Povray on ARM

[jhu]

Updated with Povray 3.7. Now back to GTA V...

 

[soccerballtux]

How much faster do you think the 8974 in the N5 would be pixel/clock?
Has about 50% more memory bandwidth.

I'm honestly a little surprised at the APQ8064's performance in the N4. E.g. that an A9 architecture (exynos 4210 bests it)

 

[jhu]

How much faster do you think the 8974 in the N5 would be pixel/clock?
Has about 50% more memory bandwidth.

I would imagine it should be faster since the code is branchy and both it and the data are too large to fit anywhere but main memory on these mobile devices.

I'm honestly a little surprised at the APQ8064's performance in the N4. E.g. that an A9 architecture (exynos 4210 bests it)

I am too. OTOH, my Nook Tablet (Cortex A9) has a lower resolution screen than my Nexua 4, yet feels a lot slower than the Nexus 4. GPU is also important for these devices.

 

[soccerballtux]

I would imagine it should be faster since the code is branchy and both it and the data are too large to fit anywhere but main memory on these mobile devices.

how much would 50% better get?
I guess we'll have to wait until you review one!

I am too. OTOH, my Nook Tablet (Cortex A9) has a lower resolution screen than my Nexua 4, yet feels a lot slower than the Nexus 4. GPU is also important for these devices.

not all A9s are alike. The Exynos was known to be a "super good version" E.g. the TI 4430 is a terrible A9 IMO. edit: oh, it looks ok here; maybe it's the android benchmarks/was poorly optimized.
Or, someone took time to commit performance stuff that can take advantage of 4210?
or maybe this is just FPU performance or something.

 

[jhu]

Just got an FX 8350 to replace my 1090T. So, IPC really has more or less caught up with Phenom II. Total system idle power draw is lower (69W vs 79W), full load power draw is higher (255W vs. 216W), but total power used is lower (33660J vs. 40791J). Interestingly, compiling with avx (-mavx) gives the same performance as without avx (I was expecting less performance).

Updated with FX 8350, more ARM, and Pentium 4 results.

 

[lamedude]

Try -mprefer-avx128.

 

[jhu]

how much would 50% better get?
I guess we'll have to wait until you review one!

Just got my Galaxy S5 (AT&T), which has a somewhat similar processor as the Nexus 5 (Snapdragon 801 vs. Snapdragon 800, both have Krait 400 cores though). Raw CPU performance of Qualcomm's chips are lower than Samsung's Exynos.

I've also just discovered that, despite not having root access, I can still run Povray on the phone.

 

[Nothingness]

Just got my Galaxy S5 (AT&T), which has a somewhat similar processor as the Nexus 5 (Snapdragon 801 vs. Snapdragon 800, both have Krait 400 cores though). Raw CPU performance of Qualcomm's chips are lower than Samsung's Exynos.

I'm surprised by how bad S80x is on this benchmark.

Can you provide me with the ARM Android binary so that I play with it?

And again thanks for keeping this up

 

[jhu]

I'm surprised by how bad S80x is on this benchmark.

Can you provide me with the ARM Android binary so that I play with it?

And again thanks for keeping this up

Here you go. Just extract and run:

./povray benchmark

If you ever feel like compiling it yourself, you cat get the source here.

 

[jhu]

Someone pointed out that Snapdragon processor results may low due to thermal throttling. So I reran Povray with 1 thread and set the CPU clock speeds at about 1 GHz. Sure enough, the per core results are much better.

 

[sm625]

After two years, where do the latest 2.5GHz smartphone chips place on the OP's chart?

 

[soccerballtux]

Someone pointed out that Snapdragon processor results may low due to thermal throttling. So I reran Povray with 1 thread and set the CPU clock speeds at about 1 GHz. Sure enough, the per core results are much better.

you can use software to adjust the thermal throttle points and the S5 throttles pretty quickly due to the plastic body. The CPU is usually throttled conservatively; I go almost exclusively off the battery with a catch at 75C. I don't game and it's rare that the CPU gets to 75C on my Nexus5.

Anandtech put the S5 in water to bypass the thermal throttling on their benchmarks. You don't have to do that...but...might as well put the IP67 to use??? :whiste:
http://www.anandtech.com/show/7942/galaxy-s5-followup

 

[soccerballtux]

After two years, where do the latest 2.5GHz smartphone chips place on the OP's chart?

look at the snapdragon 801.

this is the most exciting thread to me in years

 

[jhu]

you can use software to adjust the thermal throttle points and the S5 throttles pretty quickly due to the plastic body. The CPU is usually throttled conservatively; I go almost exclusively off the battery with a catch at 75C. I don't game and it's rare that the CPU gets to 75C on my Nexus5.

Anandtech put the S5 in water to bypass the thermal throttling on their benchmarks. You don't have to do that...but...might as well put the IP67 to use??? :whiste:
http://www.anandtech.com/show/7942/galaxy-s5-followup

Wow, never thought about putting it under water. Oh well. I returned the S5 because it couldn't be rooted and got an HTC One M8 instead. And it was cheaper!

 

[jhu]

look at the snapdragon 801.

this is the most exciting thread to me in years

Download the binary and play around with it. If you want certain compiler flags, let me know (gcc 4.6).

 

[soccerballtux]

oh I guess I need to compile it myself if I want to try diff architecture flags!

edit: nvmd. that gets way too hot way too fast. Can't even benchmark without throttling. Tell me mister anderson...

now I must mod my phone to have a heatsink...

 

[henriok]

Nice chart. But I guess that the G4 PowerPC is "7400" and not "4700" since the first exists and the latter don't.

 

[jhu]

Nice chart. But I guess that the G4 PowerPC is "7400" and not "4700" since the first exists and the latter don't.

You're right. Updated.

Also added Haswell i5. Ivy Bridge is slightly faster than Sandy Bridge. But I'm impressed that 1 thread on Haswell is within 5% of Ivy Bridge + HT. Wow!