At its SnapdragonSummit in Hawaii, Qualcomm has announced its next-generation smartphone hardware platform, with a significant naming rebrand: the Snapdragon 8 Gen 1.
Qualcomm has detailed the name changes on Nov 22, 2021, but if you missed it, “Snapdragon” is now a standalone brand separated from the Qualcomm brand. This change might make Snapdragon even more consumer-friendly and focus people’s attention on the platform and not the company. However, the chip naming is the more important part.
Instead of using numbers such as Snapdragon 888 (last year’s model), the new Snapdragon is called “Snapdragon 8 Gen1”. You can expect lower-tiers Snapdragon chips to follow the same naming scheme, and this makes the differentiation between “Tiers” (from low to high-end) and “Generation” (which year). In this example, “8” is the highest tier for smartphone chips, and Gen1 designates the 2022 generation.
The Snapdragon 8 Gen 1 chip is manufactured with Samsung’s 4nm semiconductor process, one of the most advanced available, but there’s stiff competition from TSMC as well.
*all photo credits: Eliane Fiolet, co-founder, Ubergizmo.
Snapdragon Sight: massive camera computing improvements
Let’s start with the most exciting aspect of Snapdragon 8 Gen 1 for consumers: the camera.
All the camera features are now grouped under the same roof, called “Snapdragon Sight,” so expect us to reference this more often in the future. The name is a change in communication and not a new hardware unit. It is part of the general trend at Qualcomm to make things less technical and more purpose-driven.
The raw numbers are dizzying: 240 photos per second, 3.2 Gigapixel/sec, 8K HDR 10+ video recording, and a new 18-bit triple image signal processor. With Snapdragon 8 Gen 1, Qualcomm re-affirms its commitment to mobile photography, now the #1 concern for smartphone buyers.
There’s so much that we’ll publish more details in a dedicated Snapdragon Sight article, but in the context of this platform overview let’s focus on the hardware potential first. The numbers above are primarily enabled by the new triple 18-bit ISP, and these specs reveal the overarching theme for the new camera architecture is: optimize optical flow.
Optical flow can be thought of as the physical quantity of light (= image data) gathered through the lens. Using a long exposure, having a larger sensor, having a larger aperture all increase optical flow.
However, not all this physical optical flow can be exploited digitally. There are tradeoffs between long exposure and blur or over-exposition. It means that the electro-optics (sensor+lens) are not always fully utilized.
The Snapdragon 8 Gen 1 camera architecture is built to maximize the camera’s electro-optics hardware and make the most of what that hardware can capture. Electro-optics are limited mainly by physics, which is why OEMs are in a camera hardware/specs arms race, as measured by our CAMERA HW benchmark.
“MAXIMIZES OPTICAL FLOW”
However, what’s captured by the electro-optics could be better utilized if you have the necessary compute capabilities (and know-how). For instance, this new Snapdragon maximizes optical flow by making it possible to use almost 100% of the captured light (as spliced HDR staggered photography) to levels never seen before.
We expect this to translate into significantly better HDR photography, which can capture extremely high contrast scenes in a life-like manner without losing details in dark regions or over-exposing things in bright areas.
Snapdragon 8 Gen 1’s ISP is powerful enough to handle 8K HDR10+ videos with three exposures per frame. That’s an insane imaging performance level supported by the rest of the Snapdragon 8 Gen 1 architecture (AI, GPU, CPU, etc.).
For still photography, a single HDR night photo can now be composed of 5X more frames, taken at various ISO to form a full 18-bit HDR RAW image that is then saved as is or as a .jpg file. If subjects move during a multi-frame photo, there’s hardware motion detection to help the camera deal with it at a much higher speed and power-efficiency.
With this level of performance, you’ll clearly be limited by the sensor’s performance, not by the image computation speed. That’s why it’s more important than ever to be aware of your electro-optics camera hardware capabilities (we can help because post-capture computing can magnify the quality of the incoming data.
The key takeaway is that Snapdragon 8 Gen 1 is a computational imaging Godzilla, and I can’t wait to see what kind of cameras systems OEMs will build. You can be assured that whatever data comes in, this platform will make the best of it.
This is the usual hot topic for any new generation of processors, and, of course, Qualcomm has improved all aspects of system performance.
Kryo CPU cluster
The new Kryo CPU cluster is now based on one ARM Cortex X2 (Prime, 3Ghz), three Cortex A710 (2.5Ghz), and four Cortex A510 (1.8GHz). The updated ARMv9 CPU core architecture drives most of the performance boost, but the slight increase in CPU frequency also brings an extra nudge.
Overall, Qualcomm estimates this new CPU cluster to be 20% faster and 30% more power-efficient than Snapdragon 888, the previous Android market leader. The 30% power-efficiency number is surprisingly high, and without further explanations, I’ll assume that it is based on a best-case scenario.
There’s also 50% more L3 cache that is now 6MB large to avoid expensive DDR memory access whenever possible. The memory system itself seems to remain unchanged.
Graphics and gaming
The new Adreno graphics processing unit (GPU) is said to be 30% faster while being 25% more power-efficient, which raises the baseline performance for nearly all apps that depend on the GPU, like games.
But Qualcomm goes farther if developers are willing to put some effort into optimizations. For example, the Variable Rate Shading Pro (VRS Pro) feature allows controlling the Variable Shading using an image-based input. VRS means decreasing shading quality (to boost FPS) in places where gamers won’t notice. Using image-based VRS gives even more control to developers who know best “where” to apply VRS in their games.
Another way to enjoy high FPS gaming while saving battery is the new Adreno Frame Motion Engine. As you can imagine, a 60 FPS game typically consumes twice the battery compared to a 30FPS one because it computes 2X the number of game frames.
The new Adreno Frame Motion Engine is a frame interpolator that can make 30FPS games appear as smooth as 60FPS ones. It seems to work similarly to what 120/240Hz televisions were doing by creating additional 2D frames via an intelligent interpolation, which is much less energy-intensive than actually rendering a 3D scene.
Finally, there’s Snapdragon Studios, a program under which Qualcomm works closely with developers by providing code samples, libraries, analysis/advice, or perhaps implementing hardware-specific optimizations.
Developers often can’t dive as deep into a vendor’s hardware because engineering resources (manpower) are scarce, or the required knowledge is proprietary. Qualcomm’s help can tip the scales on high-profile projects and land Snapdragon users sizeable performance perks.
“COLOSSAL COMPOUND POTENTIAL PERFORMANCE BOOSTS FROM HARDWARE AND SOFTWARE IMPROVEMENTS”
Qualcomm mentioned great examples of working with Epic to optimize well-known techniques, potentially widely used, in Unreal Engine. Using Snapdragon-specific optimizations, staples techniques such as screen-space ambient occlusion (SSAO) or temporal Anti-Aliasing (AA) got a 50%-60% speed-up on the same hardware.
Qualcomm and Epic are now collaborating on Unreal Engine 5 optimizations.
When you add up everything we talked about, you can see colossal compound potential performance boosts from hardware and software improvements.
Although not completely new is a brilliant approach because if Qualcomm’s competitors lack the resources, organization, and know-how to work at that level with developers, even having slightly better peak performance might not be enough to beat Snapdragon in actual games. In short, Qualcomm is digging a moat that helps keeping competitors at bay.
Artificial Intelligence speed-up
The AI performance saw a considerable jump, primarily due to the doubling of the computing capacity of the Hexagon processor but also thanks to the addition of 2X more memory to store neural networks without having to splice and reload them.
Most AI benchmarks should show nearly double the performance (1.7X according to Qualcomm), but it seems that in some cases, loading and unloading the neural network could cut the performance by half. That’s why doubling the AI memory can help tremendously and push the gains to 4X.
Artificial Intelligence (AI)
Performance aside, it is clear that AI keeps permeating in all aspects of computing. Wherever there’s a need for “perception” (recognize, understand, select, etc.) or tasks (denoising, de-compressing video, etc.) based on pattern recognition, AI can potentially help.
Qualcomm displayed great technology demos, such as a more power-efficient AI video decoding. Currently, this technique is experimental, but the general idea is that you can sometimes replace complex pieces of computing with more power-efficient AI computing. AI would produce something that is not always identical but “close enough” that people won’t notice.
For example, that’s how GPU Ray-Tracing denoising works on PC, and one has to admit that these techniques are awe-inspiring and unbelievably effective.
Another AI example was the Natural Language Processing demo that enables next-gen smart assistants. A Good example is to surface the “most important notifications” since we’re bombarded. I’d love to see an actual implementation as there could be many corner-cases to this concept.
AI was previously used in short-bursts, but with workloads related to imaging (including video recording), the sustained throughput and performance now seems unbounded, so expect the AI processor to continue to balloon in size and compute capacity for years to come.
There’s a 4th ISP (image signal processing) for “Always-on AI” in the Sensing Hub. The Sensing Hub is an ultra-low unit that always stays on and listens for your “OK Google” command, for example. With this ISP, the phone can always “see” if you’re looking at the phone or if someone else is peeking at your screen from behind you – without killing battery life.
The all-seeing ISP opens the door for much-heightened security and privacy as one could eventually hide notifications or lock apps if someone else is looking at the screen. This is one of the most helpful features that’s simply impossible to implement without the Sensing Hub’s ISP.
5G is perhaps the closest to the heart for Qualcomm, which has the most sophisticated broadband technology in the Premium segment. Everything presented was leading-edge, and of course, the latest WIFI-6E and Bluetooth protocols are included.
The 10 Gbps (theoretical) speed of Qualcomm’s X65 5G Rel16 modem and RF solution was announced earlier this year but is now integrated into the Snapdragon 8 Gen 1 chip. Qualcomm just revealed that it could use both sub-6 and mmWave protocols in the same session to aggregate the bandwidth.
“MAXIMIZING YOUR 5G BROADBAND CONNECTION AT ALL TIMES”
Speed aside, you’re essentially maximizing your 5G broadband connection at all times if your carrier has both sub-6 and mmWave support. Sub-6 brings better average networks when compared to 4G LTE, but mmWave is 19X faster (than sub-6) in the real world, so this is a significant development.
This year, Qualcomm is communicating more on uploads speed, which makes a lot of sense, given the rise in real-time communications and user-created content. The 8K video call with Verizon was neat and shows how far mobile broadband has come in the past five years.
The Snapdragon platform has integrated a dedicated Security Processor for some time, but there’s a new Trust Management Engine that increases security and unlocks new possibilities such as minting NFTs from the device and more.
The general idea is that any content creation could be authenticated as unique, original, and belonging to the user. The Qualcomm cloud services support the validation, and from there, a record can be added in a blockchain (like ETH), creating a new NFT, which can make its way to an NFT market.
“SAFER TO LOSE YOUR DIGITAL KEYS LOCKED BY YOUR PHONE THAN YOUR PHYSICAL KEYS”
That would be much simpler than current NFT workflows, but of course, it’s not always easy to create artwork on a mobile. And perhaps that’s why this feature will, without a doubt, make its way to Snapdragon PCs.
Security is often an obscure topic hidden from view, but with Android Ready SE and platforms like Snapdragon, it will be possible and more secure to digitize things like keys, government IDs, and more.
In many ways, it would be much safer to lose your digital keys locked by your phone than your physical keys. The same is valid for IDs as their issuance might be heavily validated and potentially verifiable in real-time.
Snapdragon 8 Gen 1 looks great, and on paper, it has everything it needs to remain the top hardware platform for Android smartphones. I like the advances in image computing and overall performance and security.
What Qualcomm is doing on the gaming front makes a lot of sense, and I predict that these developer relations efforts will be very successful in terms of adoption by game developers. You’ll notice that Ray-Tracing (RT) was not mentioned at all, but while I think that RT for mobile is great for developers, it won’t be ready for actual games for another year or two at best.
“IT HAS EVERYTHING IT NEEDS TO REMAIN THE TOP HARDWARE PLATFORM FOR ANDROID SMARTPHONES”
The 5G connectivity remains unequaled and is pretty much the best available, perhaps until next year. That’s always a reliable way to future-proof a multi-year phone purchase because networks are evolving very fast.
Finally, the camera support is phenomenal, especially for 8K video, and multi-camera recording. Single-camera photography remains limited by the lens and sensor, but every bit of captured light can be utilized to its fullest.
Expect broad support from phone makers, and from the looks of it, the first device will arrive before the end of the year. Stay tuned!
Filed in. Read more about Android, Editorspick, Processors, Qualcomm, Snapdragon, Snapdragon Summit and SoC.
Source by www.ubergizmo.com