Yes, that's depressing
NVIDIA Tegra Architecture
- Thread starter french toast
- Start date
-
- Tags
- nvidia
Since NV has optimised its GPU driver further (which is probably used for Shield only), this one should end up in GLB2.7 at around 20+fps.
Brilliantdeve
Newcomer
Review of Toshiba's 10 inch Tegra 4 powered tablet.
Summarised version, great CPU performance, before throttling kicks in. GPU performance in GLB T.rex offscreen is only 16 FPS, which shows the GPU clocks must be running at an average MHz far lower lower than in Nvidia's Shield, or running much older drivers (why) or both. The tablet does get very hot at full load, but that isn't uncommon is it now.
It used a maximum of 13.2 watts (total system draw), compared to an iPad 4 at 12.5 watts, despite the iPad being built on a larger process node, having quad memory channels and a larger die area GPU (irc) The Toshiba does have a 2560x1600 display, so has to power more pixel than the iPad, which must be factored into the equation, but still.. It as a 33 Wh battery, larger than the Shield @ 28.8, I don't know the mAh. What really seems to hold T4 back is the issue of heat dissipation, with a small / quiet fan T4 performs well. The Toshiba tablet is hardly the most svelte design, maybe they should have made the bold step and equipped it with a fan (I'm serious )
What I am still not clear about is whether the heat / TDP issues are the result of the A15 or the GPU. Android is now heavily GPU reliant, so that part is often in use, whereas 4 A15s can burn a lot of power when fully loaded, but how often in a mobile OS like Android are all 4 A15s at max usage< questions, questions. If it's CPU related then Nvidia is going to run into the same issues with T5, or are there solutions to the problem apart from running lower clocks?
http://www.notebookcheck.net/Review-Toshiba-Excite-Pro-AT10LE-A-108-Tablet.98828.0.html
Summarised version, great CPU performance, before throttling kicks in. GPU performance in GLB T.rex offscreen is only 16 FPS, which shows the GPU clocks must be running at an average MHz far lower lower than in Nvidia's Shield, or running much older drivers (why) or both. The tablet does get very hot at full load, but that isn't uncommon is it now.
It used a maximum of 13.2 watts (total system draw), compared to an iPad 4 at 12.5 watts, despite the iPad being built on a larger process node, having quad memory channels and a larger die area GPU (irc) The Toshiba does have a 2560x1600 display, so has to power more pixel than the iPad, which must be factored into the equation, but still.. It as a 33 Wh battery, larger than the Shield @ 28.8, I don't know the mAh. What really seems to hold T4 back is the issue of heat dissipation, with a small / quiet fan T4 performs well. The Toshiba tablet is hardly the most svelte design, maybe they should have made the bold step and equipped it with a fan (I'm serious
) What I am still not clear about is whether the heat / TDP issues are the result of the A15 or the GPU. Android is now heavily GPU reliant, so that part is often in use, whereas 4 A15s can burn a lot of power when fully loaded, but how often in a mobile OS like Android are all 4 A15s at max usage< questions, questions. If it's CPU related then Nvidia is going to run into the same issues with T5, or are there solutions to the problem apart from running lower clocks?
http://www.notebookcheck.net/Review-Toshiba-Excite-Pro-AT10LE-A-108-Tablet.98828.0.html
lol at the fan idea. I suppose that it's an option but Tegra 4 is supposed to be in a class that's fanless.
The CGPU unit in SoCs now is supposed to make a high-resolution Android GUI a non-issue when it comes to power consumption. Old 45nm OMAP4470 has no problem with it.
The CGPU unit in SoCs now is supposed to make a high-resolution Android GUI a non-issue when it comes to power consumption. Old 45nm OMAP4470 has no problem with it.
Another bit from that review
They say you should wander over to the Cortex A7 isle if you want something that's not a hand heater. But it's undoubtedly a combination of it being a quad core and that big gaming GPU. NV just pushed 28nm too far.
13 Watts passively cooled in your hands....
They say you should wander over to the Cortex A7 isle if you want something that's not a hand heater. But it's undoubtedly a combination of it being a quad core and that big gaming GPU. NV just pushed 28nm too far.
LOL. They looped 3dmark 10 times (!) on a thin fanless tablet and then observed that thermal overload protection kicked in? Tell us something we don't know
You did fail to mention that this T4-powered Toshiba tablet has a much faster quad-core CPU vs. the dual-core CPU in the ipad 4 (in addition to the higher-than-retina resolution display)
So a peak power consumption that is only 5.6% higher than the ipad 4 is not a bad result relative to the ipad all things considered.A thin fanless tablet is not going to spend any meaningful amount of time at 13w peak power consumption. Anyway, as I mentioned above, this is only 5.6% higher peak power consumption than the ipad 4 (which means that the Toshiba tablet actually has higher perf. per watt than the ipad 4 tablet; Kepler.M is obviously in another league in terms of perf. per watt if initial test results are anything to go by).
On a side note, even though peak temperature and peak power consumption of this Toshiba tablet is similar to the ipad 4, the flaw in the design is that the SoC is placed at or near the spot where the left hand holds the tablet. This becomes an issue because this is a horizontally oriented tablet.
Last edited by a moderator:
I haven't seen any noticable throttling on other tablets after spending a couple of hours in demanding 3D, but hey there have to be excuses for anything.
How about finding an excuse for that one also: http://www.fudzilla.com/home/item/32211-nvidia’s-tegra-business-faces-more-headwinds
How about finding an excuse for that one also: http://www.fudzilla.com/home/item/32211-nvidia’s-tegra-business-faces-more-headwinds
Silent_Buddha
Legend
LOL. They looped 3dmark 10 times (!) on a thin fanless tablet and then observed that thermal overload protection kicked in? Tell us something we don't know
And yet the Snapdragon 800 manages to handle that just fine. In fact, outperforming the actively cooled Tegra 4 while being itself passively cooled in a "thin fanless tablet."
Unless it's really cheap, I see no reason for any tablet manufacturer to use Tegra 4. But perhaps it'll do better in Nvidia's own tablet design and/or Microsoft's Surface RT v.2.
Regards,
SB
D
Deleted member 13524
Guest
LOL. They looped 3dmark 10 times (!) on a thin fanless tablet and then observed that thermal overload protection kicked in? Tell us something we don't know
Running 3dmark 10 times in a row is representative of an actual gaming session with a 3d game. A single run is not.
That toshiba tablet is obviously under-delivering in the thermal dissipation front.
LOL, running at the 3dmark "Extreme" settings looped 10 times in a row is representative of actual 3d gaming on an Android tablet? Not at all. You do realize that, even when not throttling, the average FPS at these Extreme settings is ~ 30fps (and that is with Tegra 4, which is one of the highest performance ultra mobile SoC's currently available today)? Most Android games would have a much higher framerate than this on any T4-powered device. And obviously 3dmark Extreme will push the CPU too with it's Physics test (and in fact, the CPU throttling is likely to be worse than the GPU throttling).
This Toshiba tablet has a variety of different problems (ranging from poor SoC placement to a battery hogging screen to a blurry camera to buggy software). So it will be interesting to read reviews on other T4-powered devices such as the upcoming HP Slatebook x2 and Asus Transformer Pad Infinity to see if they suffer from the same issues. Shield, on the other hand, has had absolutely fantastic user reviews.
This Toshiba tablet has a variety of different problems (ranging from poor SoC placement to a battery hogging screen to a blurry camera to buggy software). So it will be interesting to read reviews on other T4-powered devices such as the upcoming HP Slatebook x2 and Asus Transformer Pad Infinity to see if they suffer from the same issues. Shield, on the other hand, has had absolutely fantastic user reviews.
The SoC die size of S800 is ~ 50% larger than T4, so I would say that T4 should be much less expensive in comparison. T4 will be used by Asus (in the upcoming Transformer Pad Infinity) and HP (in the upcoming Slatebook x2), among others.
That is incorrect. The CPU performance of the quad Cortex A15 in Shield is better than the quad Krait 400 in S800 (and the GPU performance is pretty close too: only 1-8% difference). In fact, a passively cooled S800 will likely throttle frequencies much more than an actively cooled device such as Shield over more extended periods of processor-intensive activity.
Last edited by a moderator:
Yes the overarching one appears to be T4 though.
Given what we know about T4 in tablets, it looks like Nvidia has been pulling shenanigans with Shield. I guess that's always a problem with self-built hardware, but if this Toshiba tablet is pulling 13W I don't see how Shield with it's faster chip is only pulling 5W, which is what the battery life tests suggest.
D
Deleted member 13524
Guest
Yes, it is. 3dmark won't heat up the SoC anymore than playing NOVA3 or MC4 at 1080p.
3dmark is not the android equivalent of furmark.
This is about temperatures, not performance.
Nonetheless, how the SoC behaves with last year's games is not an absolute measurement.
If someone buys a Tegra 4 tablet and intends to play 3d games, I'm pretty sure that customers won't be willing to limit themselves to 3d games from 2012. Some of them will be playing games well into 2015.
Even if the throttling is worse on the CPU side, it's still a fact that the device gets about half of its initial performance when playing a 3d game for a long time.
There are reports of this tablet's owners complaining about sluggish gaming performance after a few minutes of usage.
Some users are actually claiming that forcing the tablet to run in power saving mode (limiting the CPU cores to 1GHz and a mild underclock to the GPU) will result in better continuous gaming performance.
Yes, my comment was about the Toshiba tablet (I thought that was obvious). Even if the Shield has active cooling, I didn't automatically assume that all fanless Tegra 4 devices will share the same overheating/throttling issues as the AT10LE.
Similar threads
