Product overview
Purpose-built accelerator for enterprise AI workloads
Designed for dense inference, model adaptation, and private AI infrastructure where predictable availability, localized supply, and software compatibility are critical.
Architecture
MUSA
Memory & Performance
High-bandwidth memory profile for AI inference and training.
VRAM capacity48
Memory typeGDDR6
Memory bandwidth768
Interconnect typeMTLink
Interconnect speed112
Architecture
Compute architecture and software execution model.
ArchitectureMUSA
Compute units8192
Power & Thermal
Data center integration requirements.
Thermal design power450
CoolingПассивное
Form factorPCIe
Pixel Rate768
Texture Rate768
Benchmarks
Peak theoretical compute for common AI precisions.
FP64
-
FP32
25
FP16
25
TF32
50
BF16 Tensor
100
FP8 Tensor
-
INT8 Tensor
200
Compatibility
Interfaces, frameworks, and deployment environment.
PCIe interfacePCIe 5.0x16
Video encoding48x1080p30fps
Video decoding96x1080p30fps
Physical Dimensions
Card dimensions for server platforms.
Slots2
Length266 mm
Height112 mm
Width39 mm
Pricing
On request
Volume pricing available for cluster deployments and pilot batches.
Documentation