NVIDIA (NASDAQ:NVDA) has released its new SoC (system on a chip) called the Tegra K1, that is highly touted for its speed. Fellow S.A. author Paulo Santos has written an excellent article on the topic. Here I will question if perhaps we should take these preliminary numbers with a grain of salt.
In my article on the upcoming A8 chip from Apple (NASDAQ:AAPL), I speculated on its features and performance, extrapolating from that of the A7, which is used in the iPhone 5s, the iPad Mini with Retina display, and the iPad Air. But, since the A8 has not been announced, let alone shipped, any comparisons here would be conjecture.
The Tegra K1
I agree with Mr. Santos that this powerful entry is a potential plus for NVIDIA if it is picked up to run successful tablets. They could run either Android, by Google (NASDAQ:GOOG) or Windows Phone 8 by Microsoft (NASDAQ:MSFT). As such, the K1 will compete against Snapdragon entries from Qualcomm (NASDAQ:QCOM).
Among other things, he notes that NVIDIA is now open to selling its GPU design to other chip developers, and that this could be a negative for Imagination Technologies (OTCPK:OTCPK:IGNMF) which currently supplies its PowerVR designs for Apple's processors.
NVIDIA of course, is famous for its graphic processing units (GPU) and associated graphics boards for personal computers. Several years ago they entered the SoC market with their Tegra line. These systems contain CPUs based on designs by Arm Holding (NASDAQ:ARMH) but with their own graphics cores. The new Tegra K1 begins the use of their Kepler GPU technology, essentially the same as that in their PC orientated GPUs.
As Santos notes:
At this point, the benchmarks seem to paint a very favorable picture for Nvidia, with the Tegra K1 showing well ahead of competition. This is especially noticeable in graphics-heavy work, as can be seen in the following benchmark
[Source: WCCF tech]
The WCCF Tech review notes:
However for the target niche the Tegra K1 was actually created; it leads with a major gap. Scoring a rock solid 60fps in an off screen 1080p Benchmark it fares significantly better than the Tegra 4.
The closest tablet tested is the iPad Air which manages a barely acceptable 27 fps (still not gaming standard) on the demanding 1080p res.
I have no doubt that the Tegra K1 is an impressive chip with very high GPU marks. It sports 192 CUDA graphic cores and is definitely cutting edge technology. Apple's A7 sports only 4 graphic cores although they are of a completely different technology, based on Imagination Technology's PowerVR series 6 (or Rogue) designs.
There are two things to be careful of here.
First is that we should not compare two SoCs based on only one benchmark. When the highly respected site AnandTec evaluated the iPad Air, they used no less than 20 benchmarks for the GPU alone, and others still for the CPU performance. That said, the GFXBench that tests HDTV (1080p) video production is a reasonable one to use, as long as we maintain perspective. We can accept it as a tentative indication of performance against other systems. In this one task at least, the K1 is certainly impressive.
More importantly, however, we have to be just a little bit skeptical of the use of a reference tablet in the testing. A reference tablet is designed to show proof of concept but does not necessarily conform to a lot of real world restrictions.
Specifically, I refer to electrical consumption and heat.
As a processor does work, it consumes electricity and produces heat. The first can run down a battery. The second, if not restricted, can burn out a processor. The way to eliminate the heat problem is first to provide heat dissipation mechanics (metallic heat sinks, passive ventilation and fans), and then to literally slow the processor down. My concern here is that NVIDIA wants to get the most performance out of their demos, and so we must wonder what they have done to put their best foot forward.
Are there larger batteries to handle power consumption? Are there larger heat sinks? extra space or even fans?
In a tablet the weight of batteries is a significant design factor, as is overall size. In a reference tablet these may not be issues, but for real production models they most certainly are.
So, while the tests here are a great indication of possible peak performance, it is too early to judge how it will perform in real life, on real production products.
AnandTech does have a review of the Tegra K1, but only of the official release party press conference information - that is, no evaluation unit to test. They have some very interesting remarks - as always.
The K1 comes in two configurations, 32-bit and 64-bit models. Of the latter model, called Denver core, they write:
The Denver option is the more interesting of the two as it not only gives us another (very unique) solution to the power problem in mobile, but it also embraces a much more sane idea of the right balance of core size vs. core count in mobile.
What they refer to here is the fact that this model uses only two CPU cores - like the Apple A7 design.
Still, they say that "the big story behind Tegra K1 is its GPU," which is not surprising. In a smartphone or tablet, so much of the experience is graphic - from the scrolling list and other interface elements, to the web page displays, photo production, and video display. Therefore, graphics processing is more important, in many ways, than is that of the traditional CPU.
Not only is the system fast, but because they use the same core technology, they have the same application programming interface (API) for developers. Prior to the new Kepler design, the Tegra chips were non-standard. AnandTech notes:
The most compelling argument in favor of putting Kepler in a mobile SoC actually has to do with its API support. In one swift move NVIDIA goes from being disappointing in API support to industry leading. Since this is a full Kepler implementation (just a lower power/performing version) Tegra K1 maintains full API compatibility with NVIDIA's flagship GeForce products. OpenGL ES 3.0 is supported but so are full OpenGL 4.4, DX11 and CUDA 6.0
And later on:
NVIDIA's challenge with Tegra has always been getting design wins. In the past NVIDIA offered quirky alternatives to Qualcomm, most of the time at a more attractive price point. With Tegra K1, NVIDIA offers a substantial feature and performance advantage thanks to its mobile Kepler GPU.
If I were in Microsoft's shoes, I'd view Tegra K1 as an opportunity to revolutionize my mobile strategy.
The A7 is what it is and cannot be changed, but looking to the future Apple does have options. In my speculative A8 article, I noted of the PowerVR Series 6 technology, which boasts (among other things):
- Highly Scalable, Programmable Processing Clusters
- The USCs [Unified Shading Clusters] are multi-threaded processing units that are optimised for the operations used in vertex and pixel shaders and also practical GPU compute tasks (video and image processing, for example). The USC architecture is highly scalable, meaning that multiple clusters can be implemented with very little overhead (performance scales almost linearly with additional clusters). [emphasis added]
This means that they will be able to scale up their processor as well. As they move to a smaller IC process technology that allows them more complexity for a given area, then they will be able to add GPU clusters. Even if Tegra is double the GPU speed of the A7, that can be matched or bested in Apple's next generation.
The new Tegra K1 is definitely a big jump in processing power from their previous models, and is likely to generate new design wins for NVIDIA. This should be good for their bottom line. Still, as for how great the improvement really is, we will have to wait until we have real production units to test.