Lilbits: A phone that charges in 9 minutes, Qualcomm v ARM, and why the Pixel 7 is 64-bit only (according to Google)

1. 2.5GHz on a Pi4 is pretty cool, seems to take a lot to get it past what you can do with just air cooling. I had 2.3GHz Pi4 running PCem with one of the teeny desktop-style pi coolers and after shrouding it so it actually forced air THROUGH the heatsink fins it was solid. The Pi4 is a relatively zippy little unit if you can cool it enough.

2. 64 bit only Android will become inevitable as ARM started dropping support for 32 bit instructions on its newest cores, but its implementation on the Pixel 7 is completely artificial and I can’t help thinking that it’s not worth the loss of 32 bit compatibility.

3. Typo: “4,300 mAy battery”, should be “4,300 mAh battery”.

4. IF what Qualcomm is claiming is true, and IF Arm are allowed to rewrite all their agreements, that has enormous implications not only for Qualcomm, or even for designers of ARM chips, but for any OEM who wants to marry Arm CPUs with GPUs etc from any other supplier – including all the big fabbing companies. If I’ve understood Qualcomm correctly, everyone would be stuck with Arm’s designs.

   Surely that’s going to cause processor design companies to accelerate the RISC-V contingency plans they must have made when Nvidia was attempting to grasp control of the ARM ISA.

   1. It’s not true, they’re exaggerating.

      The business model that ARM has it that they offer you a full-solution for very cheap. If you want less than that, it means you are allowed freedom to customise further, it has to follow certain protocols, and you need to do this yourself. And you can even go further than that and only license the protocol itself, and make a fully custom processor, and the fees are very high.

      As an example; MediaTek pays the least and gets the full package. Samsung pays a lot and is allowed to have their Mongoose Cores. While Apple pays extremely high and they have their A-line and M-line of processors.

      It’s actually brilliant, because it levels the playfield and increases competition. While ensuring minimal fragmentation. And still giving freedom for those companies who want to do interesting things (which was paramount leading up to their server solutions).

      Here’s why Qualcomm is in the wrong.
      Nuvia is made from ex-Apple engineers, who designed a new core back in 2018-ish that was similar to Apple’s. This was on a Custom License agreement temporary to ARM, who were trying to foster more competition in the market. They were due to release their products around 2019, but instead the company got acquired by Qualcomm. Now Qualcomm used to have a big R&D budget, engineering staff, and they used to pay very handsomely to ARM. They believe that acquiring Nuvia has now given them the Custom License agreement. But the terms of that contract is between the two companies, so had Qualcomm kept Nuvia separately in terms of products and finances, they would have been fine. But Qualcomm has merged them into their core business, so in-effect their is no “Nuvia” anymore. The Custom License Agreement was never open for sale, only the staff and their work was. So basically Qualcomm has to apply and obtain a Custom License Agreement from ARM again, which is costly, and ARM had changed their pricing starting with ARMv9. It may look like ARM is double-dipping here, but it’s not, as apparently Nuvia’s original cost was very cheap (to bolster competition). But being merged into Qualcomm means that competitive reasoning is gone, and there’s no reason Qualcomm should get a steep discount and hurt the market.

      Now, the other allegations they’re making sound preposterous. ARM doesn’t try to block it’s customers from using other GPU, DSP, or NPU solutions. But this might be one of the things that was adjusted in the new ARMv9 platform. To say that ARM has a monopoly in this field is to state the obvious; that’s by design so there is a cohesive ecosystem and not rampant fragmentation like we see in the general RISC processor field. They’re taking their business seriously, and made some notes from their closest competitor (x86) which has very good cohesion.

      Qualcomm’s previous efforts in making Custom Cores ranged from Scorpion-S1 (A8), Scorpion-S2 (A8+), Krait-200 (A9+), Krait-300 (A15), Krait-400 (A15+), Krait-450 (A17), and Kryo-100 (A57). There have also been the Samsung Mongoose lineup, Nvidia Tegra lineup, Apple’s lineup, and many small solutions over the years by server vendors.

      1. Arm appear to be behaving in a very authoritarian manner, and Qualcomm seem to be balking at that.

         I agree that Qualcomm’s latest allegations sound preposterous, but just because Arm hasn’t done things in a certain way to date doesn’t mean to say that they aren’t intending to do so from 2025, and that would have huge remifications across the industry. The judge needs to probe the veracity of this claim, for the sake of all the players in the industry.

         But it’s for the judge alone now to rule on what Qualcomm/Nuvia can and can’t do with the licenses they hold. What seems unequivocal though, is that the relationship between Arm and Qualcomm is sour. Calista Redmond recently announced that Qualcomm have joined the RISC-V Foundation. Arm, it seems, are set to lose [at least] one of its most lucrative customers over the next few years.

         1. Agreed.
            However, there is no way Qualcomm is exiting their current business model (ARM processors for Android phones). It’s what makes them most of their profit.

            RISC-V is vapourware at the moment. A lot of their claims about catching up to the older ARM Designs have not materialised. From the software-front they haven’t really progressed much. On top of this, they seem to already be unravelling and fragmenting. And the most disappoint is that, it can be built on an open-source hardware but vendors can make proprietary customisations and hold it against developers and users. So that’s why Qualcomm cannot and will not abandon the ARM ecosystem… after all, they’ve only recently become competitive with x86 across the board (pocketables, thin/light, mini pc, HEDT and Servers).

            There is AT LEAST a sizeable 5-YEARS for RISC-V to be viable. In my view, we will need to see something like 16nm SMIC built silicon, 4+4 hybrid design, 64-bit RISC-V that’s equivalent to A73+A53. On top of that, it will need to be ported to the main Linux Branches (Debian, RHLE, Arch) and also Vanilla AOSP. That is a good starting point for this to develop into an ecosystem. For it to be competitive, it will need to be something more like 8nm TSMC silicon, 6+6 hybrid design, and 64-bit equivalent to A78+A55. In terms of software, have official running OS from Ubuntu, openSUSE, Manjaro, and an Official AndroidOS with Google Play Services.

            1. In this last statement, Qualcomm said that they believe their Arm licenses are good for a while yet. I agree that, unless Arm does do something drastic in 2025, nobody will be using RISC-V main processors in their flagship devices in the near future – the app ecosystem isn’t there yet. However, I would assume that Qualcomm/Nuvia have already made a start on powerful RISC-V processor designs.

               We know though that Samsung is using SiFive RISC-V chips as co-processors in their kit, and there are rumours that Apple are looking to do something similar. I’d be surprised if Qualcomm doesn’t produce a processor similar to SiFive’s fairly soon.

               But what Arm seemed to be more concerned about recently, was the idea that RISC-V was already making inroads into datacentres – something they’d hoped to get good profits out of for the next decade at least. I would think that Qualcomm/Nuvia have the talent to cash in there too, and they are likely to do it now with RISC-V.

               As for Linux OSs running on RISC-V – Debian and Ubuntu certainly do, and Suse and Fedora do too, if I recall correctly. Android Open Source Project is also running on a T-Head board. And with cheaper RISC-V development boards now shipping in greater numbers, we can expect to see many more apps being rebased too.

            2. China is moving on RISC V chips, they will not kick flagship chips in to the back row, but they don’t need to. They just need to nibble the Arm market away, Bit by bit.

Comments are closed.