Enflame Cloudblazer T20 32GB
Enflame Technology Cloudblazer T20 is an AI accelerator on GCU-CARE 2.0 architecture, also known as DTU (Deep Thinking Unit) — Enflame's proprietary neural network computing architecture.
Manufactured on GlobalFoundries' 12nm process. Features 32 GB HBM2e (1800 GB/s), FP8, BF16, FP16, INT8 support.
TopsRider software stack is compatible with TensorFlow, PyTorch, and ONNX.
GCU-LARE interconnect delivers 300 GB/s between cards. Widely used for CNN/RNN model training, financial analytics (risk analysis, fraud detection), government administration systems, and high-performance data center clusters.
VRAM
32
FP32
32
TDP
300
Interface
PCIe 4.0x16
Pre-order
Product status
Memory Specifications
VRAM
32
VRAM Type
HBM2e
Memory Bandwidth
1500
Interface
PCIe 4.0x16
Interconnect Type
GCU-LARE
Interconnect Speed
300
Computing Performance
FP64 Vector-
FP32 Vector32
FP16 Vector128
TF32 Tensor128
FP/BF16 Tensor128
FP8 Tensor-
INT8 Tensor256
Graphics & Thermal
Pixel Rate (GPixels/s)
-
Texture Rate (GTexels/s)
-
TDP
300
Cooling Type
Пассивное
Form Factor
PCIe
Video Encoding
-
Video Decoding
-
Physical Dimensions
Slots
2
Length
- mm
Height
- mm
Width
- mm
Architecture
Architecture
GCU-CARE
Cores
-
Price
On request