NVIDIA Tegra Architecture

[xpea]

Xiaomi typically is a wee bit aggressive with frequencies; if you look back at Tegra4 results the T4 equipped Xiaomi Mi3 exceeds by a healthy margin more than one Tegra4 tablet:

http://gfxbench.com/result.jsp?benc...VIDIA Tegra&base=device&os-check-Windows_dx=0

The ZTE T4 smartphone performs by 20% lower than the Xiaomi.

my wife has a T4 MI3 and it doesn't run hotter than my S800 Nexus5. In fact, I see no real world difference in performance between these 2 phones. They are both blazing fast. Of course, due to his much bigger battery, the T4 Mi3 has much better autonomy.
All in one, if other OEMs are more conservative in their clock rates, it's not Nvidia/Xiaomi fault...

 

[DSC]

http://gfxbench.com/result.jsp

 

[Ailuros]

my wife has a T4 MI3 and it doesn't run hotter than my S800 Nexus5. In fact, I see no real world difference in performance between these 2 phones. They are both blazing fast. Of course, due to his much bigger battery, the T4 Mi3 has much better autonomy.

And that autonomy is being measured in standby, talk time or coincidentially in battery life in something like a demanding 3D game? The Mi3 has a bigger battery for a reason.

All in one, if other OEMs are more conservative in their clock rates, it's not Nvidia/Xiaomi fault...

Other vendors that had used T4 for their tablets don't have the chinese state to support them like Xiaomi which have the luxury to sell even at a loss. NVIDIA has no saying at all in these cases as long as the manufacturer doesn't exceed the maximum frequencies.

 

[xpea]

http://gfxbench.com/result.jsp

disruptive results here
When S805 will be available in devices ? I remember its manufactured on 20nm, so end of year time frame ?
and what's the perf update on Adreno 420 compared to 330 ? will it be able to match TK1 kepler ?

 

[xpea]

And that autonomy is being measured in standby, talk time or coincidentially in battery life in something like a demanding 3D game? The Mi3 has a bigger battery for a reason.

no scientific test, just real life usage. With my wife we have same profile, twitter & skype for me, Weibo and QQ for her, checking gmail all day long, a bit of games and web browsing. very few calls. As we work together, our usage time is the same.
Screen around 30%
Wifi 15%
Skype 14%
voice calls less than 10%

Usually, the T4 Mi3 has 20% battery when back at home when my Nexus 5 is already exhausted before leaving the office (Powerbank is my savior)

 

[Ailuros]

no scientific test, just real life usage. With my wife we have same profile, twitter & skype for me, Weibo and QQ for her, checking gmail all day long, a bit of games and web browsing. very few calls. As we work together, our usage time is the same.
Screen around 30%
Wifi 15%
Skype 14%
voice calls less than 10%

Usually, the T4 Mi3 has 20% battery when back at home when my Nexus 5 is already exhausted before leaving the office (Powerbank is my savior)

Since we are obviously talking about frequencies/Gfxbench results in this topic here you go:

http://gfxbench.com/result.jsp?benchmark=gfx30&test=559&order=score&text-filter=Xiaomi &base=device

A Mi3 with a S800 has an at least 30% higher battery life compared to its T4/Mi3 brother.

And of course this one too:

http://gfxbench.com/result.jsp?benchmark=gfx30&test=558&order=score&text-filter=Xiaomi &base=device

 

[xpea]

Since we are obviously talking about frequencies/Gfxbench results in this topic here you go:

http://gfxbench.com/result.jsp?benchmark=gfx30&test=559&order=score&text-filter=Xiaomi &base=device

A Mi3 with a S800 has an at least 30% higher battery life compared to its T4/Mi3 brother.

And of course this one too:

http://gfxbench.com/result.jsp?benchmark=gfx30&test=558&order=score&text-filter=Xiaomi &base=device

yeah but this is for gaming. our real life usage is maybe only 10% gaming...

 

[Ailuros]

disruptive results here
When S805 will be available in devices ? I remember its manufactured on 20nm, so end of year time frame ?
and what's the perf update on Adreno 420 compared to 330 ? will it be able to match TK1 kepler ?

I believe QCOM claimed Adreno330 +40% for which I assume that its valid for the same clocks between the two? In any case the highest clocked 330 right now achieves ~12 fps in Manhattan offscreen + those 40% = 17 fps. Careful though if a vendor would integrate a K1 into a high end smartphone I wouldn't expect there any 28-30 fps performance in that bench either.

 

[Ailuros]

yeah but this is for gaming. our real life usage is maybe only 10% gaming...

Of course is it for gaming. If you use a smartphone just to render the UI you don't need a high end GPU in it.

 

[Helmore]

disruptive results here
When S805 will be available in devices ? I remember its manufactured on 20nm, so end of year time frame ?
and what's the perf update on Adreno 420 compared to 330 ? will it be able to match TK1 kepler ?

S805 will be on 28 nm. Check here: http://www.qualcomm.com/snapdragon/processors/805

 

[Ailuros]

S805 will be on 28 nm. Check here: http://www.qualcomm.com/snapdragon/processors/805

28HPm for the unnecessary hairsplitting; S810 with the Adreno430 will be on 20SoC. The latter could have as many as 16 clusters but I don't expect it before fall this year and that's even an "if".

 

[Helmore]

I expect S810 phones at MWC2015 at the earliest, but that might be slightly pessimistic.

 

[xpea]

Of course is it for gaming. If you use a smartphone just to render the UI you don't need a high end GPU in it.

well, in 2014 I consider a smartphone as my main computer (at least for me). It's a device that stays with me 24 hours per day. Thus I want it to compute all the tasks as fast as possible. So I need fast CPU to handle many parallel social activities, fast GPU for gaming between work time, a good screen, and obviously good battery life (Nexus 5 is a bit short in this area).
I had a Nexus 4 and jumping to the Nexus 5 was a good improvement in daily usage (multi tasking, responsiveness, not only gaming). So yes, S800, even without a better GPU is much better for me than S600

 

[silent_guy]

Of course is it for gaming. If you use a smartphone just to render the UI you don't need a high end GPU in it.

At least xpea's experience contradicts the doom and gloom narrative that Tegra SOCs are a power disaster in everything they do. You may only be interested in the highest power 3D use case, and that's fine from a technical point of view, but let's not forget the fact that nobody is running glbench or similarly stressful 3d apps more than 10% of the time.
If they're gaming, candy crush and friend are much more common.

 

[xpea]

I see...
so it's realistic to expect S805 in Q3/14 devices and S810 in Q2/15
TK1 competitor will be S805 / Apple A8
and Erista will face S810

if projected QC performance is to be believed, it looks like Nvidia will have the edge from now on. I don't see any other company beating NV GPU, especially with Maxwell on the way. Let's see how it will translate into design wins / market share...

 

[Ailuros]

At least xpea's experience contradicts the doom and gloom narrative that Tegra SOCs are a power disaster in everything they do.

At least we have you to bump into every other conversation trying to play devil's advocate without any more substance than the above that you oppose to. While you're busy chasing your own tail it might be more useful for a change to consider that instead of assisting in keeping debates on a civilized level you're in reality rather spilling oil into the fire.

You may only be interested in the highest power 3D use case, and that's fine from a technical point of view, but let's not forget the fact that nobody is running glbench or similarly stressful 3d apps more than 10% of the time.

A technical point of view is what would suit a board like B3D best. If you want to analyze if the cassis colour matches your shoes I'm afraid you're at the wrong place.

 

[Ailuros]

well, in 2014 I consider a smartphone as my main computer (at least for me). It's a device that stays with me 24 hours per day. Thus I want it to compute all the tasks as fast as possible. So I need fast CPU to handle many parallel social activities, fast GPU for gaming between work time, a good screen, and obviously good battery life (Nexus 5 is a bit short in this area).
I had a Nexus 4 and jumping to the Nexus 5 was a good improvement in daily usage (multi tasking, responsiveness, not only gaming). So yes, S800, even without a better GPU is much better for me than S600

Nothing against that; but I can buy a PC just for office/multimedia use or I can buy a PC for high end gaming. Is in the second case system power consumption irrelevant if I spend only 10% of my total PC time for gaming? Does power consumption shrink?

Needless to say that if I'd run several tests to rate any system, mobile device or any other device's power consumption, I wouldn't use the lightest use case scenarios but the exact opposite; at the very least no GPU will clock at its peak frequency for web browsing or any other garden variety use case.

It's not that much different with CPUs in the mobile space too, with the only other difference that public benchmarks for those are in their majority more questionable in their merits than their GPU benchmark counter-applications.

 

[silent_guy]

My apologies for seeing value in actual use cases, something for which a lot of data is missing, especially for Tegra devices, because there aren't a lot of those around to begin with.

That was terribly rude of me.

 

[Ailuros]

My apologies for seeing value in actual use cases, something for which a lot of data is missing, especially for Tegra devices, because there aren't a lot of those around to begin with.

That was terribly rude of me.

My apologies for making you sound like a grumpy old spinster, but you actually do sound like one.

 

[silent_guy]

my wife has a T4 MI3 and it doesn't run hotter than my S800 Nexus5. In fact, I see no real world difference in performance between these 2 phones. They are both blazing fast. Of course, due to his much bigger battery, the T4 Mi3 has much better autonomy.

Could you do some real-world tests?

E.g.
- Charge both phones in the evening, then disconnect them at night, see how much battery is left in the morning.
- From fully charged, run glbench for 10min, check battery afterwards. Also, check differences between bench scores before and after to see impact of heat etc.