Qualcomm Krait & MSM8960 @ AnandTech

look at the Tegra4 LTE chip, the i500.

It was said that the i500 is only 40% as large as a normal baseband-chip. Despite that it has roughly an die-size of 1/4 of the Tegra4 itself, or maybe 20mm². Therefore a normal baseband-chip could be 50mm² large. Therefore all of the 40mm² size difference between S600 and S800 could be due to the LTE-system.

Link: http://www.engadget.com/2013/01/06/nvidia-tegra-4-official

Those dies are for illustrative purposes only, and almost certainly not to scale. Keep in mind that the entire T4i SoC is close to 60 mm^2 including i500 baseband modem. And the die size area of S800's integrated modem is unknown. Considering that the GPU operating frequency is "only" ~ 450MHz (according to Anandtech), Adreno 330 in S800 surely must have more execution units and larger die size area compared to Adreno 320.
 
330 for sure has more ALU lanes than 320, otherwise the performance difference wouldn't make any sense; however it still stands that ALUs are relatively cheap in hw and they CAN'T under any circumstance account for a 50% SoC die estate increase.

If memory serves well a FP32 unit under 28LP should be around 0.01mm2 for synthesis alone. Even if you'd say that the 330 added say 10mm2 in die area that's a whole damn LOT in terms of just additional FLOPs. Granted that's not all the logic they likely added, yet again the TMU amount is the same.
 
Those dies are for illustrative purposes only, and almost certainly not to scale. Keep in mind that the entire T4i SoC is close to 60 mm^2 including i500 baseband modem. And the die size area of S800's integrated modem is unknown. Considering that the GPU operating frequency is "only" ~ 450MHz (according to Anandtech), Adreno 330 in S800 surely must have more execution units and larger die size area compared to Adreno 320.


http://www.nvidia.com/docs/IO/116757/NVIDIA_i500_whitepaper_FINALv3.pdf

7 x 7mm package, 40% the size of comparable conventional 4G modem
 
Package size is irrelevant.

A bit more than one year before tape-out, i500 was expected to be measure ~15mm2 with about ~10mm2 of that being the baseband itself (as opposed to peripheral logic and I/O). The MDM9200 was 91.4mm2 on 45nm so you might indeed expect a 150Mbps baseband on 28nm to add about 30-40mm2 (or 3-4x as much as i500's baseband).

So it does seem like the die size of the AP part of S800 didn't increase much (but I'd be very surprised if it didn't increase somewhat at least, and they could have saved some space for new features by optimising different blocks of course).
 
I just want to make a fairly obvious but important point...(again) the tegra 4 development tablet as I read somewhere is equipped with dual channel ddr3L 1866 and a high end sand disk nand memory/high end controller.

Whilst im not aware of the quality of the qualcomm s800 reference tablets nand..we do know it carries an inferior dual channel lpddr3 800 memory.
http://www.slideshare.net/mobile/caulfield2600/tegra-4-outperforms-snapdragon-17176163

This would surely bring the cpu/system scores more on an even keel had both reference platforms used the same momory..also bumping up s800 graphics scores a little.

#ive said my peice.
 
Last edited by a moderator:
Can anyone comment on how the GLBench battery test works? Is it somehow performance normalized?
I can figure out it says 100% brightness but it doesn't specify if it's the 30fps or 60fps bench. My guess is on the latter. A fatter GPU would certainly allow it to run at lower frequencies in any case, achieving better battery life.

But really, it's quite a feat. Too bad most sites don't do full throttle battery drain tests, that'd be more interesting.
 
I have to take back what I said about normalization - every test but the 3D run had both full brightness and normalized brightness variants. But since those results were identical with the other S4 I think it's fair to say the display performance is the same, as well as the battery being the same. So I'd call this a big win for S800.

Some people I've talked to have been rationalizing Intel's GPU decisions in BayTrail-T by saying that they're going for small GPU to keep power budget small. Clearly this demonstrates what we already knew, that bigger GPU can mean better power consumption, not worse. This is especially true with so many mobile games being frame locked. All they really need is more options to adjust resolution and quality.
 
When was it said that original Adreno 320 had 64 ALU's and the one in s600 has 96?
http://www.gsmarena.com/china_mobiles_oppo_find_5_gets_upgraded_with_snapdragon_600-news-6310.php

Sounds like crossed wires to me. Qualcomm wouldn't release a GPU with the model name but 1.5x as many ALUs. Benchmarks show as much, the Adreno 320 in S600 devices performs more or less like the Adreno 320 in S4Pro devices, barring any advantages due to clock boosting. Adreno 330, on the other hand, would fit a 1.5x ALU scaling.
 
The comparison is to an Exynos Octa equipped S4, going off old tests the S600 S4 would still use less battery, but the S800 clearly fits just fine in a large phone. :)
 
Back
Top