On CBS MoneyWatch: Dumbest Things You Do With Money
BNET Business Network:
BNET
TechRepublic
ZDNet
TalkBack 22 of 35:
Next »
« Previous
actually...
First of all thanks everyone for all the polite and nice feedback... I was expecting all kinds of flame saying how I forgot some type of something somewhere. happy

It is interesting though that you talk about that, because I know of someone who is working on something like that for their doctoral research in CompSci. Dr. Daniel Jimenez at Rutgers University is big into branch prediction and ways to accurately measure it... and he's working on a project now with some students to characterize different branch prediction methods based on the amount of power they cause the system to consume when working on different benchmarks.

As you may or may not know branch prediction allows CPUs to guess very accurately if a jump (basically a if-then statement in machine instructions) will be taken or not. If you can predict it accurately, you can save a lot of time. Today's branch predictors are actually amazingly accurate. Most can guess correctly more than 92% of the time!

It is interesting though that usually just percentages are used, but no one usually measures the effect on the system to see if a branch predictor is making it run any better. It could be that a branch predictor has a lower accuracy, but the things it misses don't cause much work, but another one with better accuracy misses on really bad instructions that cause lots of extra work... so even though it appears to be more accurate, it doesn't help the system to run better... hence the work to measure branch predictors based on the amount of power they cause the system to use.
Posted by: deymious@...   Posted on: 09/18/05 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

The speed race is over. The power race is on.  DonnieBoy | 08/23/05
don't we all wish that...  ctk76 | 08/23/05
At last Intel started catching up  michael_t | 08/23/05
Overheating  dch48 | 08/23/05
Old Story  in-DUH-vidual | 08/23/05
Its true i reheated chineese food with it  BuckRogers_z | 08/24/05
RE: That's nice; now for speed?  Krazyken39 | 08/23/05
now for speed  linuxoverwindows | 08/23/05
Power consumption is of little interest to me  stan@... | 08/23/05
Re Power Consumption  joedokes | 08/23/05
Oh Really?  panzrwagn | 08/23/05
Really  stan@... | 08/24/05
The market's going away from your preference...  kjartan93 | 08/24/05
It usually does  stan@... | 08/24/05
Low power? So what  Raed009 | 08/23/05
what you think is speed may not be  deymious@... | 08/23/05
nice work...  linuxoverwindows | 08/23/05
Here Here  bka1959 | 08/23/05
Re: nice work  Ande_z | 08/24/05
It gets worse  CobraA1 | 08/23/05
Power vs Actual Work Done (i.e.SpecInt)  usr001 | 08/24/05
actually...  deymious@... | 09/18/05
Thank you  G Fedorchuk | 08/25/05
Intel: more power less enrgy  chanhqvo | 08/23/05
Roadmap and Apple  Mr_Dave | 08/23/05
I'll be Watching...  Jkirk3279 | 08/23/05
On sharing IBM technology...  flatliner | 08/23/05
Torvolds strikes again  Roger Ramjet | 08/24/05
New Architecture  Old Guy_z | 08/24/05
I agree speed race over  kokuryu | 08/24/05
RE: "I agree speed race over"  ajapierce | 08/24/05
Parasite-ware... good one  Mr_Dave | 08/24/05
Great for the Road... not so good for the ....  BuckRogers_z | 08/24/05
IDF shows INTEL is far far behind  sharikou | 08/26/05
Who cares?  Thore | 02/07/06

What do you think?

SponsoredWhite Papers, Webcasts, and Downloads

advertisement

SmartPlanet

Click Here