Fall IDF 2005 - Day 1: Coverage of Everything
by Anand Lal Shimpi on August 24, 2005 2:31 AM EST- Posted in
- Trade Shows
Intel's Gets into Benchmarking
Intel's Digital Home Capability Assessment Tool (isn't that a mouthful) is a new Intel benchmark that will be released this Fall. What's particularly interesting about this benchmark is that it runs actual digital home tasks (e.g. playing HD movies, transcoding video, recording video and exporting video) using real world applications.
The benchmarking tool runs these tasks one at a time as well as in parallel, and generates a performance rating based on things like the number of frames dropped while watching a HD video stream.
The idea is to be able to quantify the digital home experience a particular platform would offer, thus allowing end users to better compare systems based on overall experience. Honestly, the whole thing sounds a lot like AMD's True Performance Initiative that resulted in the model numbers we've come to live with for so long; it also sounds a lot like Intel's iComp ratings from years past.
The difference between this tool and previous attempts to characterize performance is that Intel's tool doesn't just output a performance rating, it also gives you a detailed list of what tasks the tested PC can do poorly, relatively well and perfectly:
The overall performance rating
The breakdown of what this system can and cannot do
The rating system is an interesting idea and Intel promises that it will be made extensible to include other usage patterns as they emerge.
Obviously Intel's tool isn't something we'd use here on AnandTech, but it is interesting to note that you may end up seeing the results of this tool in stores like Best Buy and by companies like Dell to help characterize the relative performance of their systems as far as a digital home user is concerned.
The tasks the tool performs are using regular, unmodified applications, so in theory they shouldn't deliberately penalize non-Intel processors, but we'd honestly be very surprised if Intel had put together a benchmark that they didn't win.
Intel's other benchmarking program was a similar experience-oriented performance tool for games, very similar to FRAPS with a few tweaks. The tool measures instantaneous frame rate over time, but then it compares those frame rates to user data expressing the quality of their gaming experience with reference to the frame rates and produces an experience rating.
The idea of the experience rating sounds interesting, but we can't help but believe that this tool wouldn't have been created had Intel not been so far behind on gaming performance.
Intel is planning on making all of these types of tools available to the public and making them somewhat transparent, hopefully showing consumers what these benchmarks actually do to generate their results. Again, these aren't things we'd use here, but they are things that may end up in a retail store near you.
14 Comments
View All Comments
Questar - Wednesday, August 24, 2005 - link
The 5-10 year part is speculation of Anand. Intel never said it would take that long. I'll bet two years. It doesn't take 5 years to write a compiler or add a chip feature.Anand Lal Shimpi - Wednesday, August 24, 2005 - link
The Intel rep that did the demo was the one that provided the 5 - 10 year estimate. This research is in its very early stages, but the promising first results means it will probably get more support.Take care,
Anand
drpepper128 - Wednesday, August 24, 2005 - link
Is it just me or are we missing something here?To me it seems that the real power of Mitosis is that companies would not have to worry about writing code that is mult-threaded. Instead they can have single-threaded code and use the compiler to multi-thread it. This is where the real power of multi-core processors could come from. Some day when we have 100 core processors we will need something like a compiler to figure things out for us; otherwise a company's costs would skyrocket. Think somewhere along the lines of graphics cards.
JarredWalton - Wednesday, August 24, 2005 - link
I was thinking if they could get Mitosis into the chips (rather than required compiler support) then it would benefit practically *any* application. The only time it wouldn't help performance would be when your CPU was either fully loaded on every core, or perhaps if the multiple threads start using up resources that could be better used on stuff other than speculative execution.