AMD CDNA
E1033639
AMD CDNA is a GPU architecture from AMD optimized for high-performance computing and data center workloads, particularly in AI and scientific applications.
All labels observed (1)
| Label | Occurrences |
|---|---|
| AMD CDNA canonical | 1 |
Statements (47)
| Predicate | Object |
|---|---|
| instanceOf |
GPU microarchitecture
ⓘ
data center GPU architecture ⓘ |
| architectureType | compute-focused GPU architecture ⓘ |
| belongsToFamily | AMD GPU architectures NERFINISHED ⓘ |
| compatibleWith | ROCm software stack NERFINISHED ⓘ |
| contrastsWith | AMD RDNA NERFINISHED ⓘ |
| designedFor |
accelerated computing
ⓘ
exascale computing systems ⓘ |
| designGoal |
maximize compute density per rack
ⓘ
optimize performance per watt in data centers ⓘ |
| developer |
AMD
NERFINISHED
ⓘ
Advanced Micro Devices NERFINISHED ⓘ |
| focusesOn |
energy efficiency for HPC
ⓘ
scalability in multi-GPU systems ⓘ throughput computing ⓘ |
| hasVariant |
AMD CDNA 2
NERFINISHED
ⓘ
AMD CDNA 3 NERFINISHED ⓘ |
| introducedFor | data center accelerators ⓘ |
| marketedAs | compute-optimized GPU architecture ⓘ |
| notOptimizedFor | consumer gaming graphics ⓘ |
| optimizedFor |
AI workloads
ⓘ
HPC workloads ⓘ data center workloads ⓘ high-performance computing ⓘ machine learning workloads ⓘ scientific computing applications ⓘ |
| segment | professional and enterprise GPUs ⓘ |
| softwareEcosystem | open-source GPU computing tools ⓘ |
| successorArchitecture |
AMD CDNA 2
NERFINISHED
ⓘ
AMD CDNA 3 NERFINISHED ⓘ |
| supports |
GPU-accelerated libraries and frameworks
ⓘ
HBM-based memory subsystems ⓘ high FP16 performance ⓘ high-bandwidth memory ⓘ matrix operations acceleration ⓘ mixed-precision compute ⓘ tensor-like compute operations ⓘ |
| targetMarket |
cloud data centers
ⓘ
enterprise AI infrastructure ⓘ supercomputing centers ⓘ |
| usedFor |
AI inference
ⓘ
AI training ⓘ large-scale parallel computing ⓘ numerical analysis workloads ⓘ scientific simulations ⓘ |
| usedIn |
AMD Instinct accelerators
NERFINISHED
ⓘ
data center GPUs ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.
subject surface form:
AMD Instinct