddit.com/r/Amd/comment...t_6_comparing/

Each Zen2 core is around 2.83MM2. The 512KB L2 cache is 0.81MM2. So that comes to around 7.28MM2 on a less denser TSMC 7NM.

25% larger surface area on a much higher density node. Also there is spacing considerations with Zen2/Zen3,ie,as they run at higher clockspeeds and voltages,to have the heat distribution more even. Each Zen2/Zen3 core will have an additional small area penalty due to the use of SMT.

If you look at Zen3 compared with Zen2,the chiplet is around 10% larger apparently:
https://twitter.com/chiakokhua/statu...67162968371200

So if that is mostly down to the cores getting larger,its still smaller per core(in transistor count) than the A14 on a less denser process node.

Has nobody noticed the massive problem with this as well. DRAM is integrated on the SoC so these new Apple laptops are limited to 16 gig of memory or 8 if you buy it like that. You can't ever upgrade as the memory is integrated in the SoC

So your memory fails - it's bin the laptop time because Apple now pairs other chips to the motherboard meaning you cannot repair it. Even Apple don't repair them nowadays they bin them. iphone 12 is the same, no spares if anything fails so if you break your screen they currently bin the lot

The warranties are short on the iPhones. IIRC its 12 months for the phone and 24 months for the battery. So if we go by what Louis Rossman has shown in his videos,Apple wants you to buy a new product when it fails.

Got to increase those margins!

[3dcandy]

OK,I looked at some measurements for the cores.

From TH:

https://www.tomshardware.com/uk/news...ionic-revealed


So a 2C/2T FireStorm core on 5NM is 9.1MM2 it appears. TSMC 5NM is apparently upto 80% denser than TSMC 7NM:
https://www.kitguru.net/components/c...ensity-vs-7nm/

So despite the "large 4MB L2 cache" the process node is much more denser than 7NM. The A14 is 11.8 billion transistors on TSMC 5NM and is around 88MM2. Renoir is 9.8 billion transistors and is 156MM2.

From Reddit:
https://www.reddit.com/r/Amd/comment...t_6_comparing/

Each Zen2 core is around 2.83MM2. The 512KB L2 cache is 0.81MM2. So that comes to around 7.28MM2 on a less denser TSMC 7NM.

25% larger surface area on a much higher density node. Also there is spacing considerations with Zen2/Zen3,ie,as they run at higher clockspeeds and voltages,to have the heat distribution more even. Each Zen2/Zen3 core will have an additional small area penalty due to the use of SMT.

If you look at Zen3 compared with Zen2,the chiplet is around 10% larger apparently:
https://twitter.com/chiakokhua/statu...67162968371200

So if that is mostly down to the cores getting larger,its still smaller per core(in transistor count) than the A14 on a less denser process node.



The warranties are short on the iPhones. IIRC its 12 months for the phone and 24 months for the battery. So if we go by what Louis Rossman has shown in his videos,Apple wants you to buy a new product when it fails.

Got to increase those margins!

indeed and I know loads of people - over 30 that have taken out a 24 month contract and still had to pay for a repair not covered under warranty before the contract has ended which has effectively resulted in a non working phone but them expected to continue to pay for it

[CAT-THE-FIFTH]

indeed and I know loads of people - over 30 that have taken out a 24 month contract and still had to pay for a repair not covered under warranty before the contract has ended which has effectively resulted in a non working phone but them expected to continue to pay for it

There used to be a time a decade ago or so,when a phone contract was reasonably priced,and was 12 months or a bit longer. You essentially got a massively subsidised phone with talktime,etc. Nowadays it seems more like an interest free loan and a poor value contract,so it makes more sense to buy it up front and get a decent SIM free contract. Sadly because a lot of these "high end" phones are very overpriced(IMHO),I would rather get a cheaper one,and get myself a decent camera with the money saved(and it still ends up cheaper!).

[DanceswithUnix]

So if that is mostly down to the cores getting larger,its still smaller per core(in transistor count) than the A14 on a less denser process node.

My point was more that Intel have been "just throwing transistors at making the chip wider" for years, usually with just a 5% performance gain from a generous helping of gates.

So yeah Apple get a nice big transistor budget, but it's what they do with it that counts.

[protagonist]

From experience it's usually the storage or the battery that tend to develop faults. The SoC/Boards for smartphone usually last for a good few years before giving up, and that's in the thermal envelope of a smartphone as well

[kompukare]

From experience it's usually the storage or the battery that tend to develop faults. The SoC/Boards for smartphone usually last for a good few years before giving up, and that's in the thermal envelope of a smartphone as well

Don't worry Apple have that covered now. First they went with proprietary connectors, now they don't even bother with that and solder their SSDs straight to the PBC making data recovery almost impossible. But then they want you to buy online storage instead.