By: Matt Waldhauer (M.Waldhauer.delete@this.gmx.de), November 18, 2010 7:38 am
Room: Moderated Discussions
someone (someone@somewhere.com) on 11/17/10 wrote:
---------------------------
>Matt Waldhauer (M.Waldhauer@gmx.de) on 11/17/10 wrote:
>---------------------------
>>Thanks for your nice speculative article.
>>
>>What do you think about the idea that Poulson might employ reliable computing by
>>issueing the same instructions twice to different execution units?
>>
>>- Matt
>
>Poulson seems to offer up the possibility of doing a kind
>of dual core lock step execution within a single core.
>
>If 1) you restrict per thread instruction issue width to two
>bundles, and 2) there is sufficient symmetry in execution
>resources, one couldd execute a single program thread
>duplicated as two hardware threads running in lock step
>on one core. It wouldn't take much extra logic to cross-
>check execution results in real time.
IIRC Intel filed a patent for doing such checks for code execution on two similarly configured groups of CPU resources.
---------------------------
>Matt Waldhauer (M.Waldhauer@gmx.de) on 11/17/10 wrote:
>---------------------------
>>Thanks for your nice speculative article.
>>
>>What do you think about the idea that Poulson might employ reliable computing by
>>issueing the same instructions twice to different execution units?
>>
>>- Matt
>
>Poulson seems to offer up the possibility of doing a kind
>of dual core lock step execution within a single core.
>
>If 1) you restrict per thread instruction issue width to two
>bundles, and 2) there is sufficient symmetry in execution
>resources, one couldd execute a single program thread
>duplicated as two hardware threads running in lock step
>on one core. It wouldn't take much extra logic to cross-
>check execution results in real time.
IIRC Intel filed a patent for doing such checks for code execution on two similarly configured groups of CPU resources.