Qualcomm’s processors for laptops and desktops could get a major speed boost next year. While the latest Qualcomm Snapdragon 8cx chips are a lot more powerful than the previous-gen, they still struggle to compete with the latest Intel, AMD, or Apple processors.
But Qualcomm has a plan for catching up. A few years ago the company acquired a startup called Nuvia that was founded by former Apple employees who had worked on that company’s processors. Qualcomm’s first chips based on Nuvia designs could arrive in 2024, and now Kuba Wojciechowski claims to have new details about what to expect.
According to Wojciechowski, the chip is code-named “Hamoa,” and it will support up to 12 cores:
- Up to 8 Performance cores at up to 3.4 GHz
- Up to 4 Efficiency cores at up to 2.5 GHz
Wojciechowski says each block of four cores has 12MB of shared L2 cache and 8MB of L3 cache, plus 12MB of system-level cache and 4MB for graphics.
The chips will allegedly feature the same Adreno 740 graphics as Qualcomm’s Snapdragon 8 Gen 2 processor, with support for DirectX 12 and Vulkan 1.3 graphics APIs, as well as OpenCL and DirectML. And Qualcomm’s highest-performance 12-core chip will also support discrete graphics thanks to 8 lanes of PCIe 4.0.
Other features are expected to include support for up to 64GB of 8-channel LPDDR5x memory, support for PCIe 4.0 NVMe and/or UFS 4.0 storage. There’s also support for WiFi 7, a Qualcomm Hexagon Tensor Processor with up to 45 TOPS of AI performance, and support for up to three Thunderbolt 4 ports and two USB 3.1 ports with support for 10 Gbps speeds.
According to Wojciechowski, the chip can decode video at up to 4K/120fps resolutions and encode 4K/60fps video and it can drive up to three displays including one 5K display and two 4K displays.
It’s probably best to take this all with a few grains of salt. Not only is it possible that some of the information in the leak could be inaccurate, but it’s also possible that Qualcomm’s plans could change in the coming year. But it sounds like Qualcomm’s chips for Windows and ChromeOS devices could be on track to deliver big performance gains in the next year or two… although how they’ll be competing with next-gen chips from Apple, Intel, and AMD when they arrive.
Leak: update on Qualcomm’s Apple M rival – 8cx Gen 4, codename Hamoa 🧵 pic.twitter.com/1tptCk8ghf
— Kuba Wojciechowski⚡ (@Za_Raczke) January 20, 2023
A 12 core Qualcomm, even with ARM stock cores, would at least perform better than M1 and M2, while being much better in power draw than the best of Intel and ARM. Adreno 740 is supposed to be, even in the Smartphone version, better than a GTX 1050. A slightly overclocked, at least with a fan, version of this chip could perform as good as a Radeon 680M, and it’s still better than anything Intel has to offer.
The problem is they take TOO DAMN LONG! Cortex X3 was introduced 6 months ago! And they are thinking of releasing this by end of this year, or beginning of next, when Intel will present 14th gen, and AMD 8000 series chips, and Apple had introduced the M3 already, and at 3nm.
Qualcomm is stuck at perpetually being behind the competition if they keep this way.
ARM stock cores do not perform better than M1, let alone M2, and Apple hasn’t released an M3 yet. It takes quite a bit of work to go from a phone silicon to a laptop or desktop type silicon. Yes, Qualcomm is taking its sweet time but the volume numbers need to stack up. When Apple put an M2 chip in the iPad Pros, MacBook Airs, MBP13s and Mac minis, they ensure tens of millions of installs. Qualcomm cannot target these sorts of volumes to recoup the investment, hence they’re behind.
I’d like to say they could start to if they made the chips systemready so they’d be easy to build things with that could easily have operating systems adapted to, but I’m sure there could always be some reason why it wouldn’t that I don’t know about and I’m just wishfully thinking because I’d like to have the battery life and low fan noise without having to give up on operating system choice, which is more important to me.
Nope.
Having something “SystemReady” does not ensure it is good, in terms of performance and battery life. Don’t get me wrong, it’s nice to have standardisation. But we could be looking at an ARM-book with 12x Cortex-A515 cores for all we know… yeah, it’s pretty slow for a laptop.
Apple has always been competitive when it comes to chipset. Sometimes they were behind, but since iOS was significantly more optimised than Android, it didn’t matter. With the iPhone 7, they began to pull ahead, and Android has been behind by 1-2 generations ever since. The latest ARM Cortex-A715 has barely caught up to the Apple A13-Bionic Large-Core in performance. The next family is the A14/A15/A16 which had small iterative upgrades on top of it.
The Cortex-X1/X2/X3 are good options for laptops, but they’re trading battery drain for performance. Whereas the Apple M1 does so at a better balance. There are alternative Large Cores from Qualcomm (Kyro) that is obsolete, from Samsung (Mongoose) which were inferior, from Nvidia (Carmel) which is discontinued, and obviously Nuvia (Phoenix) which has been altered and acquired.
Now, I believe we could be looking for an optimistic future. With the likes of TSMC-3nm, Cortex-A730 and Cortex-X4, MediaTek Dimensity in the flagship market, as well as Qualcomm releasing Nuvia-cores. We should see the field in 2024 catch up to Apple’s 2021 benchmark, and perhaps see a whole new level created by Apple’s lead in 2024 as well.
ARM does a lot of partnering with fabs to get their IP on the latest node, but Apple IP and SOCs have always been the first high-volume products on TSMC’s latest nodes. The best process node with the best fab support allows them to always be ahead of ARM (and Qualcomm). Maybe Nuvia allows Qualcomm to be a fast-follower and be aware of TSMC process details that Apple exploits. I am not aware of Nuvia’s secret sauce.
But even if Apple and ARM designs were equal, Apple is always on the “real” latest process which should allow for the highest performance.
@riddick
It’s not so much about the silicon, as it is about mathematics. By using more equations, and more advanced equations, you can estimate calculations faster, or solve for them. This requires a wide-bandwidth and fast memory. And that’s why Apple leads here, basically their “architecture”.
For example, the iPhone Xmas was launched in Q3 2018 against the Samsung S10+ in Q1 2019. They were on the same TSMC-7nm node. However, Samsung was using the QSD 855 chipset, which was revolutionary for its day, against the Apple A12-Bionic. The Cortex-A55 was a whopping 4-generations behind the Apple Tesmpest cores. And the Cortex-A76 was a solid 2-generations behind the Apple Vortex cores.
But it’s not just about the hardware, software plays a major role. Android runs mostly at a higher-level to ensure compatibility with more silicon and more device types. Whereas iOS is semi-native which means code runs lower and closer to the silicon (a la Metal), but it’s not as cross-compatible.
Again, the lithography is important because it can mean less voltage leaking (battery life), higher frequencies and more transistors (performance) but these are iterative improvements measured in the 1% to 49% range typically. Whereas the architectural improvements (ie Rapid Maths) and we’re seeing orders of improvements from x2 to x100 typically. A classic example is the leap from idTech 2 to the idTech 3 Source Engine. But advancement in the latter is harder and slower as time goes on, it requires actual Doctorates, Professors, Mathematicians to come up with newer and better algorithms for calculation, whilst the lithography side of the business has been focussed on cutting silicon sharper and sharper for the past 30 years.
Eventually we’re going to hit a wall, with hardware limitations as electrons can’t pass through silicon anymore, but we still have at least another 10 years of runway left with Silicon. We may see promises from Quantum Computing, and we may want to make the leap there, using lasers for information rather than electrons. And perhaps using the charge of them (-/0/+) to have three data states, so ternary instead of binary. And we would use trits instead of bits, or trytes instead of bytes. And instead of a 2^64 system (64-bit), we could see a very symmetrical 3^81 system (81-trit), which would have a lot more permutations allowed.
Anyways just food for thought.
As excited as I am for faster ARM chips Qualcomm need to improve their drivers. I’ve got a project volterra with their flagship 8cx gen 3 however my gpu drivers crash under load. They’ve been this way for 2-3 months now but as far as I know no fix is available, although the thinkpad x13s patch notes imply it’s fixed.
I’m surprised they are as good as the way they are, since they’re so locked down and proprietary. Whilst the Mali-GPU which is basically a Market Standard has so many driver issues.
One thing I will say, is that the Adreno 740 iGPU is competitive against Apple’s iGPU. Both are cutting edge. AMD and Nvidia offer more and faster calculations, but at a significantly more power draw.
The only question remains is the CPU. If Qualcomm uses the Cortex-X3 or a derivative, they are not going to match the Apple M1. The Nuvia cores WOULD-HAVE been competitive if they were released on-time a few years ago, but currently they’re a mystery (no current leaks). They could be worse than ARM’s designs for all we know.
The other part of the equation is, even if they hit it out of the park, and have the QC 8cx4 be better than the Apple M1/M2… that doesn’t guarantee a great product. You still need the drivers, the operating system, and the applications. Currently I don’t like how Windows 11S compares to macOS. And even if you’re agnostic about it, you still need the OEMs to have good build materials, parts, and QC speaking holistic about the end-user device (Razer, ASUS, Lenovo are good at this, not so much for MS, HP, Acer, etc etc).
It’s interesting, in the light of the current Arm-Qualcomm dispute, that Qualcomm is continuing its efforts with Nuvia designs, and intends to sell the processors (Arm is insisting that Qualcomm destroy Nuvia designs). Obviously we don’t know how long the court case will last, nor can we know the outcome, but there could potentially be a large number of Nuvia-designed Qualcomm (ARM ISA) chips in tech by the time there’s a conclusion…
“In the Qualcomm vs. Arm case, discovery is set to start on January 13th, 2023, and the trial on September 23rd, 2024”
https://www.rcrwireless.com/20221222/analyst-angle/qualcomm-vs-arm-court-dates-announced-who-has-the-luxury-of-time-will-win
I find it ironic that Qualcomm would fight against ARM over licensing issues when they made so much money off licensing in the past. Maybe that puts them at an advantage in the court case, as they can anticipate what ARM is going to do.