NVIDIA Volta architecture
E758262
NVIDIA Volta architecture is a GPU microarchitecture designed for high-performance computing and AI workloads, introducing Tensor Cores to accelerate deep learning operations.
Statements (49)
| Predicate | Object |
|---|---|
| instanceOf | GPU microarchitecture ⓘ |
| architectureGeneration | NVIDIA Volta NERFINISHED ⓘ |
| codename | Volta NERFINISHED ⓘ |
| designedFor |
artificial intelligence workloads
ⓘ
deep learning acceleration ⓘ high-performance computing ⓘ |
| developer | NVIDIA NERFINISHED ⓘ |
| firstAnnouncementDate | 2017-05 ⓘ |
| firstGPUModel | NVIDIA Tesla V100 NERFINISHED ⓘ |
| flagshipComputeGPU | NVIDIA Tesla V100 NERFINISHED ⓘ |
| focus | energy-efficient performance for HPC and AI ⓘ |
| foundry | TSMC NERFINISHED ⓘ |
| includesProduct |
NVIDIA Quadro GV100
NERFINISHED
ⓘ
NVIDIA Tesla V100 PCIe NERFINISHED ⓘ NVIDIA Tesla V100 SXM2 NERFINISHED ⓘ NVIDIA Tesla V100S NERFINISHED ⓘ NVIDIA Titan V NERFINISHED ⓘ |
| introducedFeature | Tensor Cores NERFINISHED ⓘ |
| keyInnovation |
Tensor Cores for deep learning
ⓘ
enhanced SM (Streaming Multiprocessor) design ⓘ improved NVLink interconnect ⓘ independent thread scheduling ⓘ larger register file per SM ⓘ |
| marketSegment |
AI inference
ⓘ
AI training ⓘ HPC ⓘ data center ⓘ |
| memoryType | HBM2 NERFINISHED ⓘ |
| notableGPUCore | GV100 GENERATED ⓘ |
| predecessor | NVIDIA Pascal architecture NERFINISHED ⓘ |
| processNode | 12 nm FinFET ⓘ |
| successor |
NVIDIA Ampere architecture
NERFINISHED
ⓘ
NVIDIA Turing architecture NERFINISHED ⓘ |
| supports |
CUDA
NERFINISHED
ⓘ
ECC memory ⓘ FP16 operations ⓘ FP32 operations ⓘ FP64 operations ⓘ HBM2 memory ⓘ NVIDIA CUDA 9 NERFINISHED ⓘ NVIDIA NVLink NERFINISHED ⓘ Tensor operations ⓘ mixed-precision computing ⓘ |
| targetWorkload |
data analytics
ⓘ
high-performance computing clusters ⓘ neural network training ⓘ scientific simulation ⓘ |
| usedIn |
NVIDIA DGX-1 system (Volta version)
NERFINISHED
ⓘ
NVIDIA DGX-2 system NERFINISHED ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.