NVIDIA Tegra Architecture

[Exophase]

Yeah, pretty confused about what this alleged delay is supposed to mean. Is this a delay that already happened or a new one?

And how exactly is pushing out Tegra 4 by a quarter pulling in Tegra 4i by two?

 

[ams]

The Tegra business (including Android devices, Windows RT devices, automotive, and embedded devices) units sold per fiscal qtr is as follows ( http://news.cnet.com/8301-1035_3-57579098-94/nvidias-tegra-mobile-chip-business-hits-a-wall-with-growth/ ):

Q3 FY11: 0.6 million
Q4 FY11: 1.8 million

Q1 FY12: 5.1 million
Q2 FY12: 6.2 million
Q3 FY12: 8.0 million
Q4 FY12: 4.2 million
Total FY12: 23.5 million

Q1 FY13: 5.9 million
Q2 FY13: 7.7 million
Q3 FY13: 10.3 million
Q4 FY13: 8.1 million
Total FY13: 32.0 million

Q1 FY14 (Est): 3.8 million

Considering that shipments of Tegra 4 will not begin until Q2 FY14, and shipments of Tegra 4i will not begin until Q4 FY14, while shipments of Tegra 3 already started ramping down in Q1 FY14, it makes sense that overall Tegra shipments will be depressed in Q1 FY14, and it makes sense that overall Tegra shipments will likely be flat in FY14 relative to FY13. The transition from Q4 FY13 to Q1 FY14 is somewhat similar to the transition from Q3 FY12 to Q4 FY12 in the sense that prior generation Tegra production ramps down more quickly than next generation Tegra production ramps up.

 

[ams]

Yeah, pretty confused about what this alleged delay is supposed to mean. Is this a delay that already happened or a new one?

This three month delay already happened. NVIDIA originally expected to ship Tegra 4 in Q1 FY14, but actual shipments will begin in Q2 FY14.

And how exactly is pushing out Tegra 4 by a quarter pulling in Tegra 4i by two?

I'm not really sure why or how that would be the case. NVIDIA probably did prioritize Tegra 4i to some extent in order to bring it to market as quickly as possible, but it is unclear why there is a two quarter benefit in time to market.

 

[ams]

The Android Headlines article is putting a bit too much emphasis on having an LTE solution integrated into the mobile chipset though. Yes, that was a major advantage for Qualcomm, but Qualcomm's advantage was that they were pretty much the only one with decent LTE chips. They talk about the next Nexus 7 using a Qualcomm chip because it has integrated LTE, but chances are Google/ASUS will use the Snapdragon 600 and that doesn't have integrated LTE.

Assuming that Google wants to update the Nexus 7 by ~ middle of this year, it makes sense that they would go with something like S4Pro/S600. After all, it would make little sense to use a Tegra 4 SoC in a low cost Nexus tablet that dramatically outperforms Google's own high end Nexus 10 tablet with respect to CPU and GPU performance! In addition to logistics and proper product placement, S4Pro/S600 has some time to market advantage compared to Tegra 4 (even if the difference in time to market may only be two or three months and may be a moot point if the refreshed Nexus 7 tablet arrives in the middle of the year), and likely has a peak power consumption advantage compared to Tegra 4 too.

 

[Helmore]

This three month delay already happened. NVIDIA originally expected to ship Tegra 4 in Q1 FY14, but actual shipments will begin in Q2 FY14.

I'm not really sure why or how that would be the case. NVIDIA probably did prioritize Tegra 4i to some extent in order to bring it to market as quickly as possible, but it is unclear why there is a two quarter benefit in time to market.

I think Jen-Hsun Huang just told his investors to expect Tegra 4 in Q3 instead of Q2, but that we can expect to see Tegra 4i in Q3 as well.

EDIT: I now see that you're talking about financial years for NVIDIA and I'm not sure what dates that covers. It could be that you're saying the same as I am....

 

[ams]

With respect to NVIDIA's fiscal qtr/year translating to calendar mth/year, Q1 FY13 = Feb-Apr CY12, Q2 FY13= May-Jul CY12, Q3 FY13 = Aug-Oct CY12, Q4 FY13 = Nov-Jan CY12/13, and Q1 FY14 = Feb-Apr CY13. The expectation is that Tegra 4 will not be available until late Q2 2013 calendar qtr/year at the earliest.

 

[Helmore]

The expectation is that Tegra 4 will not be available until late Q2 2013 calendar qtr/year at the earliest.

That wouldn't really be a delay in my book. I expected to see Tegra 4 around the end of Q2 at the earliest anyway. A delay would be Q3.

 

[ams]

When NVIDIA said that Tegra 4 was delayed by 3 months, they meant relative to their original expectations last year. So Tegra 4 was originally expected to be shipping by Q1 CY13.

 

[Turbotab]

http://www.engadget.com/2013/04/12/next-gen-mobile-kepler-graphics-demoed/

Surely running on Kayla.

 

[RecessionCone]

Surely running on Kayla.

That would require recompiling Battlefield for ARM, as well as having a complete DX11 stack running on ARM. I'd be surprised if all that was up and running.

 

[Turbotab]

That would require recompiling Battlefield for ARM, as well as having a complete DX11 stack running on ARM. I'd be surprised if all that was up and running.

This is the original article, no mention of Tegra or SoC, just 'Kepler Mobile' which could mean anything, there is a picture of what appears to be a PCI-e device

http://venturebeat.com/2013/04/11/nvidia-shows-off-stunning-graphics-with-kepler-mobile-chip/

 

[Ailuros]

The Tegra business (including Android devices, Windows RT devices, automotive, and embedded devices) units sold per fiscal qtr is as follows ( http://news.cnet.com/8301-1035_3-57579098-94/nvidias-tegra-mobile-chip-business-hits-a-wall-with-growth/ ):

Q3 FY11: 0.6 million
Q4 FY11: 1.8 million

Q1 FY12: 5.1 million
Q2 FY12: 6.2 million
Q3 FY12: 8.0 million
Q4 FY12: 4.2 million
Total FY12: 23.5 million

Q1 FY13: 5.9 million
Q2 FY13: 7.7 million
Q3 FY13: 10.3 million
Q4 FY13: 8.1 million
Total FY13: 32.0 million

Q1 FY14 (Est): 3.8 million

Considering that shipments of Tegra 4 will not begin until Q2 FY14, and shipments of Tegra 4i will not begin until Q4 FY14, while shipments of Tegra 3 already started ramping down in Q1 FY14, it makes sense that overall Tegra shipments will be depressed in Q1 FY14, and it makes sense that overall Tegra shipments will likely be flat in FY14 relative to FY13. The transition from Q4 FY13 to Q1 FY14 is somewhat similar to the transition from Q3 FY12 to Q4 FY12 in the sense that prior generation Tegra production ramps down more quickly than next generation Tegra production ramps up.

As long as there's automotive/embedded added to the mix which could be anything but T3 SoCs, those numbers aren't unfortunately particularly useful.

 

[silent_guy]

As long as there's automotive/embedded added to the mix which could be anything but T3 SoCs, those numbers aren't unfortunately particularly useful.

Do they still use Tegra 2 for automobile? If so, why aren't those relevant to the Tegra business?

 

[Ailuros]

Do they still use Tegra 2 for automobile? If so, why aren't those relevant to the Tegra business?

Because I was wondering about T3 sales coincidentially in my former posts based on the Hexus claims? When they say the Nexus7 sold 6M units and represents 60% of T3 sales what do you end up with? I'll re-phrase: even 10M per year just for tablets and one smartphone design win for Tegra3 are a very low number. The Nexus claim doesn't make sense to me and the chart where ams took the results of states "tegra unit shipments", which doesn't guarantee me just T3 sales.

 

[ams]

Surely running on Kayla.

The GPU was most likely the same as what was used in the Tegra 5 "Kayla" development platform, but the CPU was said to have been downclocked to match the SPECInt performance of the Tegra 5 "Logan" CPU, which implies that it was an x86 CPU. I suppose that is fine because the purpose of the demonstration was to show off GPU performance and feature set rather than CPU performance, but it does beg the question about how easy or difficult it is to recompile x86 software to run on ARM processors.

 

[Deleted member 13524]

I didn‘t notice Tegra 4 was that much late to the party.
Can nVidia‘s Tegra division survive a whole year without a new SoC?

 

[ams]

I'm not sure what you mean. Tegra 4 was delayed by about three months, not one year. The first Tegra 4-powered devices should come to market at end of Q2 2013, with the majority of devices coming out in Q3 and Q4 2013.

 

[french toast]

It doesn't matter as its dead on arrival.

Tegra 4i however is much more interesting for a upper midrange phone.

 

[Exophase]

It doesn't matter as its dead on arrival.

Tegra 4i however is much more interesting for a upper midrange phone.

I see Tegra 4i as being interesting if and only if it can reach its upper clocks under reasonable power consumption. I'm not super optimistic. You don't see Samsung running their 32nm Cortex-A9s at anywhere remotely close to these clock speeds. I doubt TSMC 28nm has such an advantage to allow such higher scaling without extreme power consumption.

Anyone want to start taking guesses on how much power a single Tegra 4i A9 will consume at 2.3GHz? My bet is > 2W. Hopefully by then Android will be better at power gating at least some unused cores, otherwise you can count on the leakage of the other three cores adding a bit more to that total..

It'll probably still look great at single threaded benchmarks. I'm sure Sunspider will still be popular then.

 

[french toast]

@ exophase. .going off topic for a second..what is your view on kraits supposedly poor sunspider performance in comparison to cortex A9s?

Im aware most of it is due to vendor specific java optimisations. .but even using the same vendor and software ie galaxy s3..krait doesnt seem to flex its muscles in this test.

Whats your thoughts here?