SAR, Cell Phones, and San Francisco

An interesting article (http://www.theglobeandmail.com/news/technology/san-francisco-passes-cell...) came across my desk this week. It talks about a recent vote in by the Board of Supervisors in San Francisco requiring that cell phone retailers post the specific absorption rate (SAR) for all cellular handsets that they sell. The article caught my attention because many cell phone manufactures use our FDTD libraries to model the electromagnetic field strengths, which are used to calculate SAR as a post processing step.

What is SAR?

SAR is a measure of the amount of power that is absorbed by a human body tissue in Watts/kg, averaged over a 1 gram or 10 gram mass of tissue. The FCC and other national government organizations are responsible for defining the safety limits for normal use.

How do you Measure SAR?

Because it is impractical (and inhumane) to insert a probe into someone’s head while they are using a cell phone, we rely on models to determine how much power the human body is absorbing. Since the absorption rate is highly dependent on the position of the cell phone antenna designers will run hundreds, thousands, or even tens of thousands of simulations to determine the SAR value for a given phone.

The other way to measure the SAR value is to create a physical model of the human head and use probes to measure the value. Ideally, simulations match the model results. Major discrepancies need to be resolved before the phone is sent to manufacturing.

Is Lower Better?

Sure. But if the value is too low, your phone will start dropping calls. Typically, antenna designers will work to minimize SAR while maximizing signal strength from the antenna.

CUDA Training

I recently finished teaching an Acceleware CUDA training course, so the timing seemed appropriate to share some of my experiences from the course and share some of the students’ thoughts as they progressed through the training material. When you are immersed in GPU technology on a daily basis, it is easy to take the fundamental concepts of CUDA for granted. Teaching the course forces me to revisit the foundations of GPU programming and gives me some insight into some of the thoughts of people approaching the GPU for the first time.

Learning the GPU – the First Probable Outcomes

When people first start working with the GPU, they invariably experience one of the following outcomes:

1. I have no idea where to begin.
2. The performance on the GPU is slower.
3. The performance on the GPU is marginally better than the CPU.
4. The performance on the GPU crushes the CPU!

I Have No Idea Where to Begin!

This is not uncommon. The GPU does not work like a normal CPU and it requires a completely different mindset to program. By way of example, I’m going to blatantly steal, I mean borrow from Mike’s analogy of home renovations. Imagine that you are building a house using the following (simplified) process:

1. Pour the foundation
2. Add the frame to the home
3. Build the roof
4. Plumbing and electrical
5. Apply the finishing touches

The Technical Details
Since Acceleware’s early days of GPU computing, we have looked for ways to speed up linear algebra solvers, in particular sparse iterative techniques. Two challenges prevented us from achieving this goal:

• Lack of double precision support on hardware
• Preconditioning algorithms tend to be difficult to parallelize

Double Precision
Performing linear algebra calculations in single precision limited the applicability of accelerated solvers on the GPU. With NVIDIA and AMD’s introduction of double precision computation units and support for the IEEE 754 standard, GPUs now match CPU based calculations. Here is the convergence curve of a linear system using a preconditioned GMRES solver:

I had the opportunity to visit DAC 2009 this year in San Francisco. DAC is an EDA conference which showcases tools and processes that perform simulation, logic synthesis, design verification, and timing analysis.  The work we do in linear algebra has direct applicability in this space, and we are eager to expand our reach in the EDA market. The show itself was well attended and there were approximately 200 exhibitors. It was great to catch up with our customers and reach out to some new ones.

Opening Thoughts
Linear algebra and matrices have always intrigued me. The first time I learned that A · B ≠ B · A, my mathematical foundation was turned upside down and I was hooked. Today, I still get to work with A’s and B’s, only instead of 2 x 2 matrices, they are 10,000,000 x 10,000,000 in size. Instead of being exercises in a textbook, the solutions to Ax=b predict how structures will react when stress is applied to them or how a circuit will behave when a signal is introduced to the system. As the complexity of the math increases, so does the practicality of the model, and unfortunately so does the time it takes for the computer to get an answer.

Why do People Care about Speeding Up Linear Algebra?
If you have gotten this far in the blog, you might already have an answer to this question. At Acceleware, we are seeing that simulations are getting more and more complex, and correspondingly the sizes of matrices are getting larger and larger. Engineers, mathematicians, and product designers are frequently limited by the processing power available. We frequently hear from our customers that their nirvana would be to simulate larger data sets, but they are forced to simplify their models so that they finish in a reasonable time. I am always eager to hear these stories as it further reinforces the need for more compute power beyond the conventional CPU and cluster.
A good example was a story told to me by a customer who was using radio waves to search for oil. The problem ends up being formulated as a system of linear equations. I asked how fast their target runtime would be to compute this problem. The response was, “Just being able to finish the simulation would be great. There’s too much processing required.”

IMS/MTT 2009 in Boston was an extremely successful show for Acceleware. This year we had the largest number of Acceleware users stop by our booth in our history of attending the show.  We were curious to hear their experiences with our products and what applications they were working on.  One of our customers positively described our software interface as ‘extremely well designed, very well documented, and easy to use.’  From the application side, end users were applying our product in a wide range of fields including the design of medical devices and predicting the performance of antennas for portable electronics.

Acceleware had a good presence at the show.  We saw a steady amount of traffic all three days, and the show was well attended, despite concerns of the economic climate.

For our fourth consecutive year, Acceleware will be exhibiting at the International Microwave Symposium (IMS) held in Boston from June 9th through June 11th.  I’m excited to see many familiar faces and to meet many new ones. We will be showcasing our products including our accelerated FDTD software, clustered solutions, and our linear algebra solvers.
IMS is one of the world’s leading microwave conferences.  We are always impressed by the diversity of industrial and academic applications that exhibit and present during Microwave Week.  Areas of focus vary from military to medical and commercial to academic.  We are excited that we can offer our products to all of these fields to speed up their designs and simulations!
Come by and visit us at Booth #2709! We look forward to showing you our products and talking about ways we can speed up your computions!