|
|
|
|
|
|
by schoolornot
2027 days ago
|
|
|
Apple has a 10 year lead time on Google for running OSes on in-house Silicon. If there is no intention to compete with Qualcomm, how are they going to get the economies of scale problem down to where the chips are affordable + performant? The volumes of Chromebooks/Pixels sold in a week vs. iPhones in a week is stark. |
|
|
Apple is a fabless designer, other vendors have access to TSMC as well, as long as they have space capacity. Realistically, other ARM vendors are about 2 years behind. A Snapdragon 875 scores the same as an Apple A12 on Single Thread Geekbench.
On GPU and TPU, Apple isn't that far ahead at all. TPUs are relatively simple devices, and the Snapdragon 888 already has almost double the A14 performance in TOPS on paper. GPU wise, the A14 (PowerVR derived) GPU is roughly equivalent to a 2016 era NVidia 1060. AMD's RDNA2 based cores for SOCs are likely more powerful and more capable (e.g. raytracing acceleration)
Much is made of how far Qualcomm is behind Apple, but like with AMD and Intel, Qualcomm has a wider market focus to address. Apple sells only a few SKUs, Qualcomm needs to make chips for a much wider array of demands, and so they're a jack of all trades, master of none. In much the same way, AMD and Intel have desktops and enterprise vendors to satisfy, including people running SMP systems. Does the M1 support multiple M1 SMP? Does it support ECC? The other vendors are kind of hamstrung by trying to make one architecture that pleases too many markets at once.
Look at the Anandtech Snapdragon 888 article: https://www.anandtech.com/show/16271/qualcomm-snapdragon-888...
"Qualcomm’s 25% generational boost is also less than Arm’s advertised 30% as the new S888 continues to use a 4MB L3 cache for the CPU cluster, versus Arm’s envisioned 8MB configuration for a high-end 5nm SoC with the new X1 cores. Qualcomm explained to us that this was simply a balance between cost, implementation effort, and diminishing returns of a higher cache configuration design."
Basically, Qualcomm needs to worry about what their customers are willing to pay for the chip and how much work it takes to integrate. Apple doesn't, if need be, they can bump the price for a more expensive SOC, Qualcomm Android vendors can't.
25% CPU performance uplift and 30% GPU performance uplift for the 888. the A14 gets 1583 Single Thread in Geekbench, the 888 will likely turn in a final Single Core perf between 1300-1450, only 9% less than the A14. The situation is the same for multi-core, roughly on par, and on Antutu, the Snapdragon 875 beats the A14 substantially, and wins on several 3dMark tests.
Apple does not have an insurmountable lead over the other vendors in terms of SOC, I'd argue that their primary advantage right now is their software has always been better. In particular, WebKit's JIT for ARM does a lot better than V8 for ARM, and Android Dalvik does not produce code that runs as well as LLVM on Swift (plus there's generational GC vs ARC+refcounting).
Otherwise, a 9% CPU advantage in single thread performance is nothing to write home about.