Posts Tagged ‘GPGPU’

Acronyms With Historical Baggage

July 9, 2011

There are some pretty silly acronyms out there. Some even parody themselves. My personal favorite is PCMCIA for People Can’t Memorize Computer Industry Acronyms.

As people build new things based on old things, the new things sometimes get names that tell a bit about their history. This eventually leads to an AWHB, or Acronym With Historical Baggage. Here, baggage refers to how the name manages to oxymoronically mention its history. An obvious example, at least for those familiar with it, is EEPROM, or Electronically Erasable Programmable Read Only Memory. In this case, programmable means data can be written to the supposedly read-only memory. Before EEPROM, there was EPROM which was erasable, but not electronically. Before that, there was PROM that could be written only once, and before that, ROM, which was manufactured with the data instead of being programmable. Thus, EEPROM is clearly an AWHB.

A newer AWHB is GPGPU, or General Purpose Graphics Processing Unit. I imagine that most people who have thought about a GPU, or Graphics Processing Unit, came to the conclusion that it was generally for the purpose of making graphics, so why make a new, longer acronym that explicitly states is it general? That is where the AWHB nature comes in. A GPGPU adds all the complex branching needed for general purpose computing to a processor with a GPU lineage. That makes a GPGPU a CPU that can do math really fast in parallel compared to most other CPUs. Some CPU designs are moving toward the same goal, like IBM’s POWER architecture and Intel’s Larrabee. Just to confuse things more, Larrabee is being called a GPGPU but is extending Intel’s CPU designs to look more like GPUs, just like POWER.

As far as I can tell, GPGPUs are a subset of CPUs and not GPUs. When GPUs came out, they were for a specific purpose. The design took what would be software on a CPU and implemented an equivalent with transistors which resulted in hardware that did predetermined mathematical computations faster than a CPU because implementing conditional branches while maintaining high performance is very difficult and takes a lot of transistors. These GPGPUs remove the limitation that made GPUs different from CPUs in the first place. Their approach to math was different from CPUs before CPUs gained SIMD, or vector, instructions.

Now, GPGPUs differ from CPUs only in what hardware they are paired with, but that isn’t inherent in the GPGPU itself. The notion of something called a GPGPU just bugs me. AWHBs just have that effect on me.

There are plenty of other AWHBs, but I don’t know them all. What AWHBs have you come across?


False Steps

The Space Race as it might have been

You Control The Action!

High Frontier

the space colony simulation game

Simple Climate

Straightforwardly explaining climate change, so you can read, react and then get on with your life.