If we're comparing it to dedicated chips, where did IB's IGP stand?
On the mobile side, its roughly comparable to the 610M.
What's interesting though:
-HD 4000 has nearly double the bandwidth with DC DDR3-1600 offering 25.6GB/s bandwidth versus 610M's 14.4GB/s
**
-HD 4000 has more than double the Flops with 307GFlops @ 1.2GHz versus 610M's 142GFlops
-However, the Texture Fillrate is not much far away between the two, with the HD 4000 having 9.6GTexels/s and 610M having 7.2GTexels/s. There's also an addendum to that
.*
*While
theoretical texture fillrate is decently higher on the HD 4000, it stays further away from theoretical in actual measurements compared to the predecessor HD 3000. That suggests the TMUs
may be clocked asynchronously to the main clock, perhaps at ~85% of the speed.
**Traditional scaling shows an unofficial rule of thumb to bandwidth and performance relationship of
30% for every 100% increase in memory bandwidth. Chipset based IGPs generally performed 30% lower than discrete counterparts despite having the same theoretical bandwidth. Perhaps the nature of sharing with the CPU makes the practical usable bandwidth 50% of the theoretical on the iGPUs compared to the discrete parts. That means HD 4000 with DC DDR3-1600 and 610M
may in fact have similar bandwidth as well.
http://www.anandtech.com/show/6332/amd-trinity-a10-5800k-a8-5600k-review-part-1/7