One of the most frustrating things about the mobile system-on-chip landscape is that the players involved are very reluctant to give out things like transistor counts and chip sizes, likely for competitive reasons. However, a question that has lingered on my mind for quite some time -- and one that may be of interest to investors and those who follow this space -- is the approximate chip size of NVIDIA's (NASDAQ:NVDA) Tegra K1. This is important, because die size is a key driver of cost and, ultimately, the viability of the product in various markets.
NVIDIA has specifically stated that it is targeting high performance niches
NVIDIA is pretty upfront about the fact that its Tegra chips will be targeted toward high performance segments of the mobile market, particularly ones where graphics performance is important. Now, in this world you never get something for nothing, and in order to stick the kind of graphics performance onto a chip that NVIDIA has requires a heap of transistors, which means die space. This means that Tegra K1 probably sells for quite a bit more than your average mobile system-on-chip.
In light of this fact, targeting high performance niches makes sense. Most Android tablets and smartphones are really aimed to be as low cost as possible, so the chip-vendors that have succeeded in this space have typically done so through tight integration of as much functionality as possible into a single chip for as cheaply as possible. Qualcomm has been particularly successful here through best-in-class integration of modem, connectivity, and other IP.
However, NVIDIA's game is high performance, relatively high ASP markets, and the Tegra K1 -- from the initial performance numbers -- looks like it does a good job there.
How big is a 192-core GPU?
NVIDIA's Tegra K1 sports five ARM (NASDAQ:ARMH) Cortex A15 CPU cores, a 192-core Kepler-based GPU, and all of the standard features you'd find on a high end mobile chip all built on TSMC's (NYSE:TSM) 28 HPm manufacturing process. The prior generation Tegra 4 (4+1 core Cortex A15, 72-core GeForce ULP GPU) weighed in at a die size of 80 square millimeters, but so far we have no clue how big the Tegra K1 is.
So, in the absence of information, it's time to try to make an educated guess. By far the biggest portion of this chip will be the 192-core Kepler GPU. Not only is each "core" of this GPU more advanced than the prior Tegra 4's, but the K1 packs nearly three times as many of them. In fact, the 192-core Kepler configuration is more or less identical to the GeForce GT 630, which was a 192-core Kepler discrete GPU. The die size for this chip was 79 square millimeters on TSMC's 28-nanometer HKMG process.
Figuring out how big Tegra K1 is
Now, let's take a look at Apple's A7, which we know sports a die-size of about 102 square millimeters, to get a rough idea of how much the "rest" of the chip could weigh in at. According to Chipworks' picture, the CPU complex makes up about 16.3% of the A7's entire die (~16.6 square millimeters), the GPU complex takes up about 20.6% of the die (~21 square millimeters). There's also an SRAM block in there that acts as an L3 cache that takes up about 5.9% of the die (~6 square millimeters).
If we assume the CPU complexes are roughly the same size and the "rest" of the chip ex-L3 cache are similar, then substituting the PowerVR graphics block with the much beefier Kepler GPU we can figure out approximately how big Tegra K1 is. Now, while the desktop GPU implementation of Kepler weighs in at 79 square millimeters, we can assume that NVIDIA's teams have figured out how to optimize it more for density, so let's call it 60-65 square millimeters for the GPU block. This leads us to roughly a die size of 135 -- 140 square millimeters.
Foolish bottom line
The Tegra K1 is likely a big chip relative to its peers, but it packs a pretty serious amount of graphics performance. For applications where graphics performance matters, the Tegra K1 should be hard to beat, and should do very well in attracting customers looking to differentiate via world-class mobile gaming performance. It should be interesting to see what other devices Tegra K1 finds a home in during the prime tablet selling season later this year.
Ashraf Eassa owns shares of ARM Holdings. The Motley Fool recommends Nvidia. The Motley Fool owns shares of Qualcomm. Try any of our Foolish newsletter services free for 30 days. We Fools may not all hold the same opinions, but we all believe that considering a diverse range of insights makes us better investors. The Motley Fool has a disclosure policy.