NVIDIA, ARM, Tegra 3, and Beyond

The Competition in ARM-Land

NVIDIA does not live in a vacuum when it comes to ARM processors.  While NVIDIA has a small leg up on the competition in actually shipping dual core Cortex A9 SOCs, there is plenty of competition around which has taken notice of the new kid.  Qualcomm and Texas Instruments are two of the larger entities out there providing ARM based processors for phones and tablets.  Both of these companies have shipped a whole lot more product than NVIDIA so far, and they have very deep relationships with manufacturers around the world.  TI did mention in their most recent earnings report that they were behind NVIDIA in shipping dual core A9 based processors, but that they are speeding up development to not just deliver this series of products much more quickly, but will advance the release schedule of future products.  The one area that NVIDIA has an advantage is certainly that of integrated graphics.


Imagination Technologies licenses out its PowerVR mobile architectures, and TI is one of the biggest customers.  The OMAP series of processors feature different SGX configurations from IT, and these are still very robust parts.  IT has not updated the cores nearly as much over the past few years as compared to the desktop and notebook battles that go on between AMD and NVIDIA.  In fact, the latest SGX543 was announced in early 2009, but the first time this graphics core technology was used was actually in late 2010.  Apple will also be utilizing this technology in their next generation ARM based SOCs which will be powering the upcoming iPad 2.  Sony also is entering the fray with their Next Gen PSP, which will feature a quad core Cortex A9 based processor with the quad version of the SGX543.  It is interesting to think that Sony is not adopting NVIDIA based graphics, since NVIDIA powers the graphics portion of the PS3.

Qualcomm has the very popular “Snapdragon” series of SOCs, and in an interesting twist they were the company which bought up AMD’s Imageon division.  Their Adreno graphics appear quite a bit slower than both the NVIDIA and IT based units, but considering the new blood that just showed up in the highly competitive cellphone processor world I would expect to see Adreno gain some muscle.

The upside of ARM for NVIDIA is the tremendous growth prospects it faces with its competitive products.  NVIDIA is not stopping just with cellphones and tablets.  Project Denver is aimed at creating a high performance ARM based CPU which will fit in standard notebook, desktop, and server TDPs.  This means products going from 18 watts TDP up to 125 watts TDP.  There is no reason why ARM CPUs cannot scale into this area, and their performance should be competitive with what AMD and Intel have in this market dominated by these two companies.  With the announcement that Microsoft will offer an ARM based version of Windows 8, the upside for NVIDIA is certainly there.  With a robust chipset design team as well, it would not be surprising if NVIDIA was able to offer an all-in-one solution for these markets.


One other area in which NVIDIA hopes to capitalize is that of consoles.  While NVIDIA is in the PS3, there are still plenty of consoles around which they are not a part of.  While the current generation is dominated by the Power architecture (Cell is Power based with 7 SPEs for the PS3, the Xbox 360 utilizes a 3 core PowerPC design, and the Wii features a single core PowerPC unit), this may not be the case with the next generation of consoles.

Sony is taking the first major shot with the Next Gen PSP.  This will be ARM based, and the specifications for this handheld device look impressive.  Sony and the gang have to be thinking that it would be in their best interest to provide a unified code base for their gaming consoles and portables.  ARM is a way to do that.  With NVIDIA working directly with ARM to create high performance multi-core processors, it is not much of a jump to consider that Sony will move away from the older PowerPC architecture and go with ARM.  While the Cell concept was interesting, it was never very effective.  Apparently the memory and cache bandwidth did not allow the effective use of more than 2 SPEs at any one time.  The CPU was simply being underutilized, and no amount of software advancements could change that.

The Nintendo DS is also based on ARM processor technology.  These utilize the ARMv7 architecture, which is still 32 bit.  While the IBM PowerPC is a good architecture, the idea of switching to an ARM ISA certainly is appealing, as this would easily fit the business plans of these companies.  The world has plenty of ARM based developers, and the competition to provide the chip for these next generation consoles would be stiff.  Add into that equation the amount of software development going into ARM powered mobile devices, and we see some very compelling reasons to move in that direction.

ARM is working on developing a 64 bit processor, and we can only assume that NVIDIA is also tapping into this resource (as well as providing its own resources to get these products up and running ASAP).  While it is a good couple of years until we see the next generation of gaming consoles hit the market, I do expect all of them to go to the ARM architecture for their CPU needs.  It just makes sense, especially with the recent explosion of mobile gaming which is overwhelmingly run on ARM based devices.

AMD and Intel are not standing still in the face of this mobile revolution.  While AMD is a bit slow coming to the party, and in fact looking a bit silly for selling off the Imageon division, they have a new architecture in Bobcat that can move into the sub-1 watt marketplace.  At 28 nm, there can conceivably be a basic dual core Bobcat based APU optimized for lower power envelopes.  This would probably run a dual core processor at 1 GHz and a 40 to 80 stream unit graphics part at 200 MHz, and if done correctly could compete with the ARM products at that time.  While the ARM processors will likely run at a higher clock speed, the design of a Bobcat based low power derivative should be able to be competitive even at a lower clockspeed.


Intel has had Atom out for three years now, and this introductory part is not exactly a barn burner.  Nor is it all that exciting if shrunk to sub 28 nm geometries.  Intel most likely is working on a new low power architecture which takes some of the power saving and performance features from Sandy Bridge, and creates something much more closely related to Bobcat than the current iteration of Atom.  Intel will have 22 nm production up and running by the end of this year, with initial products hitting the shelves in early 2012.  If Intel does in fact have a design much as I described above, with a further refinement of the Intel integrated graphics, then we could have a product in the first half of 2012 that will compare well to the low power ARM based products using the 28 nm process node.

NVIDIA has made a dramatic statement with the release of the Tegra 2.  This was sorely needed after the past few years that NVIDIA has had.  Obviously their plan for ARM has been in the works for quite some time, and the perceived lull in productivity between 2008 and the first half of 2010 is looking now like a complete retooling and rearrangement of NVIDIA’s divisions and design teams.  The recent successes of the GeForce desktop graphics cards, combined with the very successful Tegra 2 launch, has given NVIDIA a lot of impetus.  Morale is much improved inside of the company, and externally it gives people the impression of strong forward movement.  Now NVIDIA needs to keep up the momentum and continue to release new iterations of these highly sought after chips.  People who thought that NVIDIA would fall to irrelevance (including myself) look to be wrong.  If NVIDIA can stay at the forefront of this market, and expand successfully into the traditional CPU market, then the company has many years of success and prosperity ahead.