Anyone who works in or develops HPC software cringes at this. Sure it's likely you can make an ARM chip with the performance of high-end Xeons, but the world of HPC software is a million years away from XCode app development where you can flip a switch to build for a new architecture. There are so many bespoke libraries (e.g. Intel MKL) and years of optimization that have gone into getting HPC code to run fast on AVX Xeons.
Apple is a bit-player in HPC with the Mac Pro because of their 'war' with nVIDIA ( cutting off access to the compute power of their massively parallel GPUs) - it would just sideline them even more if they went ARM.
GC (Grand Central) is Apples answer, as parallelism is the answer to supercomputer power. Works fine with XCode.
Grand Central works as well as the Backplane and interprocessing designs of the CPU architecture. In short, it's not a panacea. Slap two TR or EPYC CPUs in today's Mac Pro and you've got 128 Cores/ 256 threads that nothing ARM designs with it's energy footprint limitations will ever touch. Zen 3 arrives this Fall. Sorry, but no, Apple won't invest heavy resources for a small market product. They'll decide between continuing with Intel, or finally move forward on AMD Zen.
RDNA 2.0 is about to come out and it lays to waste the GPGPUs custom designed by AMD already. The rational mind says Zen 3/RDNA 2.0/Afterburner 2.0 in the next Mac Pro.
Anyone who works in or develops HPC software cringes at this. Sure it's likely you can make an ARM chip with the performance of high-end Xeons, but the world of HPC software is a million years away from XCode app development where you can flip a switch to build for a new architecture. There are so many bespoke libraries (e.g. Intel MKL) and years of optimization that have gone into getting HPC code to run fast on AVX Xeons.
Apple is a bit-player in HPC with the Mac Pro because of their 'war' with nVIDIA ( cutting off access to the compute power of their massively parallel GPUs) - it would just sideline them even more if they went ARM.
I don't think Apple goes after the HPC market. These are mostly high powered machines for creatives. I don't build for HPC, but I think it is generally run on clusters of Linux machines. Despite that, there may be HPC applications for ARM chips on Linux. If you only burn 200 watts for the same performance of 400 watts on Xeon, that may be compelling for some HPC project. Particularly if designing something new from square one.
1. Is Apple capable of making a CPU for the Mac Pro that makes the Mac Pro a better computer than what it would b with an Intel or AMD CPU?
2. If so, does it make business sense for Apple to do that?
3. If so, will Apple actually do it?
I think the answers are:
1. Definitely yes. Apple has the talent and the money to do this. 99.9% probability here.
2. Less clear, but I think so. It depends largely on whether it makes sense to move the design of the iPhone SOC to something akin to AMD's chiplet design, where you have multiple dies together in the same package. If they went that route, the functions of the iPhone SOC could be split into "compute" and "I/O", or CPU, GPU, and I/O. The same "compute" or "CPU" die used in the iPhone could then also be used in a Mac Pro, just a larger number of them. If that makes sense for the iPhone, then the marginal cost of making a pro-desktop SOC is much lower. 75% probability here
3. This is where it gets really hard to predict. It seems that Apple has perhaps realized that the Mac isn't going to be replaced by th iPad and they need to plan for a long future for the Mac. So... maybe. I'd say coin toss.
Anyone who works in or develops HPC software cringes at this. Sure it's likely you can make an ARM chip with the performance of high-end Xeons, but the world of HPC software is a million years away from XCode app development where you can flip a switch to build for a new architecture. There are so many bespoke libraries (e.g. Intel MKL) and years of optimization that have gone into getting HPC code to run fast on AVX Xeons.
Apple is a bit-player in HPC with the Mac Pro because of their 'war' with nVIDIA ( cutting off access to the compute power of their massively parallel GPUs) - it would just sideline them even more if they went ARM.
I don't think Apple goes after the HPC market. These are mostly high powered machines for creatives. I don't build for HPC, but I think it is generally run on clusters of Linux machines. Despite that, there may be HPC applications for ARM chips on Linux. If you only burn 200 watts for the same performance of 400 watts on Xeon, that may be compelling for some HPC project. Particularly if designing something new from square one.
What they do want is the desktop machines that the researchers use to write code to run on the HPC clusters so having hardware that is closer isn't a bad thing.
Anyone who works in or develops HPC software cringes at this. Sure it's likely you can make an ARM chip with the performance of high-end Xeons, but the world of HPC software is a million years away from XCode app development where you can flip a switch to build for a new architecture. There are so many bespoke libraries (e.g. Intel MKL) and years of optimization that have gone into getting HPC code to run fast on AVX Xeons.
Apple is a bit-player in HPC with the Mac Pro because of their 'war' with nVIDIA ( cutting off access to the compute power of their massively parallel GPUs) - it would just sideline them even more if they went ARM.
I’ve been discussing this for some time and have yet to see anyone actually generate LINPACK or HPCG benchmark results let alone easy to use software. You can blame it on Apple or NVIDIA but HPC hasn’t been on Apple’s radar for years.
Fujitsu FX100. ARM based specifically made for HPC and outperforming Xeons. So it can be done.
Imagine macOS optimized to run nearly everything on ARM64 while still allowing interoperability with x86-64 applications. The two processor complexes could be connected via Thunderbolt, which at 40 Gb/sec is pretty fast.
Isn't this exactly what "Sidecar" to iPadOS already does?
Run that across PCIe which Apple has in the Aseries for USBc without the overhead for USBc and you have as much bandwidth as Intels DMI bus. Apple gets a unified hardware platform off all common protocols, USB, radios, screens, charging, input. Everything else is industry standard PCIe. Apple gets freed up to buy x86 heavy-applications processors and supporting GPUs from whoever can pass a validation test.
MacOS has been flirting with this path for a decade.
"iPadOS unleased" could handle the frontend and look Mac like doing it, Catalyst apps handle Mac apps on that frontend. Which gets the ball rolling. Legacy Mac apps run on the x86 until they are certified. Apple does in house testing of other MacStore Apps against ARM frontend and flags them as Yes/No in the downloads to the machine. Developers get a suite to tools to make their App ready to run in part or in full on the front. With a system
Macs to are ridge trucks, about time they offered articulated ones as well.
Amazon appears to have designed a pretty vanilla 64 core ARM server chip, fabbed on TSMC 7nm, that is very competitive with x86 (and really wipes the floor on performance/$).
Apple could share the same CPU design between Mac Pro and datacenter.
I would take his word with a grain of salt. He doesn't work for Apple so I just don't think this is happening that soon if at all. They need to make that kind of processor do everything that the Intel one does today. Many businesses use the Mac now because they have dual operating system capabilities. Changing to anything else would possibly be a major hit on business customers and home users as well.
Amazon appears to have designed a pretty vanilla 64 core ARM server chip, fabbed on TSMC 7nm, that is very competitive with x86 (and really wipes the floor on performance/$).
Apple could share the same CPU design between Mac Pro and datacenter.
Assuming a competent chip team, it's worth reiterating that the number one driver for chip performance is the fab that they are made on. AMD's Zen 2 and Apple's SoC chips are made on TSMC 7 nm while Intel is still fabbing its desktop chips on their 14 nm fab.
That's about a factor of 2 in transistor density between the two fabs, and you don't make up that kind of transistor budget deficit through ISA or chip microarchitecture. This is why Apple's A12X at 5 W compares well to Intel 14nm 28 W chips in the MBP, or Ryzen and TR compare well to Intel's 14nm desktop chips.
There's skepticism on Apple being able to design a workstation architecture, but I have zero doubts they can do it, and as long as they are fabbed on TSMC's leading edge fab, they will be the fastest machines in its class.
macOS/ARM is all about 3rd party software and the balance of trades with losing x86 compatibility. Things that are basically out of Apple's hands.
Anyone who works in or develops HPC software cringes at this. Sure it's likely you can make an ARM chip with the performance of high-end Xeons, but the world of HPC software is a million years away from XCode app development where you can flip a switch to build for a new architecture. There are so many bespoke libraries (e.g. Intel MKL) and years of optimization that have gone into getting HPC code to run fast on AVX Xeons.
Apple is a bit-player in HPC with the Mac Pro because of their 'war' with nVIDIA ( cutting off access to the compute power of their massively parallel GPUs) - it would just sideline them even more if they went ARM.
I cringe at your comment.
Who needs Intel MKL, just replace it with an ARM equivalent, thats why it is a library (swapping it is easy), it might even be less buggy, Apple can switch to ARM because it makes the SoC including the GPU’s and other accelerator processors with accompanying libraries and frameworks like GC and doesn’t need NVIDIA.
Fun fact: "...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
Fun fact: "...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
This forum oddly has plenty of Apple fanatics that say it can't be done despite sites like AnandTech already detailing how Apple's low-power silicon designed for a handheld tablet has been trouncing more expensive Intel chips that couldn't even be used in a tablet do to their TDP.
Fun fact: "...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
This forum oddly has plenty of Apple fanatics that say it can't be done despite sites like AnandTech already detailing how Apple's low-power silicon designed for a handheld tablet has been trouncing more expensive Intel chips that couldn't even be used in a tablet do to their TDP.
Yes. Lots of skepticism that Apple doesn't have the technical chops to do it or that the economics just can't work out. Never mind that Amazon designed competitive ARM server designs or that AMD's revenue was less than $7b for all of 2019. That's like less than Apple's AirPods revenue I think. Heck, ARM Holdings is like owned by Softbank and they are looking to dump ARM Holdings I think. They are like a sub $1b company that designs reference ARM CPU designs that powers most of the phone and IoT world, including Amazon's Graviton.
As I said earlier, the primary driver to this is fabbing on the most state of art fab. This been Intel for the about 40 years. For the past two years and the foreseeable future, it is TSMC, a company that Apple has a lot of familiarity with and has a rather large share of their capacity.
Fun fact: "...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
This forum oddly has plenty of Apple fanatics that say it can't be done despite sites like AnandTech already detailing how Apple's low-power silicon designed for a handheld tablet has been trouncing more expensive Intel chips that couldn't even be used in a tablet do to their TDP.
Yes. Lots of skepticism that Apple doesn't have the technical chops to do it or that the economics just can't work out. Never mind that Amazon designed competitive ARM server designs or that AMD's revenue was less than $7b for all of 2019. That's like less than Apple's AirPods revenue I think. Heck, ARM Holdings is like owned by Softbank and they are looking to dump ARM Holdings I think. They are like a sub $1b company that designs reference ARM CPU designs that powers most of the phone and IoT world, including Amazon's Graviton.
As I said earlier, the primary driver to this is fabbing on the most state of art fab. This been Intel for the about 40 years. For the past two years and the foreseeable future, it is TSMC, a company that Apple has a lot of familiarity with and has a rather large share of their capacity.
It is pretty strange that people think that the ARM architecture is weak at things like multitasking and SIMD. I get that most people here are not electrical/computer engineers but just common sense should show that the ARM architecture is fully modern. It is quite a bit newer than x86 and the 64-bit version is even newer.
Anyone who has used a recent iPad Pro can also attest to the performance of the A12x. The idea that the A12x with 8 cores and 10 billion transistors can't multitask well is ludicrous.
Fun fact: "...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
This forum oddly has plenty of Apple fanatics that say it can't be done despite sites like AnandTech already detailing how Apple's low-power silicon designed for a handheld tablet has been trouncing more expensive Intel chips that couldn't even be used in a tablet do to their TDP.
Yes. Lots of skepticism that Apple doesn't have the technical chops to do it or that the economics just can't work out. Never mind that Amazon designed competitive ARM server designs or that AMD's revenue was less than $7b for all of 2019. That's like less than Apple's AirPods revenue I think. Heck, ARM Holdings is like owned by Softbank and they are looking to dump ARM Holdings I think. They are like a sub $1b company that designs reference ARM CPU designs that powers most of the phone and IoT world, including Amazon's Graviton.
As I said earlier, the primary driver to this is fabbing on the most state of art fab. This been Intel for the about 40 years. For the past two years and the foreseeable future, it is TSMC, a company that Apple has a lot of familiarity with and has a rather large share of their capacity.
It is pretty strange that people think that the ARM architecture is weak at things like multitasking and SIMD. I get that most people here are not electrical/computer engineers but just common sense should show that the ARM architecture is fully modern. It is quite a bit newer than x86 and the 64-bit version is even newer.
Anyone who has used a recent iPad Pro can also attest to the performance of the A12x. The idea that the A12x with 8 cores and 10 billion transistors can't multitask well is ludicrous.
Remember how impossible it was for Apple to move from 32-bit to 64-bit. And before that Apple buying PA Semi was a foolish move that wouldn’t yield anything that others couldn’t do better than some gadget company that makes overpriced MP3 players.
I would take his word with a grain of salt. He doesn't work for Apple so I just don't think this is happening that soon if at all. They need to make that kind of processor do everything that the Intel one does today. Many businesses use the Mac now because they have dual operating system capabilities. Changing to anything else would possibly be a major hit on business customers and home users as well.
Exactly the question isn't a technical one it's market one. Why would people who need to get work done pay money for something that doesn't work as well and isn't going to work any better for the foreseeable future?
They could just keep current hardware and be no worse off. Anything software companies do to try and make their offer work better on new machines (better multi-threading) will still help pay the bills on retained hardware that already has idol multi-cores running that software anyway.
I would take his word with a grain of salt. He doesn't work for Apple so I just don't think this is happening that soon if at all. They need to make that kind of processor do everything that the Intel one does today. Many businesses use the Mac now because they have dual operating system capabilities. Changing to anything else would possibly be a major hit on business customers and home users as well.
Exactly the question isn't a technical one it's market one. Why would people who need to get work done pay money for something that doesn't work as well and isn't going to work any better for the foreseeable future?
They could just keep current hardware and be no worse off. Anything software companies do to try and make their offer work better on new machines (better multi-threading) will still help pay the bills on retained hardware that already has idol multi-cores running that software anyway.
Yes. Apple knows this. So, the question to ponder is what is gained by moving to ARM for the consumer? There are a lot of easy to answer questions regarding what Apple gains, but for buyers of Apple ARM PCs? Interesting.
Apple could make a 10 mm thick laptop as powerful as a MBP13. The iPad Pro is 6 mm thick. They have 4 mm to add a keyboard and trackpad. It's more move the battery, logic board, speakers, etc, to the base and have a 4 mm thick display lid. It could be 2 lb? A feature that sells it over x86 computers? Enough to make the lack of compatibility something people are willing to bear?
Taking it a little further, this could make 10 mm thick dual display clamshells or a foldable possible with good battery life. For iMacs, headless desktops, it would mean quieter machines. An iMac version could be fanless. An iMac could be 10 mm thin. Apple could put in dedicated hardware that is designed for their hardware, like dedicated hardware to make wireless AR glasses possible?
Comments
1. Is Apple capable of making a CPU for the Mac Pro that makes the Mac Pro a better computer than what it would b with an Intel or AMD CPU?
2. If so, does it make business sense for Apple to do that?
3. If so, will Apple actually do it?
I think the answers are:
1. Definitely yes. Apple has the talent and the money to do this. 99.9% probability here.
2. Less clear, but I think so. It depends largely on whether it makes sense to move the design of the iPhone SOC to something akin to AMD's chiplet design, where you have multiple dies together in the same package. If they went that route, the functions of the iPhone SOC could be split into "compute" and "I/O", or CPU, GPU, and I/O. The same "compute" or "CPU" die used in the iPhone could then also be used in a Mac Pro, just a larger number of them. If that makes sense for the iPhone, then the marginal cost of making a pro-desktop SOC is much lower. 75% probability here
3. This is where it gets really hard to predict. It seems that Apple has perhaps realized that the Mac isn't going to be replaced by th iPad and they need to plan for a long future for the Mac. So... maybe. I'd say coin toss.
What they do want is the desktop machines that the researchers use to write code to run on the HPC clusters so having hardware that is closer isn't a bad thing.
https://www.anandtech.com/show/15578/a-peek-into-the-physics-of-graivton2-amazons-neoverse-n1-server-chip-first-impressions
Amazon appears to have designed a pretty vanilla 64 core ARM server chip, fabbed on TSMC 7nm, that is very competitive with x86 (and really wipes the floor on performance/$).
Apple could share the same CPU design between Mac Pro and datacenter.
Apple can switch to ARM because it makes the SoC including the GPU’s and other accelerator processors with accompanying libraries and frameworks like GC and doesn’t need NVIDIA.
https://gigaom.com/2013/07/06/forget-servers-one-day-facebook-google-and-other-web-giants-will-make-their-own-custom-chips/
...and we had inexpensive Chromebooks with ARM chips performing on an Intel level several years ago and progress has been ramping up fast since.
https://www.theverge.com/2017/2/22/14691396/google-chromebook-arm-laptop-op1-processor-apptop
Fun fact:
"...an advanced chip design using 16 nanometer processes cost an average of $106.3 million, shifting to 10 nanometers pushed that up to $174.4 million, and the move to 7 nanometers costs $297.8 million, with projections for 5 nanometer chips to be on the order of $542.2 million. Nearly half of that cost has been – and continues to be – for software. "
As I said earlier, the primary driver to this is fabbing on the most state of art fab. This been Intel for the about 40 years. For the past two years and the foreseeable future, it is TSMC, a company that Apple has a lot of familiarity with and has a rather large share of their capacity.
Anyone who has used a recent iPad Pro can also attest to the performance of the A12x. The idea that the A12x with 8 cores and 10 billion transistors can't multitask well is ludicrous.
Apple could make a 10 mm thick laptop as powerful as a MBP13. The iPad Pro is 6 mm thick. They have 4 mm to add a keyboard and trackpad. It's more move the battery, logic board, speakers, etc, to the base and have a 4 mm thick display lid. It could be 2 lb? A feature that sells it over x86 computers? Enough to make the lack of compatibility something people are willing to bear?
Taking it a little further, this could make 10 mm thick dual display clamshells or a foldable possible with good battery life. For iMacs, headless desktops, it would mean quieter machines. An iMac version could be fanless. An iMac could be 10 mm thin. Apple could put in dedicated hardware that is designed for their hardware, like dedicated hardware to make wireless AR glasses possible?