Moving to AMD Zen 3 and beyond would be a seamless and zero cost direction! While increasing profits, lowering prices and increasing performance. Moving to ARM is a 5 year head ache where in fact ARM will never match nor surpass AMD.
I can't wrap my head around this constant push for AMD in light of what is happening. It's never going to happen. If that was ever on the table it was over a decade ago. And where is your evidence that ARM can neither match nor surpass AMD in performance per watt?
I always respect mdriftmeyer's opinions given his background but on this I agree with you. Why swap Intel for AMD when you have the ability to use in-house technology and be completely in control? AMD are certainly doing great things but the market share Apple will have of total hardware in use for their ARM tech will dwarf all else. So, IMHO, Apple will eventually take the lead across the board. The overall benefits in software across all devices will be massive and never again will waiting on another company's development be an issue. I am sure we all remember the history of this happening over and over again.
Moving to AMD Zen 3 and beyond would be a seamless and zero cost direction! While increasing profits, lowering prices and increasing performance. Moving to ARM is a 5 year head ache where in fact ARM will never match nor surpass AMD.
I can't wrap my head around this constant push for AMD in light of what is happening. It's never going to happen. If that was ever on the table it was over a decade ago. And where is your evidence that ARM can neither match nor surpass AMD in performance per watt?
I always respect mdriftmeyer's opinions given his background but on this I agree with you. Why swap Intel for AMD when you have the ability to use in-house technology and be completely in control? AMD are certainly doing great things but the market share Apple will have of total hardware in use for their ARM tech will dwarf all else. So, IMHO, Apple will eventually take the lead across the board. The overall benefits in software across all devices will be massive and never again will waiting on another company's development be an issue. I am sure we all remember the history of this happening over and over again.
I agree. I like mdriftmeyer and his posts, but the AMD love I can't understand.
Moving to AMD Zen 3 and beyond would be a seamless and zero cost direction! While increasing profits, lowering prices and increasing performance. Moving to ARM is a 5 year head ache where in fact ARM will never match nor surpass AMD.
But it's not about profit, it's about control.
Secondly, going with AMD doesn't give them the ability to optimise the chip for MacOS, that's where the real benefits are going to show.
And let's remember, AMD exists because Intel allows them to.
If you mean control over hardware innovation, I'd say you're right on the money. One could argue that the last meaningful innovation on the Intel platform was Sandy Bridge micro-architecture which was 2011. Recall that i7 processors from Sandy Bridge delivered 4 Core processors at 3.4Ghz. Aside from Intel's tick-tock model squeezing out some meager improvements, they have been coasting in neutral for a decade. For those of you with a Sandy Bridge based i7 MBP, after slapping an SSD in it, the computer is still surprisingly useful, almost 10 years since it was manufactured.
Saying that Apple Silicon is an ARM processor is like someone saying that MacOS is BSD Linux. Following the objectives from WWDC on what they want to do with Apple Silicon, sounds like they are looking to address problems that haven't been addressed in the current x86-64 micro-architectures.
Things that I am particularly interested in;
- How will Apple improve caches and loading caches on cache misses? Intel and AMD's approach on this is sharing registers with cores and calling it hyperthreading.
- In WWDC they listed high-bandwidth caches and unified memory? This sounds like innovation in Front side bus and backside bus or a re-architecture of how instructions are passed to CPU's from memory/cache. Would improvements in memory bus architecture be an improvement over hypthreading models regarding the problem of cache misses.
- Looking at Innovation from Sony in their PS5 in how they are innovating in storage bus to GPU/CPU, would Apple be taking a similar approach in solving problems in read/write latency to storage across a massively improved storage controller.
- All of their eco-system support for camera processors, ML accelerators and Neural engine, video processing, would this be offloading the CPU into cores dedicated to these tasks or are these optimized instructions sets accelerated within cores?
Given that Apple has "cash-on-hand" that exceeds the total Market Cap of Intel. ...and they are willing to bet the farm on this migration, I think that they probably have their own 10 year plan of what they want out of this micro-architecture. I'm optimistic that they have every opportunity to be competitive or to exceed the capability of of Intel and AMD silicon for Mobile processors and fingers crossed, for Desktop as well.
- How will Apple improve caches and loading caches on cache misses? Intel and AMD's approach on this is sharing registers with cores and calling it hyperthreading.
- In WWDC they listed high-bandwidth caches and unified memory? This sounds like innovation in Front side bus and backside bus or a re-architecture of how instructions are passed to CPU's from memory/cache. Would improvements in memory bus architecture be an improvement over hypthreading models regarding the problem of cache misses.
- Looking at Innovation from Sony in their PS5 in how they are innovating in storage bus to GPU/CPU, would Apple be taking a similar approach in solving problems in read/write latency to storage across a massively improved storage controller.
- All of their eco-system support for camera processors, ML accelerators and Neural engine, video processing, would this be offloading the CPU into cores dedicated to these tasks or are these optimized instructions sets accelerated within cores?
I bet no hyperthreading. It's really not a huge benefit for Apple's products unless they start selling servers, and it is expending power on high power components to do it. They are all-in on heterogeneous computing with big CPU cores and small CPU cores, where all the cores will be used simultaneously and developers are supposed to assign quality of service tags on their threads for the system to figure out whether they go to big cores or small cores. It's likely a bigger win in terms of power consumption?
If you look at the A13, I bet Apple's Mac silicon will have humungous amounts of cache. The A13 has 8 MB L2 for the 2 big cores, 4 MB L2 for the 4 small cores, and 16 MB of system level cache or L3 if you want to call it that. That's 28 MB of cache for a phone SoC operating at about 5W. The rumored 8+4 CPU for the first Macs could have 4x that: 64 MB of system level cache, 32 MB of L2 cache for 8 big CPU cores and 8 L2 MB of cache for 4 small cores. So about 100 MB of cache for a SoC going into the Apple Silicon version of the MBP13? At 15W?
The unified memory imo will basically be what you see in iPhone and iPad Pro SoCs. There will be one pool of main memory for all of the compute in the SoC, including the GPU. Might have to go to 4 channels of LPDDR4 or LPDDR5 to feed the GPU for a MBP13. The kicker will be whether the DRAM will be in-package or not. Or on-package or not. Not sure they can stack 32 to 64 GB of DRAM on top of the SoC, but I think they are going to really try. A lot of unknowns with what Apple's going to do for expensive Macs and say 256, 512, 2048 GB of RAM, and how the unified memory is going to feed a RDNA2 level GPU, say 32 CPU cores, etc. Some are saying HBM, but it's memory capacity limited.
I think I'll be happy if the storage controller enabled 8 GB/s read and writes. That's going to be the baseline storage performance for expensive PCs in the PCIe 4 era.
Yes to all that dedicated hardware being in the SoC. They are listed in their Apple Silicon art. If it is listed as box in the graphic, it will be a feature of Apple Silicon Macs.
Comments
Going forward it's an ARMs race.
https://www.anandtech.com/show/15869/new-1-supercomputer-fujitsus-fugaku-and-a64fx-take-arm-to-the-top-with-415-petaflops
If you mean control over hardware innovation, I'd say you're right on the money. One could argue that the last meaningful innovation on the Intel platform was Sandy Bridge micro-architecture which was 2011. Recall that i7 processors from Sandy Bridge delivered 4 Core processors at 3.4Ghz. Aside from Intel's tick-tock model squeezing out some meager improvements, they have been coasting in neutral for a decade. For those of you with a Sandy Bridge based i7 MBP, after slapping an SSD in it, the computer is still surprisingly useful, almost 10 years since it was manufactured.
Things that I am particularly interested in;
If you look at the A13, I bet Apple's Mac silicon will have humungous amounts of cache. The A13 has 8 MB L2 for the 2 big cores, 4 MB L2 for the 4 small cores, and 16 MB of system level cache or L3 if you want to call it that. That's 28 MB of cache for a phone SoC operating at about 5W. The rumored 8+4 CPU for the first Macs could have 4x that: 64 MB of system level cache, 32 MB of L2 cache for 8 big CPU cores and 8 L2 MB of cache for 4 small cores. So about 100 MB of cache for a SoC going into the Apple Silicon version of the MBP13? At 15W?
The unified memory imo will basically be what you see in iPhone and iPad Pro SoCs. There will be one pool of main memory for all of the compute in the SoC, including the GPU. Might have to go to 4 channels of LPDDR4 or LPDDR5 to feed the GPU for a MBP13. The kicker will be whether the DRAM will be in-package or not. Or on-package or not. Not sure they can stack 32 to 64 GB of DRAM on top of the SoC, but I think they are going to really try. A lot of unknowns with what Apple's going to do for expensive Macs and say 256, 512, 2048 GB of RAM, and how the unified memory is going to feed a RDNA2 level GPU, say 32 CPU cores, etc. Some are saying HBM, but it's memory capacity limited.
I think I'll be happy if the storage controller enabled 8 GB/s read and writes. That's going to be the baseline storage performance for expensive PCs in the PCIe 4 era.
Yes to all that dedicated hardware being in the SoC. They are listed in their Apple Silicon art. If it is listed as box in the graphic, it will be a feature of Apple Silicon Macs.