NVIDIA Tegra Architecture

What appears to be the first glimpse of the Tegra 4i, 2.3 GHz CPU and OpenGL 2.0 have arisen on GLBenchmark.

T-Rex offscreen result of 17.6 fps, so Snapdragon 600 level. If this chip ends up in $100 to $200 dollar phones, as Nvidia forecast it should prove popular.

http://gfxbench.com/device.jsp?benchmark=gfx27&D=NVidia+Tegra+GK20A+(laguna)&testgroup=overall

And it's been pulled....

3D looks in line with expectation. But at 2.3GHz, I want to see much higher CPU performance on A9r4 than 1.5~1.6GHz S600
 
3D looks in line with expectation. But at 2.3GHz, I want to see much higher CPU performance on A9r4 than 1.5~1.6GHz S600

It's a graphics benchmark, so the point of reference was the GPU. In terms of CPU, Krait has higher IPC than a A9, so it will interesting to see how much that changes in the R4 revision, also S600 ranges from 1.7 to 1.9 GHz.
 
Wasn't Tegra 4i supposed to have launched by now? NVIDIA claimed something about accelerating their schedule for Tegra 4i. Are there also any devices using their i500 baseband? Their baseband should have launched alongside Tegra 4 in Q2, even if Tegra 4 didn't quite keep its schedule either.
 
GK20A is the first Kepler core for Tegra.

It looks like the GPU speed was set to a similar frequency as what was demonstrated many months earlier.
 
GK20A is the first Kepler core for Tegra.

It looks like the GPU speed was set to a similar frequency as what was demonstrated many months earlier.

2.3 GHz may be the top speed bin, but that is quite a jump over the actively cooled Shield. I wonder if this improvement is just due to the move HPM, as the T4 was already built on a HKMG process, or is the newer revision of the A15 more power efficient.

Was the Logan demo board actively cooled?, the pictures I've seen online. don't appear to show any active cooling.
 
On a side note, does anyone know why Surface 2's Tegra 4 GPU performance in GFXBench 2.7 T-Rex HD is so much better Onscreen (http://images.anandtech.com/graphs/graph7439/59151.png) vs. Offscreen 1080p (http://images.anandtech.com/graphs/graph7439/59152.png)? The performance is 41% better (!) in Onscreen mode! Note that the Surface 2 is using a 1080p screen resolution, so the result is puzzling.

The GLbenchmark site is reporting a screen size of 1371 x 771 for the Surface 2. It looks like there may be a glitch, and the onscreen test isn't rendering to native resolution. But that isn't reflected on the result on GLB, it scores 16.5 offscreen and 15.5 onscreen. Maybe Anand mixed the data between T-Rex Onscreen and GFXBench 2.7 T-Rex HD Fixed Timestep - Onscreen :, the fixed timestep result is much higher 23.1 fps

http://gfxbench.com/device.jsp?benchmark=gfx27&D=Microsoft+Surface+2&testgroup=system
 
Last edited by a moderator:
2.3 GHz may be the top speed bin, but that is quite a jump over the actively cooled Shield. I wonder if this improvement is just due to the move HPM, as the T4 was already built on a HKMG process, or is the newer revision of the A15 more power efficient.

Was the Logan demo board actively cooled?, the pictures I've seen online. don't appear to show any active cooling.

Yes, the CPU speed/efficiency should be and really needs to be better on Logan than Wayne. As for the Logan demo board, I'm not sure what NVIDIA did.
 
The GFXBench database doesn't show such a discrepancy in the main T-Rex HD test, but there is quite a difference in the fixed timestep and Egypt tests.
http://gfxbench.com/device.jsp?D=Microsoft+Surface+2&testgroup=overall&benchmark=gfx27&var=score

Interestingly, the screen resolution is listed as 1371 x 771, which is 1080p with 140% scaling. Could it be that the initial WinRT version of the test had a bug which would make it render at the scaled resolution?

I believe that the GFXBench 2.7 T-Rex HD Fixed Timestep - Onscreen score was posted erroneously, in place of the regular onscreen result. The fixed timestep is listed as 23.1 FPS vs T-Rex onscreen =15.5 FPS on GLB
 
That seems unlikely to me, and it doesn't explain why the two Fixed Timestep results show a similar difference, either.
 
That seems unlikely to me, and it doesn't explain why the two Fixed Timestep results show a similar difference, either.

Well, I've quashed my own theory. This Youtube video of the Surface 2 running, the regular T-Rex 2.7 Onscreen also shows 24 FPS.

2.25 for the result.
http://www.youtube.com/watch?v=VdI4vTl1PvM

The idea that the reported resolution of only 1371 x 771, is the culprit definitely sounds convincing. Weirdly, the Surface Pro 2, which also has a 1080P display is reporting resolutions of 1371 x 771 & 1600 x 900 under GLB's system tab. This isn't boosting its Onscreen score, probably because it's already scoring close to 60 FPS, so is affected by V-Sync.

http://gfxbench.com/device.jsp?benchmark=gfx27&D=Microsoft+Surface+Pro+2&testgroup=system

I found another device running Windows RT, a Nokia P6095 with a Qualcomm S800, which also reports a display resolution of 1371 x 771. But I bet good money, that this is actually an early version of the Lumia 2520, which uses a 1080P display. Again the Onscreen result is much higher than Offscreen.
 
Last edited by a moderator:
Back
Top