Moar Cores! 48 of them.

[I-Hate-Hulse]

Intel has unveiled a prototype chip that packs 48 separate processing cores on to a chunk of silicon the size of a postage stamp.
The Single-chip Cloud Computer (SCC), as it is known, contains 1.3 billion transistors, the tiny on-off switches that underpin chip technology.
Each processing core could, in theory, run a separate operating system.



http://news.bbc.co.uk/2/hi/technology/8392392.stm




Personally, I'm still waiting for some dual core optimization for some apps I run...

[FanIn80]

Originally Posted by I-Hate-Hulse

Personally, I'm still waiting for some dual core optimization for some apps I run...

No kidding. Here's a novel idea, computer world... how about we lay off the hardware for a while and start focusing on the software?

[sclitheroe]

Originally Posted by FanIn80

No kidding. Here's a novel idea, computer world... how about we lay off the hardware for a while and start focusing on the software?

Just because YOU can't use 48 cores doesn't mean there aren't tons of applications for chips with increased thread density.

Forget about optimizing performance for a second and consider all the possibilities inherent in being able to support massively multiple OS instances in such a broad way. Imagine every app on your system running on its own OS image on its own dedicated cores. Stability, fault tolerance, security, scalability, all gain massive new possibilities.

The real advances with this design, regardless, isn't about overall core count, it's about the improved core manageability, similar overall power consumption compared to the i7 & i9, and the fact that if Intel is saying its coming in 2010, that they are able to reliably fab these at something approaching affordable (depending on your definition of affordable, of course)

Finally, its really important for hardware manufactures to get out ahead of the software when it comes to multithread and multicore performance, because once you start architecting and writing software that is multicore optimized, and once compiler and library support eases the barrier to entry, going from 2 thread to tens or hundreds of threads is strictly a function of hardware availability - its as easy to leverage 2 cores as it is 48, but the hardware has to be able to keep those cores busy which means lots of scheduler optimization work, improvements in memory and bus architecture, etc.

[I-Hate-Hulse]

I hope AMD or Intel calls the inevitable 8 core chip, the "Ocho".

[ricosuave]

http://www.mlin.net/SMPSeesaw.shtml

[sclitheroe]

Originally Posted by ricosuave

http://www.mlin.net/SMPSeesaw.shtml

Open task manager...next, click on the Processes tab. Next, click on View, Select Columns, and choose Threads. Now sort the display by thread count.

You will notice that every single application, with the exception of maybe a very few services, has more than one thread. With the possible exception of legacy games that have timing issues running on more than one core, or maybe a single threaded computationally demanding scientific app, there is practically never a situation where you are going to get better performance by locking a process to a single core.

Almost every single Windows app will spin up at least one additional thread just to handle the messages passed to the app by Windows, and there can be hundreds of these sent to the app every second. These messages are everything from requests to redraw the contents of windows, to mouse movements, keystrokes, incoming data, etc. You want these threads to be able to run on any available core to maximize the responsiveness of your apps.

Even in highly virtualized workloads, where I might have 15 or 20 virtual machines running on an 8 core server, VMWare's advice is always the same - don't mess with affinity, let the software work it out (except for some very, very specific edge cases)