Skip to main content

Enflame CloudBlaze i10

Enflame's first AI inference accelerator, based on self-developed Suisi 1.0 chip, released in December 2020, single-slot design, targeting cloud AI inference scenarios.

Core Specifications​

SpecificationValue
ArchitectureSuisi 1.0 (DTU dataflow architecture)
ProcessGF 12nm FinFET
TDP150 W
Memory16 GB HBM2 (estimated)
Memory Bandwidth512 GB/s (estimated)
FP32 Compute17.6 TFLOPS
FP16 / BF16 Compute70.4 TFLOPS
INT8 Compute140.8 TOPS (estimated)
InterfacePCIe 4.0 x16, single-slot
Launch Date2020-12-21
PriceNot disclosed (estimated ¥25,000)

Technical Highlights​

  • Single-slot design: One of the few high-performance AI inference cards with single-slot design, allowing 8 cards to be deployed in 1U servers, significantly improving inference density
  • Suisi 1.0 chip: Same chip as CloudBlaze T10, optimized software stack for inference scenarios
  • TopsInference engine: Companion TopsInference acceleration engine supporting INT8/FP16 mixed-precision inference
  • Multi-scenario coverage: Suitable for computer vision, natural language processing, speech recognition and synthesis, knowledge graph and other inference scenarios
  • High-density deployment: Single-slot design enables 1U servers to configure 8 i10 cards, greatly improving inference throughput

Product Positioning​

CloudBlaze i10 is the inference product in Enflame's "training + inference" product matrix. Its single-slot high-density design is its biggest feature. It targets NVIDIA T4 and focuses on cloud AI inference scenarios.

Application Scenarios​

  • Cloud AI inference services (CV, NLP, ASR/TTS)
  • Internet recommendation system inference
  • Smart city video analytics
  • Financial real-time risk control inference
  • Edge inference servers

Reference Price​

ChannelPriceDescription
Official pricingNot disclosedReleased in 2020, estimated ¥25,000/card
Channel estimate≈ ¥20,000–30,000Estimated based on T4 pricing ratio

References​