Samsung Orion SoC - dual-core A9 + "5 times the 3D graphics performance"

[tangey]

ipad2 isn't "Android Smarthpone Class".
It's in a tablet and it's not Android.

Soc graphics results under discussion is from a tablet. the GPU in the ipad2 soc beats it.

French Toast suggest its the fastest smartphone GPU in its class. The iphone4S GPU comfortably beats it, and running 20% slower than in the ipad2.

If one wants to segment mobile GPUs based on the mobile OS they run, then fair enough, However I'm not sure it gives any technical insight to do so, when there is a test such as GLbenchmark that allows cross platform comparisions.

Certainly Samsung would be aiming their high end smartphones to buyers that might consider purchasing an iphone, and as such they should be considered in the same "class"

 

[Ailuros]

http://www.glbenchmark.com/phonedetails.jsp?D=MagicLAB+MagicLEGO4x12&benchmark=glpro21

Looks like this might be the first outing for a 4212 (Mali400mp4@400MHz).

Although it gets twice the performance of the MagicLego 4210......

http://www.glbenchmark.com/compare.jsp?benchmark=glpro21&showhide=true&certified_only=1&D1=MagicLAB%20MagicLEGO4210&D2=MagicLAB%20MagicLEGO4x12

....the magicLego4210 had very poor numbers to begin with.

compared to a recent test of the samsung GT-P6800 (4210), there is about 30% improvment in egypt offscreen and 25% improvement in pro offscreen, so no 50% overall improvement that you might expect from a 50% clock increase, and definitely nowhere near "x2" in the marketing.


fill rate shows about 75% improvement to 775K tex/s. Still below ipad2 raw fill rate of 1G tex/s @250Mhz, nevermind ipad3, and excluding savings due to overdraw removal on ipad.

How certain are the 400MHz for the Mali400MP4 in the MagicLab 4x12 exactly? The GalaxyS2 scores 55.2fps in 720p Egypt and compared to that it doesn't sound like anywhere near 400MHz. In fact with linear frequency scaling and the GalaxyS2 GPU running at 266MHz I get 328MHz. Compared to the Tegra3 HTC OneX I expected more than that, but if its truly clocked at 320-330MHz the results are pretty much expected.

Assuming the above should be true, then at 400MHz the MP4 should surpass the iPhone4S.

 

[french toast]

Tangey; Fair point, but Apple doesn't license out A5 so it isn't a direct competitor in the SOC space..not in the same terms. we are really comparing it to the likes of OMAP,TEGRA,QUALCOMM,HAEWEI,INTEL etc.

Samsung makes a large number of Apple parts anyhow.

 

[Deleted member 13524]

Soc graphics results under discussion is from a tablet. (...)

No, it isn't.
The post you quoted mentioned HTC One X, which is a smartphone, not a tablet.
Tegra 2/3's GPU clocks vary between smartphone and tablet versions.


Nonetheless, you'll never see A5 or A5X in an Android smartphone/tablet so I don't think it's "fair" to call it the "same class".

Otherwise, when this gets final, I could just compare benchmarks between a MSI Wind 110W and an ipad 2/3, claiming it's the "fastest hardware available for that class", since both are tablets.


Nonetheless, we've seen how inaccurate the results from this "magiclego" device are, so we'd best wait for actual Galaxy S3/whatever numbers to evaluate its performance.

 

[Lazy8s]

The CPU of the MagicLEGO4x12 didn't measure to full Exynos 4212 clock speed in that test, so the clock of the GPU might not have been at the speed being assumed here either.

As Rys mentioned, Exynos 4210's Mali-400 MP4 can be run to 400 MHz; Egypt 720p Offscreen results of around 7K have briefly appeared on GLBenchmark.com in the recent past before being removed whereas stock clocks have tended to yield high 4K to low 5K.

 

[tangey]

The CPU of the MagicLEGO4x12 didn't measure to full Exynos 4212 clock speed in that test, so the clock of the GPU might not have been at the speed being assumed here either.

As Rys mentioned, Exynos 4210's Mali-400 MP4 can be run to 400 MHz; Egypt 720p Offscreen results of around 7K have briefly appeared on GLBenchmark.com in the recent past before being removed whereas stock clocks have tended to yield high 4K to low 5K.

5K->7K would be in the ballpark for a 50% clock increase. (267->400Mhz)

I note that the MagicLego4x12 gets 7.6K, which sounds about right.

 

[Nebuchadnezzar]

How certain are the 400MHz for the Mali400MP4 in the MagicLab 4x12 exactly? The GalaxyS2 scores 55.2fps in 720p Egypt and compared to that it doesn't sound like anywhere near 400MHz. In fact with linear frequency scaling and the GalaxyS2 GPU running at 266MHz I get 328MHz. Compared to the Tegra3 HTC OneX I expected more than that, but if its truly clocked at 320-330MHz the results are pretty much expected.

Assuming the above should be true, then at 400MHz the MP4 should surpass the iPhone4S.

Those scores on the MagicLEGO4210 are too low. I just ran the benchmarks on my S2 to emulate the 4212:

1200MHz ondemand / 267MHz Mali (Default 4210 clocks) :
Egypt : 5866 frames / 52 fps
Pro: 4147 frames / 83 fps

1400MHz ondemand / 400MHz Mali (Supposed 4212 clocks / same as the MagicLEGO):
Egypt : 7972 frames / 71 fps
Pro: 5789 frames / 116 fps

The Mali on the Exynos 4210 can only run in divisions of 800MHz; i.e. 400MHz, 267MHz, 200MHz, etc. It's a clock generation limitation which I doubt they removed on the 4212.

If anybody wants me to benchmark some other combination, feel free to ask.

 

[tangey]

The test I posted a link to for the magiclego4x12 pretty much match your up-clocked 4210.

http://www.glbenchmark.com/phonedetails.jsp?D=MagicLAB+MagicLEGO4x12&benchmark=glpro21

 

[french toast]

Yea looks about right then.

 

[Ailuros]

The test I posted a link to for the magiclego4x12 pretty much match your up-clocked 4210.

http://www.glbenchmark.com/phonedetails.jsp?D=MagicLAB+MagicLEGO4x12&benchmark=glpro21

Do the 4210 and 4212 SoCs have the same amount of bandwidth? If yes it could explain why the 4212 doesn't scale as expected; granted minor differences make sense but these here are wee bit larger. If you take Nebuchadnezzar's 52,0 fps and add 50% for perfect scaling you get a theoretical 78,0 fps, let alone if you take the highest Galaxy S2 score of 55,2 fps + 50% = ~83,0 fps.

68,0 to 78,0 or even 83,0 theoretical is quite a difference. That's exactly why I'm asking if the frequency truly is as high as 400MHz in the 4212 or else something else is pulling it back.

 

[ltcommander.data]

Do the 4210 and 4212 SoCs have the same amount of bandwidth? If yes it could explain why the 4212 doesn't scale as expected; granted minor differences make sense but these here are wee bit larger. If you take Nebuchadnezzar's 52,0 fps and add 50% for perfect scaling you get a theoretical 78,0 fps, let alone if you take the highest Galaxy S2 score of 55,2 fps + 50% = ~83,0 fps.

68,0 to 78,0 or even 83,0 theoretical is quite a difference. That's exactly why I'm asking if the frequency truly is as high as 400MHz in the 4212 or else something else is pulling it back.

http://forum.beyond3d.com/showpost.php?p=1622456&postcount=274

I previously pointed out that according to Samsung's brochures, the memory bandwidth between the 4210 and 4212 are unchanged at 6.4GB/s. So that could certainly be the reason.

 

[Ailuros]

Thanks ltcommander.

 

[Nebuchadnezzar]

Just FYI, I'm getting 6224 frames / 55fps stock and 8362 frames / 74 fps overclocked when simply changing the CPU governor to performance. That sort of difference just happens due to many reasons, you can't expect 100% theoretical scaling. I can actually change the bus frequency to a fixed value to see how much memory bandwidth affects it, but I need to change kernel/compile new one for that.

 

[tangey]

Just FYI, I'm getting 6224 frames / 55fps stock and 8362 frames / 74 fps overclocked when simply changing the CPU governor to performance. That sort of difference just happens due to many reasons, you can't expect 100% theoretical scaling. I can actually change the bus frequency to a fixed value to see how much memory bandwidth affects it, but I need to change kernel/compile new one for that.

what does "performance do" ?

...and what were you doing when you got the follow results a couple of days earlier (or is this the same thing with just an inter-test difference ?)

I just ran the benchmarks on my S2 to emulate the 4212:

1200MHz ondemand / 267MHz Mali (Default 4210 clocks) :
Egypt : 5866 frames / 52 fps
Pro: 4147 frames / 83 fps

1400MHz ondemand / 400MHz Mali (Supposed 4212 clocks / same as the MagicLEGO):
Egypt : 7972 frames / 71 fps
Pro: 5789 frames / 116 fps

 

[Deleted member 13524]

Could be just Gingerbread vs. ICS scores.

 

[Nebuchadnezzar]

what does "performance do" ?

...and what were you doing when you got the follow results a couple of days earlier (or is this the same thing with just an inter-test difference ?)

The CPU governor is in charge of frequency scaling/DVFS and is used too scale down the power it is using by scaling down the frequency. Ondemand for example is a demand-based governor which scales frequency up based on the CPU load in a given sampling period, for example around 80000ms. Frequency scaling has a transition time on any CPU, on the Exynos 4210 it's like 10000µS if I remember correctly, where there's no computations being done. This induces a kind of performance latency on the workload where for example it would need a higher frequency but the governor hasn't scaled up yet. This is why there is the performance difference.

The "performance" governor is simply a dumb governor which fixes the frequency at maximum all the time, and avoids any of the above logic. It's nice for benchmarking but too battery-hungry otherwise.

Could be just Gingerbread vs. ICS scores.

No, I'm running everything else the same. This is on an ICS build.

 

[Ailuros]

Just FYI, I'm getting 6224 frames / 55fps stock and 8362 frames / 74 fps overclocked when simply changing the CPU governor to performance. That sort of difference just happens due to many reasons, you can't expect 100% theoretical scaling. I can actually change the bus frequency to a fixed value to see how much memory bandwidth affects it, but I need to change kernel/compile new one for that.

http://www.glbenchmark.com/phonedetails.jsp?D=Samsung+GT-I9300&benchmark=glpro21

That's slightly above iPad2; I expected give or take that it'll break even with the MP2@250MHz and it doesn't seem all that far from it.

 

[Nebuchadnezzar]

http://www.glbenchmark.com/phonedetails.jsp?D=Samsung+GT-I9300&benchmark=glpro21

That's slightly above iPad2; I expected give or take that it'll break even with the MP2@250MHz and it doesn't seem all that far from it.

That's a huge jump on the Egypt score even from my overclocked device... 74fps to 92fps. Either the quadcore brings a lot, the GPU is running faster than 400MHz, or there's some new driver magic going on. The strange thing is that the Pro scores aren't that off from what I'm getting.

 

[Wishmaster]

That's a huge jump on the Egypt score even from my overclocked device... 74fps to 92fps. Either the quadcore brings a lot, the GPU is running faster than 400MHz, or there's some new driver magic going on. The strange thing is that the Pro scores aren't that off from what I'm getting.

Maybe they improved their drivers to score higher in Egypt test? Scores like that make it the fastest android device which is great from marketing POV.

 

[french toast]

Maybe they improved their drivers to score higher in Egypt test? Scores like that make it the fastest android device which is great from marketing POV.

Either way its going to be the fastest Android smartphone device of its generation...what does everyone think those quad A9's are going to be clocked at?

Also how come the off screen score is twice the frame rate as the standard on screen test..when the device runs at 720p?? is it a different test?