bohnsack.com

We need to write a term paper for my class in advanced computer architecture. Here's a short description of the requirements:

Due in December, the paper should be 15-20 pages long (double spaced) and include an in depth analysis of some topic related to the things we discuss during the semester. That is, select a concept, machine, technique, or mechanism; then, state whether it is good or not, defending your position with judicious use of appropriate metrics.

Further guidance:

• A topic we're interested in
• Understand what people have done (search/research)
• After search, you have an idea what's good and what's not good
• Propose something - make a coherent analysis of what we've decided to work
  on with the chosen metric
• Conclusion that defends use and choice of metric and makes a conclusion,
  based on the metric

Here's my proposal:

The proposed paper will examine the Single Instruction Multiple Data (SIMD) extensions found in modern commodity CPUs. An emphasis will be placed on the use of these features in floating point-intensive High Performance Computing (HPC) applications. Of specific interest is the application of SIMD features in the High Performance Linpack (HPL) benchmark and the BLAS DGEMM matrix-matrix multiply routine, which is the main math kernel used by HPL.

SIMD techniques compared will include the Pentium III's SSE extensions, Pentium 4's SSE2 and SSE3 extensions, and the PPC 970's AltiVec extensions.

The theoretical peak floating point operations per second (Flop/s) achievable with HPL will be derived for each CPU architecture and SIMD extension examined, by considering the DGEMM algorithm, and the details of each SIMD feature, along with overall CPU architecture, including pipelining and functional unit organization.

An experimental validation of theoretical values will be conducted by running the HPL benchmark on all the investigated platforms. Runs will be conducted with and without DGEMM routines that are optimized for each CPU's specific SIMD features. That is, the HPL performance of each CPU will be examined with and without the use of its SIMD extension.

Finally, a change to one of the CPUs analyzed that could double its performance in the HPL benchmark will be proposed. The architectural changes required will be explained and the theoretical speedup will be calculated.

I love grad school. My weekends mostly look like this:

Ajaxian.com - the world of AJAX.

What Internet censorship looks like in the United Arab Emirates (via jwz).

ri usage tip

I upgraded to Firefox 1.5 Beta 1 this afternoon. Lots of sweet new featues, including SVG support, but my form element spelling checker plugin doesn't work and that kind of sucks. The average user will want to wait.

One of the most interesting facets of podcasting, is podcasting truckers. This is either an invalidation of the trucker stereotype, a strong indication of how pervasive podcasting is becoming, or a little bit of both.

Fascinating.

In my computer architecture class, we're discussing the Tomasulo Approach, a mechanism that allows processors to dynamically execute instructions out of order. It's not too hard to understand from a big-picture kind of perspective, but I was having a lot of difficulty grokking the way it really works. Neither the book nor the lecture were clarifying....

Then a passing comment from a fellow student caused me to google for "Tomasulo applet", and I became enlightened (via this visualization in particular, but this one's also good).

This type of teaching aid is very beneficial, but isn't referenced in the book, and it wasn't used in lecture. Why not?

• Why can't students collaboratively annotate "the book", so none will have to struggle with this again - like ever again?
• Why aren't the best tools available brought to bear during lecture (especially when they're freely available)?

I was playing around with mapping software this evening and found a good image of the Rau farm where I spent a lot of my childhood (~15 miles NW of Watertown, SD):

~ -97.12888 longitude and 45.03017 latitude: Terraserver, Google Maps

You can find lots of old pictures of the farm here.

I also found good pictures of Rauville and Highland Park.

An inspirational podcast from Adrian Sannier:

An expert in human/computer interaction and three-dimensional visualization, Dr. Adrian Sannier joined the ASU community in July after four years as the Stanley Professor of Interdisciplinary Engineering at Iowa State University. Dr. Sannier’s first priority is developing a strategic plan for the application and use of information systems and technology at ASU. Dr. Sannier addressed the development process of this strategic plan and shared his thoughts on the role of information technology in the New American University at a forum for ASU IT staff.

Sannier Presentation MP3 - September 15, 2005

Power point presentation

A tasty portion that I especially enjoyed (roughly):

No medals are handed out for "Look out! Iceberg!". Whatever. That helps nobody.

'Cause we're still going to America, and if you don't tell us how to get around the iceberg, then we're just going to hit it.

OK?

Medals are given out for finding and fixing problems.

We'll what if I can only find it... Then get together with some people who can fix it, because the goal is to get there, instead of finding reasons why is was a good idea not to go.

Once the arrow has flown, we hit the target.

If you focus that way, you hit more targets than you miss. The other way leads to something called a can't do attitude.

Speaking of podcasting and IT, IT Conversations is a wonderful resource. So wonderful that I'm seriously considering gettting an iPod (probably a nano), so I can do the whole automatic download of new shows and have it to go thing. The Ruby on Rails podcast will also be automatically downloaded.

Update: Sannier's blog post about the podcast.

Four years ago. Don't forget.

Streaming Video.

RAY KURZWEIL: The InstaPundit Interview. Ray needs to be slapped with an SI clue stick (who the heck uses "quadrillion" instead of the much more familiar "peta" or 10¹⁵?), but the interview's an interesting glimpse at Ray's next book, which will be a must read for singularitarians.

I can see using the common English (tens, thousands, etc.) up to a million, but at a million the different words used by different languages start to get confusing, and almost no one knows what quadrillion, quintillion, etc. mean anyways - I don't. My suggestion: Use SI prefixes past a million.

One thing standing in the way of my suggestion to use global standards to count large things is that we're sucking wind in the turning out engineers department. From the interview...

If you haven't already, please consider giving towards the Hurricane Katrina relief effort. It's such an easy thing for us, yet it's the most important thing in the world for those effected by this disaster.

You can choose to help via one of many different charitable organizations.

I gave via the Salvation Army: $200 via their website + a trunk load of clothes to my local branch, but I'm encouraging you to give whatever you feel comfortable with via your favorite organization. It's the right thing to do.

Google maps of flooded NO.

How much of a geek do you have to be to contribute to this Wikipedia entry? I'm going to watch a little Q Who tonight.