In the site of Hybrid Memory Cube Consortiumis freely available the first version of the new memory specifications. The memory Hybrid Memory Cube (HMC) - the fruit of engineering company Micron. More precisely, Micron engineers offered their stack layout of memory chips, placing the very base of the stack memory controller, which is now located in the processors.
The consortium refuse to include companies such as Intel and NVIDIA, and without them on General Industry HMC standard speak highly premature. NVIDIA company introduced the concept of video on microarchitecture Volta. Under this concept, NVIDIA is planning to use the stack memory on a single substrate with a graphics processor(2,5 D-linking), but continues to ignore the standard Hybrid Memory Cube . True, this does not prevent them to use on their slides HMC image . The Hybrid Memory Cube standard is supported by more than 100 companies, including Microsoft, ARM, AMD, HP, Cray, Fujitsu, IBM, Marvell, ST Microelectronics and Xilinx. More importantly, HMC combines three leading memory manufacturers - Micron, Samsung and SK Hynix.
The standard provides two options for the location of HMC modules to the processor: in the short run, which means that at a distance of 8 to 10 inches (25 cm), and ultrashort - 1 to 3 inches. In the first case the rate will reach 15 Gbit / s per channel (28 Gbit / s in the next generation), the second - 10 Gb / s per channel (15 Gbit /s in the next specifications version ). There can be 4 or 8 channels, each of which will be on the 16 input and output lines (full duplex).
The first memory Hybrid Memory Cube Micron company expects to receive by June this year. It will be 2-GB and 4-GB modules with a total interface speed 160 GB / s in both directions. Memory speed increase up to 15 times, while the power consumption will be reduced by up to 70%. Reducing consumption, recall, is expected by reducing the length of connections between individual memory crystals . Related Products :
|