I'd rather hand that one out to Samsung irrelevant what the GalaxyS4 will contain in the end.
Qualcomm Krait & MSM8960 @ AnandTech
- Thread starter Arun
- Start date
I'd rather hand that one out to Samsung irrelevant what the GalaxyS4 will contain in the end.
D
Deleted member 13524
Guest
Isn't the S4 coming with a Snapdragon S600 too?
What I meant is that I don't expect any worthwhile differences at least in terms of graphics performance whether oc'ed S600 or octacore Exynos. The Galaxy S4 seems to appear with a S600 with the CPU clocked at 1.9GHz and likely the GPU clocked at 480MHz if that Samsung S600 device in the GLB2.5 database is the one; I'm not still not in the clear has Samsung cancelled their octacore entirely or is it merely another case example with one SoC for LTE markets and the other SoC for non-LTE markets. If that should turn out to be case after all I wouldn't be one bit surprised if both variants of the GalaxyS4 would have similar CPU and GPU frequencies.
http://www.fudzilla.com/home/item/30593-galaxy-s4-comes-on-march-14 just a tad more than 2 weeks to go.
french toast
Veteran
I still dont get where the 480mhz number comes from ailuros..
We know adreno 320 is 400mhz...any speed advantages are likely due to lpddr3 and drivers which should come with Android 4.2.x ...hence why htc one does not see any advantages due to the absence of the above.
I've explained it already in a past post and no I won't explain it twice sorry.
We have a new Snapdragon 600 result over at GLBenchmark (1.7 GHz / Adreno 320), from a Sharp FIH-VKY.
It scores 38.6 FPS in GLBenchmark 2.5 Egypt HD C24Z16 - Offscreen (1080p), quite a bump from the multiple 34 FPS from the same chipset in the HTC One. Luckily this time, we also have low-level tests results, unlike the Samsung SHV-E300S result.
The Sharp is running 4.2.1, and what looks like newer Qualcomm drivers than the HTC. We know that the One is using LPDDR2, and not LPDDR3, which leads me to question the Fill Rate tests.
Fill Rate 1080p Offscreen
HTC One - 963 Mtexels/s
Sharp - 790 Mtexels/s
As the Sharp scores higher in Egypt HD, than the One, it surely can't have worse memory bandwidth?, but if the One was being bottlenecked by its LPDDR2 then wouldn't it score in the Fill Rate tests be much lower? Also, doesn't the fill rate result suggest that the Sharp's GPU is not clocked higher the HTC One.
Of course we don't anything about the Sharp FIH-VKY, back ground processes / logging / test environment etc. My question do these results add credence to the assertion, that the performance difference between the One and Samsung / Sharp is actually just driver related,
HTC One low-level
http://www.glbenchmark.com/phonedetails.jsp?benchmark=glpro25&D=HTC+One&testgroup=lowlevel
Sharp
http://www.glbenchmark.com/phonedetails.jsp?benchmark=glpro25&D=Sharp+FIH-VKY&testgroup=lowlevel
It scores 38.6 FPS in GLBenchmark 2.5 Egypt HD C24Z16 - Offscreen (1080p), quite a bump from the multiple 34 FPS from the same chipset in the HTC One. Luckily this time, we also have low-level tests results, unlike the Samsung SHV-E300S result.
The Sharp is running 4.2.1, and what looks like newer Qualcomm drivers than the HTC. We know that the One is using LPDDR2, and not LPDDR3, which leads me to question the Fill Rate tests.
Fill Rate 1080p Offscreen
HTC One - 963 Mtexels/s
Sharp - 790 Mtexels/s
As the Sharp scores higher in Egypt HD, than the One, it surely can't have worse memory bandwidth?, but if the One was being bottlenecked by its LPDDR2 then wouldn't it score in the Fill Rate tests be much lower? Also, doesn't the fill rate result suggest that the Sharp's GPU is not clocked higher the HTC One.
Of course we don't anything about the Sharp FIH-VKY, back ground processes / logging / test environment etc. My question do these results add credence to the assertion, that the performance difference between the One and Samsung / Sharp is actually just driver related,
HTC One low-level
http://www.glbenchmark.com/phonedetails.jsp?benchmark=glpro25&D=HTC+One&testgroup=lowlevel
Sharp
http://www.glbenchmark.com/phonedetails.jsp?benchmark=glpro25&D=Sharp+FIH-VKY&testgroup=lowlevel
french toast
Veteran
Perhaps the sharp benefits from the new drivers which make one hell of a difference. .therfore allowing them to use lower clocks and/or lower bandwidth?
Thats the only explanation I can think off...gl benchmark 2.5 @ 1080p wouldn't really benefit from having loads of fill rate would it?...where as a big bump in shader efficiency would likely affect that benchmark alot.
Thats the only explanation I can think off...gl benchmark 2.5 @ 1080p wouldn't really benefit from having loads of fill rate would it?...where as a big bump in shader efficiency would likely affect that benchmark alot.
french toast
Veteran
Stumbled across something small but interesting. ..
" GPU clocks appear unchanged, which is contrary to what I was told at the launch of Krait 300 but it’s entirely possible that we’ll see implementation with higher GPU clocks."
http://anandtech.com/show/6851/the-htc-one-a-remarkable-device-anands-mini-review/3
I was under the impression that adreno 320 carried the same clocks from S4~s600? Hence the same naming scheme.
That only the 330 carried a 50mhz clock bump...this was seemingly consolidated by the fortress fire demo and the leaked qualcomm roadmap a while back?.
Anand would obviously have access to better information than many..# confused :/
" GPU clocks appear unchanged, which is contrary to what I was told at the launch of Krait 300 but it’s entirely possible that we’ll see implementation with higher GPU clocks."
http://anandtech.com/show/6851/the-htc-one-a-remarkable-device-anands-mini-review/3
I was under the impression that adreno 320 carried the same clocks from S4~s600? Hence the same naming scheme.
That only the 330 carried a 50mhz clock bump...this was seemingly consolidated by the fortress fire demo and the leaked qualcomm roadmap a while back?.
Anand would obviously have access to better information than many..# confused :/
Over at XDA, a member in Germany rooted his new HTC One and ran an app (Faux123 Kernel Enhancement) which read various device info. This also confirmed that the max GPU clock of 400 MHz. I strongly believe that GS4 (qualcomm) is also running a 400 MHz clock, if the Fill Rate of the Samsung SHV 300 is anything to go by.
Using the Nexus 4 as starting point, the Snapdragon 600 GPU should be stable at up to 487 MHz, with an undervolt and down-clocking the CPU a couple of hundred MHz for those so inclined
french toast
Veteran
Over at XDA, a member in Germany rooted his new HTC One and ran an app (Faux123 Kernel Enhancement) which read various device info. This also confirmed that the max GPU clock of 400 MHz. I strongly believe that GS4 (qualcomm) is also running a 400 MHz clock, if the Fill Rate of the Samsung SHV 300 is anything to go by.
Using the Nexus 4 as starting point, the Snapdragon 600 GPU should be stable at up to 487 MHz, with an undervolt and down-clocking the CPU a couple of hundred MHz for those so inclined
Good find
just a personal opinion for what ever thats worth, but I think qualcomm have got the correct set up for smartphones. ..everything about their soc design makes total sense for that form factor running android. Adreno 320 is a very nice chip it seems, adreno 330 leaked clock speed would also fit in with your overclock scenario. .450mhz.
Anyone got any ideas where the claimed 2x compute increase will come from on that chip?
http://news.techeye.net/chips/qualcomm-tsmc-unveil-28nm-hpm-chips
The article is roughly a month old, but it does give us some idea of what power consumption for their Snapdragon 800 chipset might be like.
They're stating around 750 mW per core at 2 to 2.3 GHz.
The article is roughly a month old, but it does give us some idea of what power consumption for their Snapdragon 800 chipset might be like.
They're stating around 750 mW per core at 2 to 2.3 GHz.
If that power, and early AnTuTu figure of 32146 are correct, it looks to another successful product.
http://gsminsider.com/2013/03/24/sn...k-result-leaks-scores-more-than-32000-points/
french toast
Veteran
I will be suprised if we see s800 clocked at full speed inside a phone...especially with decent power consumption.
Looks very good though.
Looks very good though.
At this point, I almost feel as if chip OEMs are just quoting Turbo clock speeds as its unlikely the CPU will ever sustain those speeds for long for battery reasons, just like desktop/laptop chips are limited because of thermals.
....and we have a new kid on the Kishonti Number One spot: http://glbenchmark.com/phonedetails.jsp?D=LG+D801
Please welcome the Adreno330 by 29% faster than the iPad4.
Please welcome the Adreno330 by 29% faster than the iPad4.
french toast
Veteran
Little tid bit.
http://www.engadget.com/2013/03/26/unity-to-bring-qualcomm-optimization/
Basically unity- the developer of the popular mobile 3D game engine for mobile devices, - is teaming up with qualcomm to optimise its engine for Adreno.
How much difference this will make is open to debate..but its nice to see a better collaboration between hardware/software on mobile.
http://www.engadget.com/2013/03/26/unity-to-bring-qualcomm-optimization/
Basically unity- the developer of the popular mobile 3D game engine for mobile devices, - is teaming up with qualcomm to optimise its engine for Adreno.
How much difference this will make is open to debate..but its nice to see a better collaboration between hardware/software on mobile.
Anandtech's 3dmark on Adroid: http://www.anandtech.com/show/6875/3dmark-for-android-performance-preview
While it is nice to be able to compare and constrast 720p and 1080p rendered resolution results across various Android devices, I do think that Futuremark missed an opportunity here by not giving users and reviewers the ability to test at the natively rendered resolution.
Notice that when comparing the Default settings (720p rendered resolution, normal quality textures, and normal quality post-processing effects) to the Extreme settings (1080p rendered resolution, high quality textures, and high quality post-processing effects), the Adreno 320 GPU (used in the Google Nexus 4) and the ULP Tegra 3 Geforce GPU (used in the HTC One X+) have a much greater performance hit than the Mali T604 GPU (used in the Google Nexus 10).
Comparing Adreno 320 to Mali T604 at Default settings (720p with normal quality textures and normal quality post-processing), the performance improvement is as follows:
+33.3% in Graphics Test 1
+49.8% in Graphics Test 2
+41.9% in Physics Test
+49.1% in Demo Test
Comparing Adreno 320 to Mali T604 at Extreme settings (1080p with high quality textures and high quality post-processing), the performance improvement is as follows:
+9.2% in Graphics Test 1
+25.3% in Graphics Test 2
-11.2% in Physics Test
+19.5% in Demo Test
There does appear to be room for improvement for different mobile GPU's by using newer drivers. For example, one HTC Droid DNA owner at Anandtech claimed that his 3DMark score was nearly 30% higher than what Anand measured. There was also one LG Optimus G owner at Anandtech who claimed that his 3DMark score was nearly 60% (!) higher than what Anand measured. So I think Anand needs to remeasure using the latest drivers.
At the Extreme settings, none of the currently available mobile GPU's for Android offer framerates that would be considered "playable" in the 3DMark Ice Storm test.
Notice that when comparing the Default settings (720p rendered resolution, normal quality textures, and normal quality post-processing effects) to the Extreme settings (1080p rendered resolution, high quality textures, and high quality post-processing effects), the Adreno 320 GPU (used in the Google Nexus 4) and the ULP Tegra 3 Geforce GPU (used in the HTC One X+) have a much greater performance hit than the Mali T604 GPU (used in the Google Nexus 10).
Comparing Adreno 320 to Mali T604 at Default settings (720p with normal quality textures and normal quality post-processing), the performance improvement is as follows:
+33.3% in Graphics Test 1
+49.8% in Graphics Test 2
+41.9% in Physics Test
+49.1% in Demo Test
Comparing Adreno 320 to Mali T604 at Extreme settings (1080p with high quality textures and high quality post-processing), the performance improvement is as follows:
+9.2% in Graphics Test 1
+25.3% in Graphics Test 2
-11.2% in Physics Test
+19.5% in Demo Test
There does appear to be room for improvement for different mobile GPU's by using newer drivers. For example, one HTC Droid DNA owner at Anandtech claimed that his 3DMark score was nearly 30% higher than what Anand measured. There was also one LG Optimus G owner at Anandtech who claimed that his 3DMark score was nearly 60% (!) higher than what Anand measured. So I think Anand needs to remeasure using the latest drivers.
At the Extreme settings, none of the currently available mobile GPU's for Android offer framerates that would be considered "playable" in the 3DMark Ice Storm test.
Last edited by a moderator:
french toast
Veteran
@ ams. . Could that be just down to bandwidth though?
All adreno on that chart just have access to lpddr2 1066...t604 uses lpddr3 1600 (12.8 gb/s).
I would like the test to be redone with galaxy s4 and lonovo k900.
Edit..found this supposed. Galaxy S4 score.
http://www.enet.com.cn/article/2013/0403/A20130403273397.shtml
Its in chinese but it looks very low compared to other adrenos..could this be exynos 5410?
Edit 2..yea had a proper look and it does say clearly exynos and sgx 544 mp3.
Dissapointing graphics performance if you ask me...especially when you consider it has access to lpddr3 also.
All adreno on that chart just have access to lpddr2 1066...t604 uses lpddr3 1600 (12.8 gb/s).
I would like the test to be redone with galaxy s4 and lonovo k900.
Edit..found this supposed. Galaxy S4 score.
http://www.enet.com.cn/article/2013/0403/A20130403273397.shtml
Its in chinese but it looks very low compared to other adrenos..could this be exynos 5410?
Edit 2..yea had a proper look and it does say clearly exynos and sgx 544 mp3.
Dissapointing graphics performance if you ask me...especially when you consider it has access to lpddr3 also.
Last edited by a moderator:
