5 ms·
It’s interesting that Apple’s decision to transition to their own chips was probably lead, at least partly, by the fact they knew the could lock down all this n
by jonplackett 5y ago
It’s interesting that Apple’s decision to transition to their own chips was probably lead, at least partly, by the fact they knew the could lock down all this next gen production.
If everyone else had access to cutting edge TSMC I’m sure Apple’s chips would still be good but I don’t think we’d be quite as impressed.
- sudosysgen 5y agoIf everyone had access to cutting edge processes, Apple silicon would be behind. They're not even cleanly ahead despite a generation leap in RAM and in processes in CPUs and they're still behind in GPU performance per watt compared to NVidia on 8nm. If Apple wants to lead in performance they can't allow NVidia and AMD to be on the same process. They would even risk Qualcomm chips catching up.
- tambourine_man 5y agoSource?
- akmarinov 5y agoWell that's a load of bull that you just wrote. Qualcomm uses the exact same process that Apple uses for the 888 and TSMC produces it, yet it's inferior in every way to the A14
- sudosysgen 5y agoThe 888 is produced on Samsung 5nm LPE, not TSMC : https://www.anandtech.com/show/16463/snapdragon-888-vs-exynos-2100-galaxy-s21-ultra https://www.anandtech.com/show/16463/snapdragon-888-vs-exyno... Despite an inferior process, the 888+ and 888 are only 10-20% behind the A14 in most metrics including energy efficiency, despite being on a process with 35% lower density.
- akmarinov 5y agoThat didn’t work out and they switched to TSMC - https://www.notebookcheck.net/TSMC-to-manufacture-some-of-Qualcomm-s-5nm-and-6nm-parts.531207.0.html https://www.notebookcheck.net/TSMC-to-manufacture-some-of-Qu...
- Dylan16807 5y agoI can't find anything saying those chips have reached market, let alone a benchmark. Everything I can find suggests new model numbers too.
- gruturo 5y agoExtraordinary claims require extraordinary evidence.... but I guess we could settle for just ordinary evidence, if you could care to provide some. Because, in its absence, what you wrote doesn't ring at all true.
- sudosysgen 5y agoHow is this the case? AMD mobile CPUs, core for core, are anywhere from 13% slower to 20% faster (between Geekbench and Cinebench r20 and everything in between), and the M1 Max has a generational process advantage and 2-3x faster RAM, despite a similar TDP. In multicore, the Apple processors plainly consume way more power (62W vs 35W sustained) than any of the latest generation AMD processors yet released. Clearly the gap is within to ~25% process advantage and that's without taking into account much faster RAM. Same goes in phones. The A14 is around 10-20% faster in all metrics than an 888 and has around that much of a process advantage.
- wtallis 5y agoYou seem to be comparing the nominal TDP of AMD chips against the actual measured power of Apple chips. It's really not safe to assume that a chip's advertised TDP is well-correlated with real power consumption for any particular workload; a sound comparison requires matched power and performance measurements. This is especially necessary when discussing benchmarks that have a short duration or are a variable workload, because a single TDP number cannot convey turbo behavior.
- sudosysgen 5y agoI am not assuming that. The 5980HS actually consumes only 35W for sustained loads. It can boost up to 62W for 1-2 seconds and then drops to 42W and soon drops to 35W after a maximum of 5 minutes. The SPECInt suite lasts from 1000 to 3000 seconds. Therefore the 5980HS was at 35W for the vast majority of the test.
- wtallis 5y ago
- floatboth 5y ago> generation leap in RAM You mean the exact same LPDDR4X that everyone else uses now? Maybe not in huge quantities but there are a lot of AMD Renoir and Intel Tiger Lake laptops with the same 16GB setup.
- sudosysgen 5y agoThe M1 Max and M1 Pro uses LPDDR5, not LPDDR4X. LPDDR5 is around 6 times faster than LPDDR4X but the CPU can use a bit less than half of that. If you compare the M1 that uses LPDDR4X it loses or ties in multicore and is barely faster in single core.
- gautamcgoel 5y agoWait, how is LPDDR5 6 times faster than LPDDR4X? I thought the bandwidth of DDR5 is roughly 2x that of DDR4 - and that is assuming you run the DDR5 at it's highest rated clockspeed, which no ones does yet since the technology is not fully mature.
- rasz 5y agoI think he meant whole systems. M1 Pro comes with 256bit bus (2 custom LPDDR5 chips) and 200 GB/s, M1 Max with 512 (4 custom LPDDR5 chips) and 400 GB/s. M1 is 128bit bus (2 custom LPDDR4 chips) and 60 GB/s? 400/60 = ~6
- floatboth 5y ago> LPDDR5 is around 6 times faster than LPDDR4X 6 times?! What? No generational improvement in the last decade (if not two) was six-fold. https://www.samsung.com/semiconductor/dram/lpddr5/ https://www.samsung.com/semiconductor/dram/lpddr5/ Samsung marketing says 1.5 times.
- Zigurd 5y agoIt is an integrated strategy: volume + design + supply chain + market power. Apple's CPUs are designed for scaling and binning. The chip designs, along with products like iMac and Mini, can be adjusted to respond to manufacturability and yield. Apple draws a winning hand no matter how the deck is shuffled.
- fundad 5y agoExactly Apple made themselves first because otherwise AMD and Qualcomm would have been hogging the capacity but it’s only unfair if Apple does it.
- sudosysgen 5y agoWhen did AMD and Qualcomm ever exclusively operate a cutting edge node? Apple would have had to deal with reduced supply, but so would every other player.
- fundad 5y agoNot that we know of, for hardware supply. Microsoft can't supply ARM Windows to Qualcomm's competitors. Windows can't be ported to anyone else's ARM chips because of an exclusivity agreement with Microsoft.
- pram 5y agoIt's "interesting" that I never read any posts about how unfair it was when AMD was using TSMC 7nm to beat Intel. Makes you think, it does.
- sudosysgen 5y agoUnlike AMD for 5nm, Intel was free to fab their chips on 7nm at any time and chose not to. It's not an analogous situation.
- smoldesu 5y agoProbably because TSMC 7nm =! Intel 10nm
- Dylan16807 5y agoBecause Intel had full control of their own design and fabrication. And while they had problems advancing their fabs, they also could have bought some TSMC 7nm capacity if they wanted to. That's fair. On top of them earning a loss by sticking with stagnant designs for so long.
- rsynnott 5y ago> It’s interesting that Apple’s decision to transition to their own chips was probably lead, at least partly, by the fact they knew the could lock down all this next gen production. Eh? The vast majority of Apple chips that will be made on this process will go into phones, and Apple has been doing its own phone chips for about a decade. The M1 won't be quite a rounding error, but it won't be far off; they just don't sell that many Macs.