TSMC starting production of 3nm chips for Mac, iPhone

AppleInsider · December 26, 2022 1:34PM

Within days, TSMC will start mass production of its 3-nanometer chip process for the next generations of Mac, iPhone, and other Apple devices.

TSMC starts 3nm chips

According to a report on Monday from DigiTimes, TSMC will hold a ceremony at Fab 18 at the Southern Taiwan Science Park (STSP) on December 29 to mark the occasion. It will also reveal plans to expand 3-nanometer chip production at the fab.

Apple currently uses 4-nanometer chips from TSMC in the iPhone 14 Pro models, which is the A16 Bionic chip.

TSMC began testing the 3-nanometer process in December 2021 at Fab 18. A rumor in June claimed that Apple could use the new chip process in its M2 Pro chip in certain Macs that will ship in 2023.

The chip may appear in an updated Mac mini, 14-inch MacBook Pro, and 16-inch MacBook Pro expected to debut in 2023.

TSMC's 3nm processes would provide multiple benefits to Apple, including a predicted 15% speed improvement at the same power level as a chip made using 5nm processes, or a 30% reduction in power consumption for similar speeds.

Read on AppleInsider

lkrupp · December 26, 2022 2:51PM

So how close is the competition to this or are TSMC and Apple now leaving the others in the dust? What are Qualcomm, Samsung, and Intel up to? Still speed+tons of heat?

macxpress · December 26, 2022 4:09PM

lkrupp said:

So how close is the competition to this or are TSMC and Apple now leaving the others in the dust? What are Qualcomm, Samsung, and Intel up to? Still speed+tons of heat?

Sounds like their competitors are developing an M1 killer! lol

As always their competition focus's on the current and not the future. Their competition shouldn't be developing an M1 killer...they should be focusing on an M3 killer. By the time they have an M1 killer out Apple will have an M3 or M4 released.

nubus · December 26, 2022 4:14PM

It was expected for TSMC to drop the N3 process due to limited performance improvements and problems with energy consumption. Instead we get it. If this is more than "we promised to start production before end of year" then MacBook Pro and Studio are prime candidates for N3. M2 will likely never end up in.a MacBook Pro and M3 might be "1st generation 3nm".

mikethemartian · December 26, 2022 5:48PM

lkrupp said:

So how close is the competition to this or are TSMC and Apple now leaving the others in the dust? What are Qualcomm, Samsung, and Intel up to? Still speed+tons of heat?

Well ASML makes the machines that TSMC uses and the very first of ASML’s next generation high numerical aperture machines is going to Intel. It all depends on what sort of process advantages TSMC might have on other fabs that use the ASML equipment and how well optimized their PDK is.

JinTech · December 26, 2022 6:23PM

nubus said:

It was expected for TSMC to drop the N3 process due to limited performance improvements and problems with energy consumption. Instead we get it. If this is more than "we promised to start production before end of year" then MacBook Pro and Studio are prime candidates for N3. M2 will likely never end up in.a MacBook Pro and M3 might be "1st generation 3nm".

It’s possible these chips are indeed M3 chips based on what the iPhone 15 Pro Max phones will have. If they are putting M series chips in iPads, perhaps they will start putting them in iPhones. Kind of a convoluted mess but hey, as long as we get the speed, efficiency and the fastest chips out there, who cares how Apple does it? That’s for marketing to sort out

tbornot · December 26, 2022 6:51PM

The 3nm process is for the new GPU ray tracing, which is why 3nm is delayed. Must have cost Apple a pretty penny to delay the 3nm line.

tbornot · December 26, 2022 7:01PM

I think people are making a fundamental mistake thinking events at Apple are disjoint. I think everything that has happened in the last year focuses on the most non-Apple event, Apple gave a large donation to Blender, and sent engineers. Try that again, Apple sent engineers to Blender. Everything revolves around this action.

9secondkox2 · December 26, 2022 7:25PM

tbornot said:

The 3nm process is for the new GPU ray tracing, which is why 3nm is delayed. Must have cost Apple a pretty penny to delay the 3nm line.

Pretty much. Apple has a plan. Disappointing SOCs, when the advent of Apple Silicon has become a huge sales driver, is not part of that plan.

No doubt the lockdown policies associated with COVID caused some significant development issues and delays.

I can see Apple forgoing multiple M2 derivatives in order to fast track the M3 in the hands of customers. The M2 platform (based on a16 with gimped GPU) isn’t going to cut it for the Mac Pro, big dog iMac, and be seen as an upgrade to M1 Max MacBook Pros. Taking the financial hit to avoid a bigger financial hit resulting from being seen as sluggish with their SOCs. A significant CPU performance jump over M2 with a generational leap over the GPU would keep the momentum flowing and be worth the losses associated with 3nm delays and forgoing production of m2 derivative designs.

timmillea · December 26, 2022 8:26PM

3nM is around the theoretical limit of feasibility using silicon. I wonder if there will be associated reliability/longevity issues? I would prefer some form of ECC baked in. It won't be a good day when you discover that a small percentage of your files have been slightly corrupted by an ageing 3nM SOC.

peterhart · December 26, 2022 11:32PM

What happens when they finally get to a 1nm process? What’s next?

blastdoor · December 27, 2022 7:58AM

I think it’s still more likely that m2 pro will be based on one of the 5/4nm processes

Marvin · December 27, 2022 11:34AM

tbornot said:

I think people are making a fundamental mistake thinking events at Apple are disjoint. I think everything that has happened in the last year focuses on the most non-Apple event, Apple gave a large donation to Blender, and sent engineers. Try that again, Apple sent engineers to Blender. Everything revolves around this action.

Other companies do this too, Nvidia has people working on Blender. It's a good way for them to keep their silicon competitive because they can see exactly the algorithms that need accelerated. On the following page, the M1 Ultra GPU is on page 4:

Blender GPU benchmarks

Intel's new dedicated GPU A770 scores just above it (A770 has hardware raytracing). The M1 Ultra (21TFLOPs) scores 1371. The Nvidia 4090 (82 TFLOPs) scores 12142.

The hardware raytracing makes up for about 2-3x the performance difference and the 4090 uses 2.5x the power.

An M3 Ultra would be about 35 TFLOPs and if it had hardware raytracing, it could potentially score around 7000 at the same power level as the M1 Ultra. While that would be short of the 4090, it's competitive and runs at a fraction of the power. A Duo version of this for a Mac Pro would be able to top a 4090.

I doubt that Apple working on Blender is one of the most significant things they are doing, they only seem to have a couple of people on it but for things like AR, having an efficient engine for generating high quality real-time content is a must and the hardware needs to be designed to support this.

peterhart said:

What happens when they finally get to a 1nm process? What’s next?

A 1nm chip would be 5x the performance of what we have now. This would mean MBPs are as fast as the high-end Mac Pros and Macbook Airs (and iPads) are as fast as high-end MBPs and hardware raytracing will be everywhere. At this point, most people will be content with that performance level because it will handle photoreal real-time 3D at 4K/8K. Chip manufacturers could comfortably stop at 1nm.

This is why it's important for hardware companies to diversify their company with things like content and services, including supporting content creation.

dk49 · December 27, 2022 1:33PM

The product which needs this 3nm chip the most is Apple Watch. It will improve the battery life dramatically (more than 2x just by this without increasing performance); as S6/7/8 uses 7nm.

blastdoor · December 27, 2022 3:07PM

peterhart said:

What happens when they finally get to a 1nm process? What’s next?

Hint: processes used to be measured in microns.

9secondkox2 · December 27, 2022 7:22PM

peterhart said:

What happens when they finally get to a 1nm process? What’s next?

3D stacks probably.

9secondkox2 · December 27, 2022 7:24PM

JinTech said:

nubus said:

It was expected for TSMC to drop the N3 process due to limited performance improvements and problems with energy consumption. Instead we get it. If this is more than "we promised to start production before end of year" then MacBook Pro and Studio are prime candidates for N3. M2 will likely never end up in.a MacBook Pro and M3 might be "1st generation 3nm".

It’s possible these chips are indeed M3 chips based on what the iPhone 15 Pro Max phones will have. If they are putting M series chips in iPads, perhaps they will start putting them in iPhones. Kind of a convoluted mess but hey, as long as we get the speed, efficiency and the fastest chips out there, who cares how Apple does it? That’s for marketing to sort out

Not enough mass to dissipate M series hear in an iPhone yet - though I can see Apple moving the iPhone Pro to a larger size and going with a 3nm process m2.

9secondkox2 · December 27, 2022 7:29PM

Marvin said:

tbornot said:

I think people are making a fundamental mistake thinking events at Apple are disjoint. I think everything that has happened in the last year focuses on the most non-Apple event, Apple gave a large donation to Blender, and sent engineers. Try that again, Apple sent engineers to Blender. Everything revolves around this action.

Other companies do this too, Nvidia has people working on Blender. It's a good way for them to keep their silicon competitive because they can see exactly the algorithms that need accelerated. On the following page, the M1 Ultra GPU is on page 4:

Blender GPU benchmarks

Intel's new dedicated GPU A770 scores just above it (A770 has hardware raytracing). The M1 Ultra (21TFLOPs) scores 1371. The Nvidia 4090 (82 TFLOPs) scores 12142.

The hardware raytracing makes up for about 2-3x the performance difference and the 4090 uses 2.5x the power.

An M3 Ultra would be about 35 TFLOPs and if it had hardware raytracing, it could potentially score around 7000 at the same power level as the M1 Ultra. While that would be short of the 4090, it's competitive and runs at a fraction of the power. A Duo version of this for a Mac Pro would be able to top a 4090.

I doubt that Apple working on Blender is one of the most significant things they are doing, they only seem to have a couple of people on it but for things like AR, having an efficient engine for generating high quality real-time content is a must and the hardware needs to be designed to support this.

peterhart said:

What happens when they finally get to a 1nm process? What’s next?

A 1nm chip would be 5x the performance of what we have now. This would mean MBPs are as fast as the high-end Mac Pros and Macbook Airs (and iPads) are as fast as high-end MBPs and hardware raytracing will be everywhere. At this point, most people will be content with that performance level because it will handle photoreal real-time 3D at 4K/8K. Chip manufacturers could comfortably stop at 1nm.

This is why it's important for hardware companies to diversify their company with things like content and services, including supporting content creation.

People won’t ever “arrive” at a stopping point for cpu speed and power. There will be other advances. It won’t be the same game as today.

Some theories that seem impossible now involve light in lieu of electricity in addition to using 3D stacks to theoretically multiply the speed. AMD is already using that idea to double and triple server caches. There are a lot of ideas being researched and developed concerning where to go next.

TSMC starting production of 3nm chips for Mac, iPhone

Comments