Now AMD is looking into a new technology that might help them get around this barrier. According to a leaked presentation AMD is looking into Stacked DRAM as a way to increase the memory bandwidth while reducing latency. This is all part of an initiative to introduce High Bandwidth Memory into their future chip designs and should be happening alongside their Heterogeneous System Architecture initiative and also their work on streamlining APIs.
This holy trinity of projects could (if done right) result in new processors (APU) that are unlike what we are used to today. Think about it this way: GDDR5 can provide up to 28GB/s of bandwidth while even a single stack of DRAM would offer as much as 64GB/s and two would double this to 128GB/s. That level of memory bandwidth could be a game changer if it can be incorporated into the APU correctly.
It is an impressive goal to reach for when you stop and think about it. Before you start drooling about the potential for gaming and other desktop processes we need to let you know that this technology is being investigated for high volume architectures at the moment. This means it is aimed at the big data center and large (exascale) systems. It is not likely that this will trickle down to the desktop/laptop for a few years after it hits the high-end server market.
Still it is nice to know that AMD understands their weak point and is actively working on a solution. Perhaps the day will come when AMD really can stand toe to toe with Intel again. I really hope this is the case. I would hate to think that the years of the Athlon 64 were a fluke. Getting AMD back into a position to compete with Intel performance wise will be good for everyone… fingers crossed.
Tell us what you think in our Forum