parvadomus
Senior member
- Dec 11, 2012
- 685
- 14
- 81
I can see HSA and MANTLE helping a lot on AMD's Kaveri APUs, giving AMD time to develop a better big core.
First, HSA will help with lot of parallelized things like video encoding, image processing, etc. and at the same time it could work as a dedicated physics co-processor for games (ofc they need to develop on SW side for this, but it should work better than physx).
With MANTLE, you can make games to use the APU's GPU as a co-processor for effects, for example using it for AA, shadows, etc. Helping another low end GPU to get better frame rates, and avoiding the traditional AFR problems.
About the CB score, it gained like 10% IPC, but that was without adding FPU resources, maybe for the CB workload BD/PD were doing already OK in terms of branch prediction/cache structures etc. That alone don't define IPC. In fact super PI and wprime showed that for other workloads the IPC can be about ~30%.
First, HSA will help with lot of parallelized things like video encoding, image processing, etc. and at the same time it could work as a dedicated physics co-processor for games (ofc they need to develop on SW side for this, but it should work better than physx).
With MANTLE, you can make games to use the APU's GPU as a co-processor for effects, for example using it for AA, shadows, etc. Helping another low end GPU to get better frame rates, and avoiding the traditional AFR problems.
About the CB score, it gained like 10% IPC, but that was without adding FPU resources, maybe for the CB workload BD/PD were doing already OK in terms of branch prediction/cache structures etc. That alone don't define IPC. In fact super PI and wprime showed that for other workloads the IPC can be about ~30%.