Statements (48)
| Predicate | Object |
|---|---|
| gptkbp:instanceOf |
gptkb:personal_computer
|
| gptkbp:abbreviation |
gptkb:High_Bandwidth_Memory
|
| gptkbp:application |
gptkb:machine_learning
data centers supercomputers gaming GPUs |
| gptkbp:competitor |
gptkb:GDDR6
gptkb:GDDR5X |
| gptkbp:connectedTo |
silicon interposer
|
| gptkbp:dataRate |
up to 1 Gbps per pin (HBM1)
up to 2.4 Gbps per pin (HBM2) up to 3.2 Gbps per pin (HBM2E) up to 6.4 Gbps per pin (HBM3) |
| gptkbp:developedBy |
gptkb:JEDEC
|
| gptkbp:feature |
low power consumption
high density high bandwidth wide interface high energy efficiency parallel data access 3D stacking higher cost than GDDR limited capacity per stack lower latency than GDDR reduced PCB footprint short signal paths thermal challenges used in ASICs used in FPGAs |
| https://www.w3.org/2000/01/rdf-schema#label |
HBM (High Bandwidth Memory)
|
| gptkbp:introducedIn |
2013
|
| gptkbp:memoryBusWidth |
wide I/O
|
| gptkbp:memoryStackHeight |
up to 8 dies
|
| gptkbp:memoryType |
stacked DRAM
|
| gptkbp:predecessor |
gptkb:HBM2
gptkb:HBM2E gptkb:HBM3 |
| gptkbp:standardizedBy |
gptkb:JEDEC
gptkb:JEDEC_JESD235 |
| gptkbp:successor |
gptkb:GDDR5
|
| gptkbp:usedBy |
gptkb:NVIDIA_Tesla_P100
gptkb:AMD_Fiji_GPUs |
| gptkbp:usedIn |
high-performance computing
graphics cards artificial intelligence accelerators |
| gptkbp:width |
1024 bits
|
| gptkbp:bfsParent |
gptkb:Neural_Engine
|
| gptkbp:bfsLayer |
6
|