NVIDIA Tegra Architecture

[Jubei]

FYI, the GLBenchmark 2.5 Egypt HD Offscreen 1080p results for Adreno 320 are slower than even the crippled Dalmore Tegra 4 test system. The top GLBenchmark results for Adreno 320 are highly misleading too. According to Anandtech, the LG Optimus G (with Adreno 320 GPU) "can't complete a single, continuous run of GLBenchmark 2.5 - the app will run out of texture memory and crash if you try to run through the entire suite in a single setting. The outcome is that the Optimus G avoids some otherwise nasty throttling". When using an Adreno 320 equipped smartphone that is actually able to achieve a continuous run of GLBenchmark 2.5 by throttling (such as the Google Nexus 4 phone), the GLBenchmark 2.5 Egypt HD Offscreen 1080p results for Adreno 320 are only 18 fps! See the data here: http://images.anandtech.com/graphs/graph6425/51288.png

You paint a highly misleading picture. The Nexus 4 throttles at a fairly low temperature (36 celsius IIRC) wich is more of a case of Qualcomm/LG setting the governor too low rather than abnormal heat being produced. Given that Optimus G also has problems im going to say this is related to whatever LG has done

There are 5-6 other Adreno 320 phones in glbenchmark, are you insinuating they are all crashing when running the test?

 

[french toast]

Ams. Ive just read the optimus g performance preview on anandtech...its praises the adreno 320 in gl benchmark as awesome...no mention of crashing or issues in that test...however perhaps you reference the nexus review?..
http://www.anandtech.com/show/6305/lg-optimus-g-performance-preview

I severley doubt any serious issues with adreno 320 too be honest considering the strong (ish) performance of last gen adreno 225, the likely rubbish software optimisatons lg is known for and googles attemps to appease oems in regards to both soc performance.. (galaxy nexus ti 4460 underclocked) and features like low storage 8/16gb.
Also qualcomm did mention a significant power saving with adreno 330 over "last generation" despite the promised performance improvements and likely only minor improvements in manufacturing efficiency (15% with 28nm hkmg??)

All in all adreno 320 only being able to hit 18fps due to excessive throttling seems unrealistic IMO.

 

[ams]

Comparing Dalmore Tegra 4 against Nexus 7 Tegra 3

Take a look at the Dalmore result (http://webcache.googleusercontent.com/search?q=cache:BHeFcdcCZt0J:www.glbenchmark.com/phonedetails.jsp%3Fbenchmark%3Dglpro25%26D%3DDalmore%2BDalmore%26testgroup%3Dlowlevel+http://www.glbenchmark.com/phonedetails.jsp%3Fbenchmark%3Dglpro25%26D%3DDalmore%2BDalmore%26testgroup%3Dlowlevel&cd=1&hl=en&ct=clnk&gl=uk) and compare it to the Asus Transformer Pad TF700T Infinity median result (http://www.glbenchmark.com/phonedet...former+Pad+TF700T+Infinity&testgroup=lowlevel).

Comparing Dalmore to the Asus Transformer Pad TF700T Infinity median result, I see the following differences:

Fill rate - Offscreen (1080p) : 1.20x
Triangle throughput: Textured - Offscreen (1080p) : 1.00x
Triangle throughput: Textured, vertex lit - Offscreen (1080p) : 1.59x
Triangle throughput: Textured, fragment lit - Offscreen (1080p) : 1.00x

So I was a bit off in saying that the texture fillrate is equalized between the Dalmore Tegra 4 test system and Tegra 3 (although there is only a small difference in texture fillrate), but the triangle texture rates are indeed equalized in two out of three instances between the Dalmore Tegra 4 test system and the Tegra 3 variant used in the Transformer Pad Infinity.

 

[ams]

You paint a highly misleading picture. The Nexus 4 throttles at a fairly low temperature (36 celsius IIRC) wich is more of a case of Qualcomm/LG setting the governor too low rather than abnormal heat being produced. Given that Optimus G also has problems im going to say this is related to whatever LG has done

Here is what Anandtech had to say about the Nexus 4 (http://www.anandtech.com/show/6425/google-nexus-4-and-nexus-10-review/2): "Despite being based on the same hardware, the Optimus G is able to post a much higher score here than the Nexus 4. The explanation is simple: the Optimus G can't complete a single, continuous run of GLBenchmark 2.5 - the app will run out of texture memory and crash if you try to run through the entire suite in a single setting. The outcome is that the Optimus G avoids some otherwise nasty throttling. The Nexus 4 on the other hand manages to complete everything, but likely quickly throttles its clocks down due to thermal constraints. The Nexus 4 was really hot by the end of our GLBenchmark run, which does point to some thermal throttling going on here. I do wonder if the Snapdragon S4 Pro is a bit too much for a smartphone, and is better suited for a tablet at 28nm."

I'm not saying that the Nexus 4 throttling is a serious issue in day-to-day use, I am just offering Anandtech's explanation as to why the Optimus G was able to post much higher GPU benchmark scores than the Nexus 4.

 

[Ailuros]

IMHO for those Dalmore results to make sense the frequency used must have been relatively low. If it was clocked at similar frequencies as the T3 ULP GF than there might be some serious bottleneck in the geometry pipeline that only extravagant frequencies can rectify.

It's anything but uncommon that early test platforms use quite humble frequencies; engineers know how results would look like at final target frequencies.

 

[ams]

All in all adreno 320 only being able to hit 18fps due to excessive throttling seems unrealistic IMO.

It is what it is. The reason why LG Optimus G was able to post a relatively impressive 29fps on this same test (vs. the 18fps for the Nexus 4) is because the GPU benchmarks were run in parts (rather than continuously) which didn't push the Optimus G into thermal throttling. Here is another explanation from Anandtech (http://www.anandtech.com/show/6440/google-nexus-4-review/3): "For those wondering why the LG Optimus G wasn't affected in spite of it having the same platform, the reason is because the results from the Optimus G were run in parts due to some instability affecting its ability to run a complete set of tests without crashing. The Nexus 4 has newer drivers that don't crash during a full GLBenchmark 2.5 run but as a result run the device long enough for thermal throttling to kick in".

 

[Jubei]

Here is what Anandtech had to say about the Nexus 4 (http://www.anandtech.com/show/6425/google-nexus-4-and-nexus-10-review/2): "Despite being based on the same hardware, the Optimus G is able to post a much higher score here than the Nexus 4. The explanation is simple: the Optimus G can't complete a single, continuous run of GLBenchmark 2.5 - the app will run out of texture memory and crash if you try to run through the entire suite in a single setting. The outcome is that the Optimus G avoids some otherwise nasty throttling. The Nexus 4 on the other hand manages to complete everything, but likely quickly throttles its clocks down due to thermal constraints. The Nexus 4 was really hot by the end of our GLBenchmark run, which does point to some thermal throttling going on here. I do wonder if the Snapdragon S4 Pro is a bit too much for a smartphone, and is better suited for a tablet at 28nm."

I'm not saying that the Nexus 4 throttling is a serious issue in day-to-day use, I am just offering Anandtech's explanation as to why the Optimus G was able to post much higher GPU benchmark scores than the Nexus 4.

There is a difference between saying two specific devices have issues and "Adreno 320 results are misleading in glbenchmark".

HTC DNA, Asus Padfone 2, Xiaomi MI 2 and upcoming Xperia Z and Zl all use Adreno 320 and the two first (since they have been released) can run glbenchmark without throttling or crashes

So whatever the issue is its with LG, not with Adreno 320 itself

NordicHardware benched Asus Padfone 2 and it reached 38 degrees celsius under load with no throttling. In comparison Nexus 4 throttles already at 36 degrees celsius

And it finished glbenchmark without crashing

Also Brian from Anandtech pretty much agrees Nexus 4 suffers from aggressive governor

http://img.tapatalk.com/d/12/12/01/ru8u6u9a.jpg

 

[french toast]

It is what it is. The reason why LG Optimus G was able to post a relatively impressive 29fps on this same test (vs. the 18fps for the Nexus 4) is because the GPU benchmarks were run in parts (rather than continuously) which didn't push the Optimus G into thermal throttling. Here is another explanation from Anandtech (http://www.anandtech.com/show/6440/google-nexus-4-review/3): "For those wondering why the LG Optimus G wasn't affected in spite of it having the same platform, the reason is because the results from the Optimus G were run in parts due to some instability affecting its ability to run a complete set of tests without crashing. The Nexus 4 has newer drivers that don't crash during a full GLBenchmark 2.5 run but as a result run the device long enough for thermal throttling to kick in".

Yea I can see where your coming from from what you have put on the table... but as pointed out above rather well, those devices seem to have there own issues and dont seem to show the true performance of adreno 320.

Im now very excited for adreno 330...I really hope anandtech does a re run of the power consumption of all main socs...

 

[Lazy8s]

People surprised by nVidia's relative standing in mobile graphics performance, considering their "desktop reputation", have always seemed to miss the fact that the mobile market is far more competitive. nVidia haven't had to face PowerVR on the desktop for a long while; things were different back when they did, with regard to relative, comparable performance. And Tegra 4 won't be upsetting the top of the charts.

Its GeForce does, however, compare more favorably to Mali and especially Adreno this time around, so it is a commendable design. The die area comparisons to A6X, though, are meaningless as mentioned in, now, countless prior discussions. Give nVidia the silicon budget of Apple and they will undoubtedly strike a better balance for power efficiency, leading to some increased performance, but the power consumption and thermal limits being approached by their architecture preclude any kind of proportionate performance increase in such comparisons.

 

[french toast]

People surprised by nVidia's relative standing in mobile graphics performance, considering their "desktop reputation", have always seemed to miss the fact that the mobile market is far more competitive. nVidia haven't had to face PowerVR on the desktop for a long while; things were different back when they did, with regard to relative, comparable performance. And Tegra 4 won't be upsetting the top of the charts.

Its GeForce does, however, compare more favorably to Mali and especially Adreno this time around, so it is a commendable design. The die area comparisons to A6X, though, are meaningless as mentioned in, now, countless prior discussions. Give nVidia the silicon budget of Apple and they will undoubtedly strike a better balance for power efficiency, leading to some increased performance, but the power consumption and thermal limits being approached by their architecture preclude any kind of proportionate performance increase in such comparisons.

Well we await for final clocks and power consumption...but if nvidia just breaks even with a6x on pure graphics...as long as the other factors such as the cpus....then it can be considered a complete success...

 

[ams]

CES 2013 interview with NVIDIA's CEO: http://www.youtube.com/watch?v=LvlyW1Rx3zE . Q&A discussion includes questions about Tegra 4, Shield, Cloud Gaming, Console Gaming, Project Denver, Smartphones, etc.

 

[Ailuros]

Well I wasted about 38 minutes of my time for nothing new in the hope to hear something about Project Denver. Jensen doesn't intend to create a new ecosystem, for now at least

Other than that Q&As/interviews with all kinds of companies and their respective representatives are the same. Nothing that a close observer wouldn't know already and a whole damn lot of useless marketing wash with a topping of substance free yadda yadda. "Why are you not excited damnit it's project oompa loompa..."

 

[ams]

Some clever and funny youtube videos here:

http://www.youtube.com/watch?v=yp0ajn99BEg&feature=player_embedded

http://www.youtube.com/watch?v=xs9fG9v7zVU

 

[Ailuros]

No idea where the material comes from or how accurate it is, but in the first video it's stated pretty clear "slightly faster than iPad4", else within current expectations unless we'll see any surprises.

As for the 2nd video and it's all about consoles or to stretch it a bit about home entertainment devices, here's what Gabe Newell said just recently: http://www.polygon.com/2013/1/30/39...m-boxs-biggest-threat-isnt-consoles-its-apple

 

[Ailuros]

http://translate.google.gr/translat...angeblich-erster-abnehmer-des-nvidia-tegra-4/

Obviously nothing is set in stone yet and it's way too early to jump to any conclusions, but Acer and Asus having supposedly 2nd thoughts isn't necessarily a good sign.

 

[Deleted member 13524]

http://translate.google.gr/translat...angeblich-erster-abnehmer-des-nvidia-tegra-4/

Obviously nothing is set in stone yet and it's way too early to jump to any conclusions, but Acer and Asus having supposedly 2nd thoughts isn't necessarily a good sign.

I guess the heatsink and fan in the Shield prototypes scared the hell out of them all.

 

[Ailuros]

I guess the heatsink and fan in the Shield prototypes scared the hell out of them all.

Don't run away I've got my HDR on

Jokes aside, NV knows and most of us also that it's not going to be an easy battle against giants like Qualcomm and Samsung. If NV manages to get at least one high volume deal with Grey I suppose they're going to do fine for where they are now.

 

[ninelven]

What is Grey supposed to be exactly? Dual core A-15?

 

[Helmore]

What is Grey supposed to be exactly? Dual core A-15?

I thought Grey was pretty similar to Tegra 3, just on 32/28 nm and with the baseband integrated on the chip. Quad core Cortex-A9 in other words.

 

[Ailuros]

I thought Grey was pretty similar to Tegra 3, just on 32/28 nm and with the baseband integrated on the chip. Quad core Cortex-A9 in other words.

Considering Qualcomm and Mediatek's mainstream SoCs both rely at the moment at quad A7 CPUs at 1.2GHz, it's not necessarily a disadvantage.