french toast
Veteran
Thermal throttling!?? I thought these phones had 'liquid cooling' :smile: also I've seen other videos from LBS on YouTube which do a crude throttling check and the 820 in the mi5 throttles less through certain benchmark runs than the exynos 8890 with its cooling setup.
Samsung might have gimped the s820 for some reason or it may need some optimisation, but one thing is clear, in other phones running the same processor it's real world day to day usage is barely any faster than last year's processors (outside graphics), which has to be considered a fail when you consider the new manufacturing process, I really do wish Qualcomm had gone for a cortex a72 setup, hexacore with 2 a72s @2.5-2.7 GHz and 4 a53s @ 1.8-2.0 GHz, in a smartphone form factor I actually think such a setup would be perfect on 14/16nm.
I also could see another array of a35s possibly clocked @ 1.2ghz in a setup similar to media techs upcoming x20, but the scheduler would have to be pretty good I imagine to see real benefits of using so many cores, whether so many threads used concurrently would improve performance on android remains to be seen, but I could see legitimate power saving scenarios of using each cluster for its optimised workload.
I do however like what Samsung has done with the exynos 8890, in boosting 2 cores to 2.6 GHz in certain workloads, I wonder if cortex a72 could be setup in such a way? Or is that a result of the special circuity in the M1 cores?
Back on topic, is this performance stumbling block of kryo a result of its unbalanced design towards floating point performance over INT? I like what they have done with the memory performance, trying to keep the resources fed, just wish they had gone as wide as something like a a57/a72, even if it meant dropping the clock speed and a slight regression in floating point performance.
Samsung might have gimped the s820 for some reason or it may need some optimisation, but one thing is clear, in other phones running the same processor it's real world day to day usage is barely any faster than last year's processors (outside graphics), which has to be considered a fail when you consider the new manufacturing process, I really do wish Qualcomm had gone for a cortex a72 setup, hexacore with 2 a72s @2.5-2.7 GHz and 4 a53s @ 1.8-2.0 GHz, in a smartphone form factor I actually think such a setup would be perfect on 14/16nm.
I also could see another array of a35s possibly clocked @ 1.2ghz in a setup similar to media techs upcoming x20, but the scheduler would have to be pretty good I imagine to see real benefits of using so many cores, whether so many threads used concurrently would improve performance on android remains to be seen, but I could see legitimate power saving scenarios of using each cluster for its optimised workload.
I do however like what Samsung has done with the exynos 8890, in boosting 2 cores to 2.6 GHz in certain workloads, I wonder if cortex a72 could be setup in such a way? Or is that a result of the special circuity in the M1 cores?
Back on topic, is this performance stumbling block of kryo a result of its unbalanced design towards floating point performance over INT? I like what they have done with the memory performance, trying to keep the resources fed, just wish they had gone as wide as something like a a57/a72, even if it meant dropping the clock speed and a slight regression in floating point performance.