HWM (Malaysia) - - LEARN -

While these HBM stacks are not phys­i­cally in­te­grated onto the GPU die, they are so closely con­nected that they are al­most in­dis­tin­guish­able from on-die in­te­grated RAM. This close prox­im­ity en­ables ex­tremely wide mem­ory bus widths, and con­se­quently al­lows much lower mem­ory clock speeds to be used to achieve the re­quired per­for­mance.

This is be­cause with GDDR5 mem­ory, in­di­vid­ual DRAM chips are placed side by side and con­nected via long cop­per traces on the PCB. How­ever, HBM dies are so close to the GPU that the traces that con­nect them are much shorter, which thus en­ables much wider mem­ory bus widths.

On the AMD Radeon R9 Fury X and Fury, each HBM tower has a 1,024-bit mem­ory bus width, and the four tow­ers flank­ing the GPU die com­bine for a 4,096-bit bus width, the high­est on any GPU so far. From buses just hun­dreds of bits wide, we've rock­eted to see­ing fig­ures in the thou­sands, and this is just a first-gen­er­a­tion prod­uct.

Both of AMD's HBM-equipped GPUs have a 512GB/s of mem­ory band­width, beat­ing even the 336.5GB/s of the NVIDIA GeForce GTX TI­TAN X.

Power con­sump­tion is also much re­duced be­cause the closer prox­im­ity en­ables more ef­fi­cient com­mu­ni­ca­tion be­tween the mem­ory and GPU. In­put volt­age is down to 1.3V, from 1.5V on tra­di­tional GDDR5 mem­ory.

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.