Which is the faster processor for the Galaxy S8?: Exynos 8895 vs Snapdragon 835
It's that time of the year again. Galaxy S8 will be released into the wild a couple weeks from now, and the old questions will strike again:
- Is Exynos still better than Snapdragon?
- Should I import the international variant of the Galaxy S8?
- S8 exynos vs S8 snapdragon: Which is faster and which one has better battery life?
I decided to wait for some time for things to stabilize and benchmarking sites to have more data for a statistically meaningful comparison.
Disclaimer: S8 is still not released and running on pre-production software. The final software may increase final scores, but the relative difference between these processors should still remain unchanged. If the final software shows some changes, I will update the changes in this article.
Please note that some of the tests will be exclusive simulations based on similar hardware's performance, which will be extrapolated since they are not available yet.
CPU performance
This is one of the most important aspects of a phone, since app install times(compiles), load times(executions) as well as general performance is mostly CPU dependent. Geekbench will be used for this test.
Geekbench scores are compiler dependent, hence cross-platform benchmarking is irrelevant, which is a good reason I won't be including Apple's A10 in the benchmark scores here.
The scores below are the best scores from G955U(Snapdragon) and G955F(exynos) yet from the Geekbench database. So both are at their best here.
The scores below are the best scores from G955U(Snapdragon) and G955F(exynos) yet from the Geekbench database. So both are at their best here.
As we can see, the Exynos 8895 handily beats the SD835 in these tests, which was expected given Samsung's lead in CPU performance for a few years now. However, the Snapdragon 835 is closer in multi-core performance than it has ever been since the Galaxy S5 days.
GPU performance(simulated)
This is the part where it gets tricky. Since GFXBench doesn't update as frequently and show live results as Geekbench browser does, it doesn't have the S8 in its database yet. However, since GPU scores vary linearly with clock speeds and frequency, we can get a good idea of the performance from the devices tested(Huawei mate 9 and Snapdragon 835 reference device).
Huawei Mate 9 uses the same G71 GPU, with 8 cores and 1037MHz frequency on each core. On the other hand, Galaxy S7 uses G71 with 20 cores and 546 MHz on each core. Using a wider implementation means that the Samsung implementation will throttle a lot less and consume less power, while giving higher performance(will cover this in a later article).
Since GPU performance scales linearly across even thousands of cores, we won't have much issues in simulating the performance G71MP20(31.2% higher performance than the huawei mate 9 based on core count and frequency). For adreno 540, we can just take the best score from the development board for SD835.
Enough talk, let's delve into the scores:
T-rex is one of the least stressful benchmarks for these devices. I remember when my S4 struggled to put out 15fps in this test when it came out. We have got 10x performance boost in 4 short years, which is impressive.
Manhattan came out around 3 years ago, and is still a demanding benchmark to run, and these are the only processors to clear 60fps on this test easily.
Adreno GPUs have a lot of compute power, and when the load gets higher, the performance increases compared to other GPUs. I expect to see the difference diminish as we approach newer and harder tests.
As we can see, the difference in performance is diminishing as the load increases.
Both are pretty much neck and neck here in car chase. Either the mali G71 performance tapers off very quickly here, or the Huawei mate 9 is throttling heavily here. GFXBench tests show really heavy throttling in this test for the mate 9.
Android games have closer requirements to T-rex rather than car chase here. So you can expect both GPUs to be faster than needed. But the one with more headroom will stay at lower voltage bins and hence consume less power.
Huawei Mate 9 uses the same G71 GPU, with 8 cores and 1037MHz frequency on each core. On the other hand, Galaxy S7 uses G71 with 20 cores and 546 MHz on each core. Using a wider implementation means that the Samsung implementation will throttle a lot less and consume less power, while giving higher performance(will cover this in a later article).
Since GPU performance scales linearly across even thousands of cores, we won't have much issues in simulating the performance G71MP20(31.2% higher performance than the huawei mate 9 based on core count and frequency). For adreno 540, we can just take the best score from the development board for SD835.
Enough talk, let's delve into the scores:
T-rex is one of the least stressful benchmarks for these devices. I remember when my S4 struggled to put out 15fps in this test when it came out. We have got 10x performance boost in 4 short years, which is impressive.
Manhattan came out around 3 years ago, and is still a demanding benchmark to run, and these are the only processors to clear 60fps on this test easily.
Adreno GPUs have a lot of compute power, and when the load gets higher, the performance increases compared to other GPUs. I expect to see the difference diminish as we approach newer and harder tests.
As we can see, the difference in performance is diminishing as the load increases.
Both are pretty much neck and neck here in car chase. Either the mali G71 performance tapers off very quickly here, or the Huawei mate 9 is throttling heavily here. GFXBench tests show really heavy throttling in this test for the mate 9.
Android games have closer requirements to T-rex rather than car chase here. So you can expect both GPUs to be faster than needed. But the one with more headroom will stay at lower voltage bins and hence consume less power.
Conclusion
As we can see, the Exynos is yet again faster than Snapdragon this year. Battery life results should be similar, but you can expect the Exynos to be a bit better due to Samsung's experience with big.LITTLE and HMP for 3 years now, compared to 1 year for Qualcomm. Samsung also vertically integrates their own tech, so it should hold up better than the Qualcomm variant with time as seen with the S7 series.
However, the difference between the performance of the 2 variants is negligible this time around. So purely based on performance, going for one over the other seems pointless. I would suggest waiting for the battery life and power figures in a few weeks.
Please mention which processor you like more in the comments section.
Comments
Post a Comment