[Previous] Funniest acknowledgment mail...

 [Next] Lawrence of Cybernetica...

Some slightly related from Technorati and Google.

[Ix.stanford.edu] iX Design Weblog: August 2003 Archives: We use a mouseover function to multibrowse the country name to a projector, and a click function to multibrowse the newspaper link to a vertical screen. However, eBeam hardware difficulties prevent rollover without clicking, causing the rollover to be effectively an "onClick" function, just like the newspaper multibrowse function. We have kept the rollover code so that in the future, should the hardware difficulties be overcome, rollover will automatically work. As it is, each region calls the appropriate javascript function when selected by the eBeam puck, projecting the country name to iLoft1 and multibrowsing the country's news site to iLoft2.

[Ix.stanford.edu] iX Design Weblog: July 2003 Archives: The SenseTable (mentioned in my previous entry about object tracking) used two Wacom Intuos tablets to track objects on their table. The paper said it worked well because it sensed the position of objects with 1000 dpi resolution (I'm guessing this is pretty accurate then) and with a very low latency (i.e. delay between movement and the tablet sensing it) compared to computer vision.

 [Osxaudio.com] OSXAudio -- View topic - MOTU Traveller anybody?: the traveler seems to be pretty solidly built despite its small weight - the only thing that's a bit scary is that when you press the potentiometers closest to the LCD screen (for example, pushing the top right knob sets the channel cuemix vol to either maximum or mute), the LCD bends a tiny bit backwards from that side... which suggests that it's mounted to the same bit of PCB as the knobs. might be no problem at all even on the long run but it looks weird after you've noticed it.

 [Jroller.com] murphee's Rant: The cost of branches and pipeline bubbles (which gets higher with longer pipelines) has been reduced, or at least kept at bay with more sophisticated branch prediction, but that‘ll also have its theoretical limits. TLP (thread level parallelism, as in SMT or Hyperthreading) is another optimization, which mostly helps to get better use out of the execution units and reduces the cost of I/O latency (if a load instruction takes a long time, it stalls the pipline; without SMT, the following instructions would simply be backed up and nothing would get done; with SMT, pipeline just executes more insructions from the other thread and doesn‘t have to twiddle its thumbs). But…TLP/SMT need something: multiple threads. Even if you crank up SMT to allow more than 2 threads, or if you cram more and more CPUs on one die, you‘ll need to fill all those parallel CPU pipelines with instructions.

Reflected tags on Technorati: Blog, Cybernetics, Kybernetica.com