By: Vincent Diepeveen (diep.delete@this.xs4all.nl), August 31, 2007 11:44 am
Room: Moderated Discussions
Richard Cownie (tich@pobox.com) on 8/28/07 wrote:
---------------------------
>Thanks for the great article.
>
>First impressions:
>
>1) It appears to have a lot of useful functionality
>(especially the skew-compensation stuff which only
>arrives in the AMD/HT world with HT3.0).
>
>2) It seems to be rather complicated relative to HT,
>and I would be nervous that this might lead to
>higher latency. Probably hard to figure out the
>real system latency until we see actual hardware.
>
>3) They'll have to be careful about the link-recalibration.
>It might sound reasonable to spend 10uS every 10mS
>for recalibration of each link, but this could become
>a serious problem if either you're doing something
>real-time with hard latency limits, or you have a big
>system with many links and the recalibration isn't
>synchronized.
>
>Anyway, good that CSI is getting closer to products.
What latency can we typical expect according to your best guess, when the system is not idle?
Thanks,
Vincent
---------------------------
>Thanks for the great article.
>
>First impressions:
>
>1) It appears to have a lot of useful functionality
>(especially the skew-compensation stuff which only
>arrives in the AMD/HT world with HT3.0).
>
>2) It seems to be rather complicated relative to HT,
>and I would be nervous that this might lead to
>higher latency. Probably hard to figure out the
>real system latency until we see actual hardware.
>
>3) They'll have to be careful about the link-recalibration.
>It might sound reasonable to spend 10uS every 10mS
>for recalibration of each link, but this could become
>a serious problem if either you're doing something
>real-time with hard latency limits, or you have a big
>system with many links and the recalibration isn't
>synchronized.
>
>Anyway, good that CSI is getting closer to products.
What latency can we typical expect according to your best guess, when the system is not idle?
Thanks,
Vincent