NVIDIA Tegra Architecture

[ams]

Anandtech loops GLBenchmark until the battery is exhausted for their 3D battery life benchmark

Anandtech ran their 3D battery life test with GLBench 2.5 Egypt HD (1080p offscreen) test. That benchmark is way less stressful than 3dmark Ice Storm Extreme. And when capped at 30fps as Anand did for his 3D battery life test, the A6X GPU (or even Tegra 4's GPU) will absolutely cruise along on that benchmark.

On the other hand, by looping the entire 3dmark Ice Storm Extreme benchmark (including the CPU-heavy Physics test) 10 times, the CPU throttling was quite a bit worse than even the GPU throttling on that Toshiba tablet...and even before throttling, the average fps barely reaches 30fps. How is that representative of real world gaming on these devices? Games that are actually smooth and playable on these devices will have an average fps that is well above 30fps, and will not have ridiculously massive CPU throttling either. And don't forget that the Toshiba tablet's screen appears to consume a lot of power and is a battery life hog too. I would expect HP's Slatebook x2 tablet to have better real world performance and better battery life (even without the keyboard dock) because the screen is not as power hungry in comparison.

 

[ams]

Just as much as the supposed "exclusivity" for the VW group while for some strange reason Jacinto SoCs should then also be developed by NV.

You do realize that modern-day cars such as Audi's can have hundreds of processors inside? NVIDIA's Tegra powers the heart of Audi's infotainment platform (MMX - Multimedia Extension), while Texas Instruments' Jacinto powers the Radio and Car Control Unit (RCC) element of Audi's infotainment platform. It is easy to be dismissive of the automotive business, but Tegra is now used in more than 20 automotive brands and more than 100 models, and that business will keep doubling year-over-year over the next three fiscal years.

 

[Ailuros]

You do realize that modern-day cars such as Audi's can have hundreds of processors inside? NVIDIA's Tegra powers the heart of Audi's infotainment platform (MMX - Multimedia Extension), while Texas Instruments' Jacinto powers the Radio and Car Control Unit (RCC) element of Audi's infotainment platform. It is easy to be dismissive of the automotive business, but Tegra is now used in more than 20 automotive brands and more than 100 models, and that business will keep doubling year-over-year over the next three fiscal years.

If you can answer the quite simple and quite reasonable question Exophase posted preferably with your own words it would be a eulogy and a kickstart for a real conversation.

Other than that all the real big players dealing in the embedded/automotive markets just waited for NVIDIA to participate in it only to roll over and wave a white flag. How about a statistic or something that shows marketshare in the specific market other than NV's own claims?

 

[Xmas]

On the other hand, by looping the entire 3dmark Ice Storm Extreme benchmark (including the CPU-heavy Physics test) 10 times, the CPU throttling was quite a bit worse than even the GPU throttling on that Toshiba tablet...and even before throttling, the average fps barely reaches 30fps. How is that representative of real world gaming on these devices? Games that are actually smooth and playable on these devices will have an average fps that is well above 30fps, and will not have ridiculously massive CPU throttling either.

There are two aspects to gaming performance:
- battery life with current low/medium complexity content, which high-end GPUs are likely to run at a solid 60 fps (or 30 if fast movement is not a concern) with cycles to spare.
- the capacity to run future/high complexity content where the GPU can only just sustain a playable frame rate and has no idle time to cool down.

It doesn't matter all that much that this specific test barely reaches playable frame rates (which is a matter of context anyway; 30 fps average or even lower can be well playable if it's constant). Just imagine that the same workload were split over more frames, such that the frame rate is higher but not high enough to get any idle cycles.

You keep pointing out that the CPU throttles more but I'm not sure what your point is.

 

[french toast]

Anandtech ran their 3D battery life test with GLBench 2.5 Egypt HD (1080p offscreen) test. That benchmark is way less stressful than 3dmark Ice Storm Extreme. And when capped at 30fps as Anand did for his 3D battery life test, the A6X GPU (or even Tegra 4's GPU) will absolutely cruise along on that benchmark.

On the other hand, by looping the entire 3dmark Ice Storm Extreme benchmark (including the CPU-heavy Physics test) 10 times, the CPU throttling was quite a bit worse than even the GPU throttling on that Toshiba tablet...and even before throttling, the average fps barely reaches 30fps. How is that representative of real world gaming on these devices? Games that are actually smooth and playable on these devices will have an average fps that is well above 30fps, and will not have ridiculously massive CPU throttling either. And don't forget that the Toshiba tablet's screen appears to consume a lot of power and is a battery life hog too. I would expect HP's Slatebook x2 tablet to have better real world performance and better battery life (even without the keyboard dock) because the screen is not as power hungry in comparison.

@ams..sorry to keep banging on about this but it seems you still need convincing.

Tegra 4 doesnt just throttle through stress testing benchmarks, it throttles quite severely just playing need for Speed most wanted after only 10 minutes. .reducing the game unplayable after that period.

https://www.youtube.com/watch?v=jIdxnFo3MuQ&feature=youtube_gdata_player

Although it seemed to play other less demanding games at playable frame rates, such as temple run 2, It was still throttling and dropping frames, the tablet actually locked up and froze due to the excessive heat whilst playing real racing 3.

Also although overall ui and app experience was generally snappy, google chrome ran very choppy and benchmarks were in the snapdragon s600 region..with antutu pushing 27, 000.
Quadrant 12, 000+ and sunspider in the mid 700ms...
These benchmarks are hardly respected around here for good reason (speak to exophase about antutu), but they do at least offer a comparison to competing socs...and it comes out quite average.

Also those benchmarks are a world away from the tegra 4 reference tablet and shield....and you wonder why nobody round here trusts nvidia?? :/

 

[xpea]

Also those benchmarks are a world away from the tegra 4 reference tablet and shield....and you wonder why nobody round here trusts nvidia?? :/

funny, the way you say it, I understand that the reference tablet and SHIELD have no problem (SHIELD have been on the wild for nearly a month and no throttling so far and with excellent performance, right?).
Up to now, only thing we can say is that Toshiba did a poor T4 implementation and NV did a good one with SHIELD.
Let's see how Asus and HP models will turn out before killing the baby...

 

[french toast]

funny, the way you say it, I understand that the reference tablet and SHIELD have no problem (SHIELD have been on the wild for nearly a month and no throttling so far and with excellent performance, right?).
Up to now, only thing we can say is that Toshiba did a poor T4 implementation and NV did a good one with SHIELD.
Let's see how Asus and HP models will turn out before killing the baby...

Who knows whats going on..but rest assured something IS up.
Shield is fan cooled and heatsink? In a more spacious chassis than a tablet.. So throw that one out of the window.

Th reference tablet I obviously dont know..but nvidia designed it...has it also got an elaborate heatsink? Likely dual channel 64bit 2133 ddr3L ram or at least 1866.

So I will only look at shipping products from third party OEMs, certainly nothing from nvidia.
The complete lack of design wins perhaps points this fact no?

Edit: well ive done some digging and found another review of that toshiba and it was laggy.

But I wanted another third party OEM such as the one you mentioned...so I looked up reviews and benchmark of the HP slatebook pro.
https://www.youtube.com/watch?v=oGgVAfmVgSc&feature=youtube_gdata_player

If you scroll down after the video and look at the information you will see a Selection of other tests carried out.

Basically the less than impressive benchmarks we saw with the toshiba excite are carried over with the HP.
Those are around or just above snapdragon s600..which as you are aware is qualcomms current higher midrange smartphone soc....in short pretty shit..and no where near nvidias marketing mumbo jumbo or their fan assisted frankenstein SHIELD.

Scores of 28, 000 antutu, 13, 000 ish quadrant, 13, 000 3d mark ice storm normal are hardly setting the world alight.

To be fair .ost short written reviews hav come out possitive f om what I read (I only quickly rad through the gadgeshow...and watched a video or two)..with the web browser performing much better than toshibas woefull implementation, but battery life was only rated for about a day...which doesnt sound bad untill you factor in the TWO batteries of tablet and keyboard.

Unfortunately I couldnt find any long gaming tests like the toshiba to gage the throttling like the toshiba review, so who knows on that one.

 

[xpea]

Who knows whats going on..but rest assured something IS up.
Shield is fan cooled and heatsink? In a more spacious chassis than a tablet.. So throw that one out of the window.

Th reference tablet I obviously dont know..but nvidia designed it...has it also got an elaborate heatsink? Likely dual channel 64bit 2133 ddr3L ram or at least 1866.

So I will only look at shipping products from third party OEMs, certainly nothing from nvidia.
The complete lack of design wins perhaps points this fact no?

So you throw away the good stuff and only keep the bad one ? doesn't look like a fair selection to me
Like it or not, you can buy SHIELD at newegg, best buy and so on, so it's a legitimate product.
NV may also come with is own 7" T4 tablet. You will also discard it ?

finally its pathetic reading again and again the same bulls*t argument of "completely lack of design win" when 4 of the TOP10 computer companies (HP, Asus, Acer, Toshiba) offer or will offer soon T4 tablets. Distortion of reality ?
Comon guys, you can hate NV as much as you want but don't make it black or white, it brings the discussion to a kit-garden level...

 

[Alexko]

funny, the way you say it, I understand that the reference tablet and SHIELD have no problem (SHIELD have been on the wild for nearly a month and no throttling so far and with excellent performance, right?).
Up to now, only thing we can say is that Toshiba did a poor T4 implementation and NV did a good one with SHIELD.
Let's see how Asus and HP models will turn out before killing the baby...

Silly Toshiba, all they needed to do was to put a fan in their <9mm-thick tablet. How did they miss that?

 

[french toast]

I have edited my post so please read...also re read my previous posts about tegra 4 not winning ANY smartphone design wins..like its predecessor which it is supposed to replace.

The only third party OEM designs have shown tegra 4 in a bad light and bare no resemblance to nvidias boasting or an elaborate fan cooled/heatsink designed nvidia device like shield...and I throw that out because it is not representative of what a tablet or smartphone OEM would implement, due to likely bulk and cost reasons.

Ive made my point.

 

[ams]

You are a funny guy french toast, especially for someone who has never even laid their hands on a single T4 product. While Toshiba's T4 tablet is clearly overpriced and uses a battery hogging screen (among other issues such as blurry camera that are not even related to T4), what's wrong with other products such as HP's Slatebook x2? That tablet clearly has a more powerful CPU and GPU compared to any S600 tablet, doesn't appear to have any stability problems, has a 1080p screen, and includes a keyboard and extra battery for $455-$479 USD shipped. NVIDIA's 7" Tegra Tab should provide very good performance for a very reasonable price (assuming it comes to market). Same with HP's Slate 8. Even the new Asus Transformer Pad Infinity is reported to have a base price of $399 USD. Most LTE-enabled S800 devices will be more expensive in comparison. To make any headway in smartphones requires a strong baseband strategy, so that is NVIDIA's biggest challenge in that area.

And just FYI, Google's Chrome browser is choppy even on Shield. NVIDIA's stock browser is wickedly fast though on all T4 devices. You can see this in the Slatebook x2 video linked to earlier in this thread.

 

[ams]

You keep pointing out that the CPU throttles more but I'm not sure what your point is.

The 3dmark Ice Storm Extreme test looped 10 times is maxing the CPU, then maxing the GPU, then maxing the CPU, then maxing the GPU, again and again nearly a dozen times (where both the CPU and GPU are progressively getting hotter and hotter and hotter without enough chance to cool down). Real world gaming would typically never cycle through maxing out the [quad-core] CPU and then maxing out the GPU over and over and over again in this manner. In fact, other than Need for Speed, even the less-than-perfect Toshiba tablet appears to handle virtually all Android games reasonably well and with smooth and playable framerates.

 

[xpea]

The only third party OEM designs have shown tegra 4 in a bad light and bare no resemblance to nvidias boasting or an elaborate fan cooled/heatsink designed nvidia device like shield...and I throw that out because it is not representative of what a tablet or smartphone OEM would implement, due to likely bulk and cost reasons.

Ive made my point.

well I don't agree but it's your opinion.

BTW, from this discussion, I was wondering, where are all these S800 tablets ? From general opinion here, this SoC is so much better than T4 that design wins should be counted by dozens, no ?
On T4 side, so far we have:
Toshiba Excite,
Asus Transformer pad infinity,
HP Slatebook X2 and Slate 8 Pro,
Acer TA2,
Vizio 10" that will ship for back to school.

And for S800 ? someone can put up a list ?

 

[Silent_Buddha]

Anandtech ran their 3D battery life test with GLBench 2.5 Egypt HD (1080p offscreen) test. That benchmark is way less stressful than 3dmark Ice Storm Extreme. And when capped at 30fps as Anand did for his 3D battery life test, the A6X GPU (or even Tegra 4's GPU) will absolutely cruise along on that benchmark.

On the other hand, by looping the entire 3dmark Ice Storm Extreme benchmark (including the CPU-heavy Physics test) 10 times, the CPU throttling was quite a bit worse than even the GPU throttling on that Toshiba tablet...and even before throttling, the average fps barely reaches 30fps. How is that representative of real world gaming on these devices? Games that are actually smooth and playable on these devices will have an average fps that is well above 30fps, and will not have ridiculously massive CPU throttling either. And don't forget that the Toshiba tablet's screen appears to consume a lot of power and is a battery life hog too. I would expect HP's Slatebook x2 tablet to have better real world performance and better battery life (even without the keyboard dock) because the screen is not as power hungry in comparison.

You do know that Anandtech loops and interates all of their 3D tests, right? They've been doing that for well over a decade now on desktops and mobile devices.

They do it specifically to take out any run where the benchmarked number isn't representative of the actual performance of the device whether due to loading of level assets or thermal throttling. Historically this meant that on desktop PCs, the first run was usually discarded because it often had a lower score than subsequent runs due to loading of game assets. But they've also made note of thermal throttling for some CPUs and GPUs over the years as well.

For Tegra 4 in a tablet form factor like the Toshiba, if they ever get around to benchmarking it, you can bet they will discard the early scores and report on the repeated run score if it exhibits the same behavior in their lab. Note that for Qualcomm's Snapdragon 800 preview (reference tablet design), they didn't have to do that.

Regards,
SB

 

[Xmas]

The 3dmark Ice Storm Extreme test looped 10 times is maxing the CPU, then maxing the GPU, then maxing the CPU, then maxing the GPU, again and again nearly a dozen times (where both the CPU and GPU are progressively getting hotter and hotter and hotter without enough chance to cool down). Real world gaming would typically never cycle through maxing out the [quad-core] CPU and then maxing out the GPU over and over and over again in this manner. In fact, other than Need for Speed, even the less-than-perfect Toshiba tablet appears to handle virtually all Android games reasonably well and with smooth and playable framerates.

Sure, but as I wrote above, for current low/medium complexity content where you hit 60 fps the only "performance" consideration is battery life.

Future games will at some point max out the GPU while potentially handing a substantial workload to the CPU. As for "hotter and hotter and hotter", surely throttling is meant to limit power draw to what the cooling system can dissipate at a certain temperature, thus stopping the temperature from rising. Once you are at that point, how you got there shouldn't really matter.
Are you arguing that just maxing out the GPU by running the graphics test in a loop for a sufficient amount of time (or the real usage scenario of running a highly demanding game for hours) would not result in the same level of GPU throttling?

 

[french toast]

@ams lol, you still are not getting it.

Tegra 4 runs hot in at least one third party tablet. ..cant play 3d games for very long. =epic fail.

Tegra 4 does not replicate the boasted benchmarks from nvidia in any tested third party products so far =nvidia miss leading consumers.

Hp slatebook x2 has not been tested long enough to answer the throttling issue in gaming..that doesn't mean it isnt doing it..just no one has done a thorough gsming test that I have read.
Power consumption - T4 definitely has a question mark over it, feel free to prove otherwise.

@xpea s800 has only just been launched, have no doubt it will make plenty of tablets, it already has been bench marked/reviewed in SMARTPHONES trouncing tegra 4...likely consuming less power whilst doing so.

Tegra 4 does not have on die baseband..so its either 3g or add an external one = consume more power.

Tegra 4 is a small performance bump over s600..a midrange smartphone soc launched 4 months ago.

 

[swaaye]

What do you guys think of T4's allocation of pixel/vertex units? How does this design compare to something like G70?

 

[ams]

Are you arguing that just maxing out the GPU by running the graphics test in a loop for a sufficient amount of time (or the real usage scenario of running a highly demanding game for hours) would not result in the same level of GPU throttling?

These ultra mobile handheld devices tend to have strict system level power and thermal constraints (ie. CPU + GPU + screen + anything else that consumes power). By constantly ramping up CPU, then GPU, then CPU, then GPU in a benchmark that is already quite tough even under cool operating conditions, the system level power and thermal constraints will clearly be reached, and as a result the system will be forced to clamp down on CPU and GPU clock operating frequencies. There is no way to get around this with a relatively high performance quad-core CPU and a relatively high performance GPU. With real world gaming, looking at games that are clearly playable on these devices, the power and thermal headroom should be noticeably better than this previously described loop with the vast majority of gaming applications because the CPU should not be maxing out in a similar fashion and because the GPU should have higher framerates to begin with.

 

[xpea]

@ams lol, you still are not getting it.

Tegra 4 runs hot in at least one third party tablet. ..cant play 3d games for very long. =epic fail.

Tegra 4 does not replicate the boasted benchmarks from nvidia in any tested third party products so far =nvidia miss leading consumers.

Hp slatebook x2 has not been tested long enough to answer the throttling issue in gaming..that doesn't mean it isnt doing it..just no one has done a thorough gsming test that I have read.
Power consumption - T4 definitely has a question mark over it, feel free to prove otherwise.

@xpea s800 has only just been launched, have no doubt it will make plenty of tablets, it already has been bench marked/reviewed in SMARTPHONES trouncing tegra 4...likely consuming less power whilst doing so.

Tegra 4 does not have on die baseband..so its either 3g or add an external one = consume more power.

Tegra 4 is a small performance bump over s600..a midrange smartphone soc launched 4 months ago.

here we go again... definitive words like "epic fail" and "misleading customers" despite the product just been launched.... you'll never get it
it's so easy to use this semantic.... I can do it too:
S800 is very late to market, announced in January 8th, but 8 months later, no tablet yet available = misleading customers and PR bullshit, that's why here we hate Qualcomm so much.
S800 has no design win in tablets, it's a product no OEM wants. It has to be something deeply wrong. it's a complete disaster.
S800 has poor CPU performance, they have to bump the frequency to 2.3GHz to be competitive and still they can't beat T4 at 1.8GHz. It's the bulldozer of SoC world.
S800 in Xperia Ultra Z has disappointing battery life from Engadget review, despite the Gargantua 3000maH battery. EPIC FAIL !
this product is FAIL after FAIL after FAIL, Qualcomm is doomed. Will they still be there in 2015 ?


edit: I don't know why I'm still replying. I feel like a troll now... BTW "never argue with a troll, they will drag you down to their level and beat you with experience"
The crazy thing is that I'm not a NV fan (I don't own any green graphic card) but I can't keep my mouth shut up when I read so much hate and biased opinions

 

[Blazkowicz]

Tegra 4 has no PCIe.
If they wanted to make a high powered SoC they should have gone that extra line (Tegra 3 had four lines) which would have made it useful in laptops - or netbooks since they have to go by that name when they are small and cheap.