First, let me hit you with the really important stat. A single Hybrid Memory Cube (HMC) module should deliver fifteen times the performance of a single DDR3 module. And it will do it using less energy and much, much less space.
According to the HMC Consortium – a group of companies gathered together to try and nail down a global standard for this new type of memory – modern technology has hit a wall. The improvements being sought for traditional DRAM are getting harder and harder to achieve and so we’re seeing only very slight boosts to energy efficiency and bandwidth. This is seen as a barrier to progress at a system level.
The move to DDR4 memory, which we’ll possibly see next year, is only going to provide small improvements to performance and efficiency. The HMC Consortium calls this an evolution compared to the revolutionary nature of Hybrid Memory Cube.
The HMC Consortium has now finalised the spec for HMC, and this standardisation means that companies now have a point from which to begin designs for future products which can incorporate HMC technology. Given that two of the big-hitters behind HMC – namely Samsung and Micron – are both huge names in system memory it’s no surprise that it’s taken them just seventeen months to move from the consortium’s inception to a finalised standard.
“This milestone marks the tearing down of the memory wall,” claims Micron’s VP for DRAM marketing, Robert Feurle.
And, to be fair, the way a HMC module is built is radically different from traditional DRAM. There’s a small logic layer at the base of the cube, with stacks of DRAM dies arranged one on top of the other above it. Passing through all of these layers are direct connects through the silicon itself.
Because of the stacked nature of the HMC it is estimated that a module of this sort will take up almost 90% less space than existing memory DIMMs. And with them also having that headline-grabbing 15x performance boost, as well as needing around 70% less energy per bit compared with existing DRAM tech, the claims to revolutionary status sure seem to hold up.
It’s going to be a while before we actually see the fruits of the Hybrid Memory Cube Consortium in our desktop machines, but the benefits to system memory as well as graphics memory could be huge. And a finalised standard is at least a pretty good start.
According to the HMC Consortium – a group of companies gathered together to try and nail down a global standard for this new type of memory – modern technology has hit a wall. The improvements being sought for traditional DRAM are getting harder and harder to achieve and so we’re seeing only very slight boosts to energy efficiency and bandwidth. This is seen as a barrier to progress at a system level.
The move to DDR4 memory, which we’ll possibly see next year, is only going to provide small improvements to performance and efficiency. The HMC Consortium calls this an evolution compared to the revolutionary nature of Hybrid Memory Cube.
The HMC Consortium has now finalised the spec for HMC, and this standardisation means that companies now have a point from which to begin designs for future products which can incorporate HMC technology. Given that two of the big-hitters behind HMC – namely Samsung and Micron – are both huge names in system memory it’s no surprise that it’s taken them just seventeen months to move from the consortium’s inception to a finalised standard.
“This milestone marks the tearing down of the memory wall,” claims Micron’s VP for DRAM marketing, Robert Feurle.
And, to be fair, the way a HMC module is built is radically different from traditional DRAM. There’s a small logic layer at the base of the cube, with stacks of DRAM dies arranged one on top of the other above it. Passing through all of these layers are direct connects through the silicon itself.
Because of the stacked nature of the HMC it is estimated that a module of this sort will take up almost 90% less space than existing memory DIMMs. And with them also having that headline-grabbing 15x performance boost, as well as needing around 70% less energy per bit compared with existing DRAM tech, the claims to revolutionary status sure seem to hold up.
It’s going to be a while before we actually see the fruits of the Hybrid Memory Cube Consortium in our desktop machines, but the benefits to system memory as well as graphics memory could be huge. And a finalised standard is at least a pretty good start.
