By: Mark Roulo (nothanks.delete@this.xxx.com), December 13, 2012 9:11 am
Room: Moderated Discussions
Exophase (exophase.delete@this.gmail.com) on December 12, 2012 11:46 pm wrote:
> Mark Roulo (nothanks.delete@this.xxx.com) on December 12, 2012 3:53 pm wrote:
> > Maybe this is new only to me, but the iGPU clock speed seems to have
> > almost doubled in Haswell compared to Ivy Bridge: 1250 vs. 650.
> >
> > *IF* Intel has also increased the iGPU in other ways (widening it, whatever ...) *AND*
> > can solve the bandwidth starvation issue (Crystalwell maybe??), then it looks like
> > the iGPU could easily double or more the performance compared to Ivy Bridge.
> >
> > More pressure on AMD and the low end discrete cards ...
> >
>
> If the leaked numbers are correct it's possible that they refer to turbo frequency, not base frequency.
> The column title "dynamic frequency" does imply this. In this case it's only a little bit higher than the
> highest bin on IB (1.15GHz), and lower that the highest bin on SB (1.35Hz). Combined with a higher EU count
> (20 vs 16) that gives a peak 35.9% improvement, given no other adjustments. In practice it'd tend to be
> less if the TMU capabilities haven't changed and due to main memory bandwidth staying the same.
I think you are right.
I thought that dynamic frequency was a strange name for GPU base frequency :-)
So maybe we only get ~30% GPU improvement ... which means that we don't need a story about the iGPU not getting killed by memory bandwidth.
> Mark Roulo (nothanks.delete@this.xxx.com) on December 12, 2012 3:53 pm wrote:
> > Maybe this is new only to me, but the iGPU clock speed seems to have
> > almost doubled in Haswell compared to Ivy Bridge: 1250 vs. 650.
> >
> > *IF* Intel has also increased the iGPU in other ways (widening it, whatever ...) *AND*
> > can solve the bandwidth starvation issue (Crystalwell maybe??), then it looks like
> > the iGPU could easily double or more the performance compared to Ivy Bridge.
> >
> > More pressure on AMD and the low end discrete cards ...
> >
>
> If the leaked numbers are correct it's possible that they refer to turbo frequency, not base frequency.
> The column title "dynamic frequency" does imply this. In this case it's only a little bit higher than the
> highest bin on IB (1.15GHz), and lower that the highest bin on SB (1.35Hz). Combined with a higher EU count
> (20 vs 16) that gives a peak 35.9% improvement, given no other adjustments. In practice it'd tend to be
> less if the TMU capabilities haven't changed and due to main memory bandwidth staying the same.
I think you are right.
I thought that dynamic frequency was a strange name for GPU base frequency :-)
So maybe we only get ~30% GPU improvement ... which means that we don't need a story about the iGPU not getting killed by memory bandwidth.
