By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), January 2, 2014 7:57 am
Room: Moderated Discussions
Maynard Handley (name99.delete@this.redheron9.com) on December 31, 2013 7:49 pm wrote:
> We know that the chip was abandoned shortly after Oracle bought Sun, and we know that Ellison claimed
> that it was abandoned because ran hot and slow. But we don't, as far as I know, have independent confirmation
> of this claim (after all, Ellison had his own agenda and vision for Sun once it became part of Oracle)
> or any insight into the extent to which what Ellison was seeing as slow and hot was in fact not the
> final chip but, e.g., some sort of FPGA version being used to verify the design.
Unfortunately besides Ellison's comment little and less has been known about Rock's actual performance (or lack of). The only thing I heard - and I can't remember where I heard it - was that Sun encountered some significant performance issues in multi-socket scenarios. Considering Rock was supposed to be effectively running with HLE on at all times (due to scout threading and TM) having poor scalability due to a design flaw sounds like a possible explanation. Though it might have not have been the only one.
Anyway, being a design which tried to introduce multiple radical changes at the same time it was very high risk and it's somewhat unsurprising that it ultimately crashed and burnt.
> We know that the chip was abandoned shortly after Oracle bought Sun, and we know that Ellison claimed
> that it was abandoned because ran hot and slow. But we don't, as far as I know, have independent confirmation
> of this claim (after all, Ellison had his own agenda and vision for Sun once it became part of Oracle)
> or any insight into the extent to which what Ellison was seeing as slow and hot was in fact not the
> final chip but, e.g., some sort of FPGA version being used to verify the design.
Unfortunately besides Ellison's comment little and less has been known about Rock's actual performance (or lack of). The only thing I heard - and I can't remember where I heard it - was that Sun encountered some significant performance issues in multi-socket scenarios. Considering Rock was supposed to be effectively running with HLE on at all times (due to scout threading and TM) having poor scalability due to a design flaw sounds like a possible explanation. Though it might have not have been the only one.
Anyway, being a design which tried to introduce multiple radical changes at the same time it was very high risk and it's somewhat unsurprising that it ultimately crashed and burnt.
