benchmark new smartphones: (Old Thread)

[royb]

Julien,

Did you use Droidfish for this result? I have Droidfish with SF 7 as the engine and I get nowhere near what you see. I have the Snapdragon 820-based Galaxy S7 Edge and after one minute with 1 GB of Hash and 4 threads (supposed to be 4 cores on Snapdragon 820), I see this performance in NPS for the starting position: 1096K nodes per second.

I'm wondering if I've got something set up incorrectly with SF 7 that is causing this low a number compared to the result you posted. Yes, your CPU is Exynos 8-core but some have said performance tests have shown that the two are close and the Snapdragon 820 is even a bit faster in some tests.

Roy

My guess is that SF was running on a single core. Snapdragon 820 should perform close or maybe even above Exynos.

I just put OS Monitor App on my phone. It says Droidfish is using only 1 CPU. So, must be a bug in Droidfish. Perhaps it's due to the new processor and it does not know how to properly set the number of cores to use with the new CPU? Guessing here, but maybe someone here know the author of Droidfish and could have him check into this?


Could it be that despite my setting the number of threads to 4, it was still using a single core? I do not know how to tell on the phone what is really going on when running SF 7 on it. Perhaps others with a Snapdragon 820 phone can test it and then we may learn more about this seemingly strange result.

No way it's 1096K. In fact I hope it's above Exynos.

[Laskos]

Julien,

Did you use Droidfish for this result? I have Droidfish with SF 7 as the engine and I get nowhere near what you see. I have the Snapdragon 820-based Galaxy S7 Edge and after one minute with 1 GB of Hash and 4 threads (supposed to be 4 cores on Snapdragon 820), I see this performance in NPS for the starting position: 1096K nodes per second.

I'm wondering if I've got something set up incorrectly with SF 7 that is causing this low a number compared to the result you posted. Yes, your CPU is Exynos 8-core but some have said performance tests have shown that the two are close and the Snapdragon 820 is even a bit faster in some tests.

Roy

My guess is that SF was running on a single core. Snapdragon 820 should perform close or maybe even above Exynos.

I just put OS Monitor App on my phone. It says Droidfish is using only 1 CPU. So, must be a bug in Droidfish. Perhaps it's due to the new processor and it does not know how to properly set the number of cores to use with the new CPU? Guessing here, but maybe someone here know the author of Droidfish and could have him check into this?


Could it be that despite my setting the number of threads to 4, it was still using a single core? I do not know how to tell on the phone what is really going on when running SF 7 on it. Perhaps others with a Snapdragon 820 phone can test it and then we may learn more about this seemingly strange result.

No way it's 1096K. In fact I hope it's above Exynos.

Well, it indicates that 4 core Snapdragon 820 result can come VERY strong, at something like 3.5Mnps, if there is no huge throttling .

[royb]

Julien,

Did you use Droidfish for this result? I have Droidfish with SF 7 as the engine and I get nowhere near what you see. I have the Snapdragon 820-based Galaxy S7 Edge and after one minute with 1 GB of Hash and 4 threads (supposed to be 4 cores on Snapdragon 820), I see this performance in NPS for the starting position: 1096K nodes per second.

I'm wondering if I've got something set up incorrectly with SF 7 that is causing this low a number compared to the result you posted. Yes, your CPU is Exynos 8-core but some have said performance tests have shown that the two are close and the Snapdragon 820 is even a bit faster in some tests.

Roy

My guess is that SF was running on a single core. Snapdragon 820 should perform close or maybe even above Exynos.

I just put OS Monitor App on my phone. It says Droidfish is using only 1 CPU. So, must be a bug in Droidfish. Perhaps it's due to the new processor and it does not know how to properly set the number of cores to use with the new CPU? Guessing here, but maybe someone here know the author of Droidfish and could have him check into this?


Could it be that despite my setting the number of threads to 4, it was still using a single core? I do not know how to tell on the phone what is really going on when running SF 7 on it. Perhaps others with a Snapdragon 820 phone can test it and then we may learn more about this seemingly strange result.

No way it's 1096K. In fact I hope it's above Exynos.

Well, it indicates that 4 core Snapdragon 820 result can come VERY strong, at something like 3.5Mnps, if there is no huge throttling .

Well, the mystery deepens. I configured Droidfish to use 4 cores and ran OS monitor at the same time and it said that SF 7 used only 1 cpu (100% shown rarely with 98% a common value, and once 100.5% shown). Then, I configured SF 7 to use only 1 thread and the nps figure shown dropped by about almost a factor of 4 (about 3.8 actually).

And, while running with 1 thread, the OS Monitor says I'm using 25% of the CPU; so I've just learned that this particular tool (unlike top on Linux which I'm used to) reports what percentage of the total available CPU you are using (in this case, 100% means you are using all 4 cores). So ... that makes the Exynos CPU look HUGELY faster than the Snapdragon 820, which flies in the face of everything else I've seen reported on the Internet. So, I'm going with something is not quite right with my setup, but I'm not able to see what yet. Powersaving mode is NOT enabled on my phone and the phone is plugged in while running the tests. Perhaps there is a big thermal throttle on the CPU that is kicking in?

[Laskos]

Well, the mystery deepens. I configured Droidfish to use 4 cores and ran OS monitor at the same time and it said that SF 7 used only 1 cpu (100% shown rarely with 98% a common value, and once 100.5% shown). Then, I configured SF 7 to use only 1 thread and the nps figure shown dropped by about almost a factor of 4 (about 3.8 actually).

And, while running with 1 thread, the OS Monitor says I'm using 25% of the CPU; so I've just learned that this particular tool (unlike top on Linux which I'm used to) reports what percentage of the total available CPU you are using (in this case, 100% means you are using all 4 cores). So ... that makes the Exynos CPU look HUGELY faster than the Snapdragon 820, which flies in the face of everything else I've seen reported on the Internet. So, I'm going with something is not quite right with my setup, but I'm not able to see what yet. Powersaving mode is NOT enabled on my phone and the phone is plugged in while running the tests. Perhaps there is a big thermal throttle on the CPU that is kicking in?

Very weird. So your SF on one core runs below 300Knps? This is hardly believable. It seems no large throttling to multicore, as the multiplier is 3.8 on 4 cores, with throttling this could become 2. On single core, do you see a spike to say at least 500Knps for a brief second at the beginning of the run?

[royb]

Well, the mystery deepens. I configured Droidfish to use 4 cores and ran OS monitor at the same time and it said that SF 7 used only 1 cpu (100% shown rarely with 98% a common value, and once 100.5% shown). Then, I configured SF 7 to use only 1 thread and the nps figure shown dropped by about almost a factor of 4 (about 3.8 actually).

And, while running with 1 thread, the OS Monitor says I'm using 25% of the CPU; so I've just learned that this particular tool (unlike top on Linux which I'm used to) reports what percentage of the total available CPU you are using (in this case, 100% means you are using all 4 cores). So ... that makes the Exynos CPU look HUGELY faster than the Snapdragon 820, which flies in the face of everything else I've seen reported on the Internet. So, I'm going with something is not quite right with my setup, but I'm not able to see what yet. Powersaving mode is NOT enabled on my phone and the phone is plugged in while running the tests. Perhaps there is a big thermal throttle on the CPU that is kicking in?

Very weird. So your SF on one core runs below 300Knps? This is hardly believable. It seems no large throttling to multicore, as the multiplier is 3.8 on 4 cores, with throttling this could become 2.

On my phone, SF 7 starts out at 500 knps (for a few seconds that holds up) but after a minute or so, it settles into the mid to high 200's. That seems like it may be some sort of throttling to me. The snapdragon 810 had serious thermal issues. Everything I've read about the 820 indicated the thermal issues were resolved from the 810 and no such issue was present with the 820. This data makes me wonder!

[Laskos]

Well, the mystery deepens. I configured Droidfish to use 4 cores and ran OS monitor at the same time and it said that SF 7 used only 1 cpu (100% shown rarely with 98% a common value, and once 100.5% shown). Then, I configured SF 7 to use only 1 thread and the nps figure shown dropped by about almost a factor of 4 (about 3.8 actually).

And, while running with 1 thread, the OS Monitor says I'm using 25% of the CPU; so I've just learned that this particular tool (unlike top on Linux which I'm used to) reports what percentage of the total available CPU you are using (in this case, 100% means you are using all 4 cores). So ... that makes the Exynos CPU look HUGELY faster than the Snapdragon 820, which flies in the face of everything else I've seen reported on the Internet. So, I'm going with something is not quite right with my setup, but I'm not able to see what yet. Powersaving mode is NOT enabled on my phone and the phone is plugged in while running the tests. Perhaps there is a big thermal throttle on the CPU that is kicking in?

Very weird. So your SF on one core runs below 300Knps? This is hardly believable. It seems no large throttling to multicore, as the multiplier is 3.8 on 4 cores, with throttling this could become 2.

On my phone, SF 7 starts out at 500 knps (for a few seconds that holds up) but after a minute or so, it settles into the mid to high 200's. That seems like it may be some sort of throttling to me. The snapdragon 810 had serious thermal issues. Everything I've read about the 820 indicated the thermal issues were resolved from the 810 and no such issue was present with the 820. This data makes me wonder!

I added this question in the edit to my post. Seems a huge throttling, and from a low value of 500K anyway. I was expecting 800-900K on single core and low throttling. If this holds, it's EXTREMELY disappointing from Snapdragon 820. Or maybe your chip is defective, it seems to contradict all benchmarks on internet.

[royb]

Well, the mystery deepens. I configured Droidfish to use 4 cores and ran OS monitor at the same time and it said that SF 7 used only 1 cpu (100% shown rarely with 98% a common value, and once 100.5% shown). Then, I configured SF 7 to use only 1 thread and the nps figure shown dropped by about almost a factor of 4 (about 3.8 actually).

And, while running with 1 thread, the OS Monitor says I'm using 25% of the CPU; so I've just learned that this particular tool (unlike top on Linux which I'm used to) reports what percentage of the total available CPU you are using (in this case, 100% means you are using all 4 cores). So ... that makes the Exynos CPU look HUGELY faster than the Snapdragon 820, which flies in the face of everything else I've seen reported on the Internet. So, I'm going with something is not quite right with my setup, but I'm not able to see what yet. Powersaving mode is NOT enabled on my phone and the phone is plugged in while running the tests. Perhaps there is a big thermal throttle on the CPU that is kicking in?

Very weird. So your SF on one core runs below 300Knps? This is hardly believable. It seems no large throttling to multicore, as the multiplier is 3.8 on 4 cores, with throttling this could become 2.

On my phone, SF 7 starts out at 500 knps (for a few seconds that holds up) but after a minute or so, it settles into the mid to high 200's. That seems like it may be some sort of throttling to me. The snapdragon 810 had serious thermal issues. Everything I've read about the 820 indicated the thermal issues were resolved from the 810 and no such issue was present with the 820. This data makes me wonder!

I added this question in the edit to my post. Seems a huge throttling, and from a low value of 500K anyway. I was expecting 800-900K on single core and low throttling. If this holds, it's EXTREMELY disappointing from Snapdragon 820. Or maybe your chip is defective, it seems to contradict all benchmarks on internet.

I see in OS Monitor that core 1-4 have something called "Governor" associated with them. The program says this: "Governor: interactive".

This seems an Android OS thing though, not something done in the hardware. Still, sounds like a throttling mechanism to me.

[Laskos]

I see in OS Monitor that core 1-4 have something called "Governor" associated with them. The program says this: "Governor: interactive".

This seems an Android OS thing though, not something done in the hardware. Still, sounds like a throttling mechanism to me.

Can you run Antutu 6.0 benchmark?

[royb]

I see in OS Monitor that core 1-4 have something called "Governor" associated with them. The program says this: "Governor: interactive".

This seems an Android OS thing though, not something done in the hardware. Still, sounds like a throttling mechanism to me.

Can you run Antutu 6.0 benchmark?

I downloaded AnTuTu 6.0.1 and these are the results:

Score: 97504
Device: SM-G935V

3D: 25869
UX: 34883
CPU: 30276
RAM: 6476

I have no idea what those all mean, but that's what the tool gave as output.

[royb]

I see in OS Monitor that core 1-4 have something called "Governor" associated with them. The program says this: "Governor: interactive".

This seems an Android OS thing though, not something done in the hardware. Still, sounds like a throttling mechanism to me.

Can you run Antutu 6.0 benchmark?

I downloaded AnTuTu 6.0.1 and these are the results:

Score: 97504
Device: SM-G935V

3D: 25869
UX: 34883
CPU: 30276
RAM: 6476

I have no idea what those all mean, but that's what the tool gave as output.

OK, I think I found the problem with AnTuTu performance. I had turned on Developer extensions. When I switched them off and re-ran AnTuTu 6.0.1 I got these results:

Score: 134867
Device: SM-G935V


3D: 57356
UX: 40318
CPU: 31106
RAM: 6087


No huge difference though in Droidfish running the SF 7 engine with 4-threads: 1418K nodes per second. I am beginning to think that Droidfish has some issue that's not obvious other than not performing nearly as well as would be expected on this hardware.