Qualcomm’s new Snapdragon 8 Gen 2 is here, but what does it signal for the next generation of flagships?
The Qualcomm Snapdragon 8 Gen 2 was announced just last week at the company’s tech summit in Hawaii. Qualcomm’s latest chipset features upgraded specifications and is manufactured on the TSMC process, which, if the 8 Plus Gen 1 is any indication, should result in some efficiency gains.
Furthermore, while the company was hesitant to provide in-depth technical details in some areas (such as failing to mention the name of an Adreno or Kryo version), we were still able to run a variety of popular benchmarks on the Snapdragon 8 Gen 2 reference device. These benchmarks help to establish the performance expectations for upcoming flagships in 2023, giving us something to look forward to.
How we benchmarked the Snapdragon 8 Gen 2
We ran one holistic benchmark (AnTuTu), a CPU-centric benchmark (Geekbench), a GPU-centric benchmark (GFXBench), and MLPerf benchmarks on Qualcomm’s Snapdragon 8 Gen 2 reference device. Each benchmark was run three times, and the average of the three results was used. We left the “UI Perf Mode” option enabled by Qualcomm by default.
It effectively forces the benchmarking apps to run on Prime cores in order to achieve a slightly higher score in certain benchmarks, so keep this in mind when reviewing these results. It’s also worth noting that we’ll be rerunning these benchmarks once we get our hands on a commercial device with the Qualcomm Snapdragon 8 Gen 2.
Based on its own testing, Qualcomm provided us with a set of expected benchmark scores. We only used this for reference, and a table with the benchmark scores that Qualcomm expected the reference device to achieve is available at the bottom of this article.
Snapdragon 8 Gen 2 benchmarks overview
- AnTuTu: This is a comprehensive benchmark. AnTuTu measures CPU, GPU, and memory performance while also including abstract tests and, more recently, relatable user experience simulations. The designer’s considerations are weighted in the final score.
- GeekBench: This is a CPU-centric test that includes encryption, compression (text and images), rendering, physics simulations, computer vision, ray tracing, speech recognition, and convolutional neural network inference on images. Specific metrics are provided in the score breakdown. The final score is weighted based on the designer’s considerations, with integer performance (65%), float performance (30%), and cryptography (5%) receiving the most weight.
- GFXBench: This aims to simulate video game graphics rendering using the most recent APIs, including a plethora of onscreen effects and high-quality textures. Vulkan is used in newer tests, while OpenGL ES 3.1 is used in legacy tests. Instead of a weighted score, the outputs are frames during the test and frames per second (the other number divided by the test length).
Aztec Ruins: These are the most computationally intensive tests provided by GFXBench. Top mobile chipsets cannot currently sustain 30 frames per second. The test, in particular, provides extremely high polygon count geometry, hardware tessellation, high-resolution textures, global illumination and plenty of shadow mapping, copious particle effects, and bloom and depth of field effects. The majority of these techniques will put the processor’s shader computing capabilities to the test.
Manhattan ES 3.0/3.1: This test is still relevant because modern games have already arrived at the proposed graphical fidelity and use the same techniques. It has a complex geometry, multiple render targets, reflections (cubic maps), mesh rendering, numerous deferred lighting sources, and bloom and depth of field in a post-processing pass.
- MLPerf Mobile: MLPerf Mobile is an open-source benchmark for testing the performance of mobile AI. MLCommons, a non-profit open engineering consortium, created it to “deliver transparency and a level playing field for comparing ML systems, software, and solutions.” The first iteration of MLPerf Mobile provides an inference-performance benchmark for a few computer vision and natural language processing tasks. For more information, see “MLPerf Mobile Inference Benchmark: Why Mobile AI Benchmarking Is Difficult and What to Do About It.“
Image segmentation: In this test, an input image is partitioned into labelled objects. Self-driving and remote sensing are two common applications. DeepLab v3+ with 2M parameters was used as the reference model, the dataset was ADE20K (512512), and the quality target was 93% of FP32 (0.244 mAP).
Object detection: This test involves drawing bounding boxes around objects and labelling them. Camera input is commonly used in use cases such as hazard detection or traffic analysis while driving. SSD-MobileNet v2 with 17M parameters is the reference model, the dataset is COCO 2017 (300300), and the quality target is 97% of FP32 (54.8% mIoU).
This test involves responding to questions in a colloquial manner. Online search engines are common use cases. MobileBERT with 25M parameters is the reference model, the dataset is mini Squad (Stanford Question Answering Dataset) v1.1 dev, and the quality target is 93% of FP32 (93.98% F1).
See also; Vivo X Note vs Samsung Galaxy S22 Ultra: Which is better
Benchmark results
Snapdragon 8 Gen 2 Benchmark: Antutu
As in previous years, we’re seeing a 10% increase in AnTuTu score with this year’s Snapdragon 8 Gen 2. This is a significant enough improvement to suggest that the Snapdragon 8 Gen 2 is a more powerful chipset than any other Qualcomm chip to date. It’s not quite in line with the 35% faster CPU performance, but AnTuTu is a holistic benchmark, so it may not fully reflect any CPU gains.
Snapdragon 8 Gen 2 Benchmark: Geekbench 5
Geekbench, on the other hand, is a CPU-centric performance metric. We see nearly 30% improvements in multi-core performance, which appears to be on track with Qualcomm’s advertised 35% improvement. Benchmarks may not always reflect the gains that Qualcomm measures, but this is due to measurement differences. Every tool has its own method for calculating scores and testing chipsets, and Geekbench’s method may not always reveal Qualcomm’s improvements. A 30% increase that is reflected in a year-on-year increase is still impressive.
Snapdragon 8 Gen 2 Benchmark: GFXBench
Qualcomm hasn’t revealed much about the Adreno GPU in the Snapdragon 8 Gen 2, so we can only talk about its performance gains. We don’t know how many cores there are, or how frequently they occur, and we don’t even have a version number. This is a change that occurred with the Snapdragon 8 Gen 1, and it is inconvenient when comparing GPUs. It’s much easier to explain differences in the context of version numbers than it is to name the specific chip every time.
However, aside from GFXBench’s T-Rex test, the findings bizarrely show an improvement in graphics performance overall. Considering that this test is low-intensity, other than the fact that it has a lower frame rate, I wouldn’t place much weight in it. It’s entirely possible that it’s just an optimization and that the other, more thorough tests yield much better results. The Snapdragon 8 Gen 1 achieved an average framerate of 179 FPS in the Manhattan test by GFXBench, which renders a 1080p scene offscreen using the OpenGL ES 3.1 API. The Snapdragon 8 Gen 2 achieved 222 FPS in contrast.
Read also; Oppo Find N2 Key Specifications Leaked, May Feature
The Snapdragon 8 Gen 1 had an average framerate of 49 frames per second in the GFXBench Aztec Ruins test, which produces a 1080p scene offscreen and makes use of the Vulkan graphics API. The Snapdragon 8 Gen 2 produced 65 FPS in contrast. Graphics performance has undoubtedly increased, and some of these improvements are significant. That represents a 24% improvement in the Manhattan test and a 44% improvement in the Aztec Ruins Vulkan exam.
See also; Snapdragon 8 Gen 2 — 5 Killer Camera Features
Only a select few top-tier Android games demand a lot of GPU power, but enhanced GPU performance has applications outside of gaming.
| GFXBench Test | Score (in FPS) |
|---|---|
| Manhattan 3.0 Offscreen 1080p | 329-332 |
| Manhattan 3.1 Offscreen 1080p | 224-226 |
| T-Rex Offscreen 1080p | 481-484 |
| Car Chase Offscreen 1080p | 129-130 |
| Aztec Ruins Offscreen 1080p (Normal Tier) | 178-179 |
| Aztec Ruins Vulkan Offscreen 1440p (High Tier) | 65 |
| Aztec Ruins OpenGL Offscreen 1440p (High Tier) | 60 |
MLPerf
Qualcomm has always been particularly tight-lipped about specifics about artificial intelligence advancements. We don’t have any TOPS (Trillion Operations Per Second) figures, but the company has provided us with information on some tangible improvements, such as a 435% increase in artificial intelligence performance and a 65% increase in performance per watt. The results above demonstrate how the Snapdragon 8 Gen 2 performs in AI, and you can compare it to other devices tested by MLCommons.
Conclusion and expected scores on Snapdragon 8 Gen 2 Benchmark
The table that Qualcomm provided us with expected benchmark scores is shown below, and it is most consistent with the results that we obtained above.
| Benchmark | Version | Method | Expected Score Range | |
|---|---|---|---|---|
| System | Geekbench ST | v5.4.4 | Average of 3 iterations | ~1485 – 1495 |
| System | Geekbench MT | v5.4.4 | Average of 3 iterations | ~5050 – 5200 |
| System | AnTuTu | v9.3.0 | 1st run: ~1.27 – 1.28m Avg of 3 iterations: ~1.26m | |
| System | PCMark | v3.0.4061 | Average of 3 iterations | ~18.5 – 18.9k |
| Browser (Chrome v95.0.4638.74 64-bit) | JetStream | v2.0 | Average of 3 iterations | ~167 – 170 |
| Browser | SpeedoMeter | v2.0 | Average of 3 iterations | ~144 – 146 |
| Browser | WebXPRT | v3.0 | Average of 3 iterations | ~219 – 220 |
| Graphics | GFXBench Manhattan 3.0 Offscreen (1080p) (FPS) | v5.0 | Average of 3 iterations | ~329 – 332 FPS |
| Graphics | GFXBench T-Rex – Offscreen (1080p) (FPS) | v5.0 | Average of 3 iterations | ~481 – 484 FPS |
| Graphics | GFXBench Manhattan 3.1 Offscreen (1080p) (FPS) | v5.0 | Average of 3 iterations | ~224 – 226 FPS |
| Graphics | GFXBench Car Chase Offscreen (1080p) ES3.1 (FPS) | v5.0 | Average of 3 iterations | ~129 – 130 FPS |
| Graphics | GFXBench Aztec Ruins Vulkan (High Tier) Offscreen (1440p) (FPS) | v5.0 | Average of 3 iterations | ~60 FPS |
| Graphics | GFXBench Aztec Ruins OpenGL (High Tier) Offscreen (1080p) (FPS) | v5.0 | Average of 3 iterations | ~178 – 179 FPS |
| Graphics | 3DMark Wild Life Unlimited | v2.2.4786 | Average of 3 iterations | 82 |
| Graphics | 3DMark Wild Life Extreme Unlimited | v2.2.4786 | Average of 3 iterations | 23 |
| AI | MLPerf | v2.1 | Image classification: 3915 – 3920 Object detection: 1765 – 1800 V2.0 Image segmentation: 945 – 950 Language understanding: 185 Image classification (offline): 4980 – 5020 |
Qualcomm claims that the first Snapdragon 8 Gen 2 devices will be available by the end of 2022. We’ll be watching to see how the Snapdragon 8 Gen 2 compares to chips like the MediaTek Dimensity 9200. If you’re upgrading from a device at least two years old, the improvements will most likely be noticeable, though the massive gains in AI performance will most likely go unnoticed. When it comes to Qualcomm’s chipsets, companies rarely use AI to its full potential, and the same is likely to be true here.
Qualcomm confirmed that Redmagic, Honor, ZTE, Xiaomi, Meizu, Vivo, Sony, Redmi, OPPO, nubia, Motorola, OnePlus, Sharp, Asus, and iQOO will launch Snapdragon 8 Gen 2-powered devices. We’re excited to test this chipset in a more controlled environment in commercial devices in the future.
