Available DPUs for your production infrastructure
DPU cards based on the Kalray MPPA® architecture, engineered for AI inference, storage, networking, and security workloads.
Coolidge v2 Processor
The MPPA® Coolidge v2 processor at the heart of Kalray's K300 and TC4 products. Available for design-in as a standalone processor for OEMs and system builders integrating a DPU directly on their board.
PROCESSOR
MPPA® Coolidge v2 @ 1 GHz
NUMBER OF CORES
80 RISC-V cores
ACCELERATION
Integrated AI accelerator, cryptography, FEC, data compression
INTERFACES
Ethernet, PCIe Gen4, DDR4
K200-LP
DPU card offering an unprecedented level of performance and programmability, a game-changing solution in performance per Watt and per dollar. Optimized for storage, security, and networking workloads.
PROCESSOR
1× MPPA® Coolidge v1 @ 1 GHz
MEMORY
8GB DDR4 @ 3200MT/s
HOST INTERFACE
PCIe Gen4 16-lane
ACCELERATION
NVMe-oF (RoCE / TCP), encryption, compression, erasure coding
K300
Versatile DPU card delivering exceptional performance and energy efficiency across AI, smart vision, networking, security, and telecoms. Integrates seamlessly with Kalray's open programming environment.
PROCESSOR
1× MPPA® Coolidge v2 @ 1 GHz
MEMORY
16GB DDR4 @ 3200MT/s
HOST INTERFACE
PCIe Gen4 16-lane
ACCELERATION
FEC (LDPC), cryptography, data compression
PERFORMANCE
40 TOPS (INT8) · 20 TFLOPS (FP16)
TC4
DPU card designed to merge classical and AI-based processing for demanding industrial AI applications. Offers complementary architecture to GPUs, large numbers of parallel, asynchronous operations processed efficiently.
PROCESSOR
4× MPPA® Coolidge v2 @ 1 GHz
HOST INTERFACE
PCIe Gen4 16-lane
PERFORMANCE
160 TOPS (INT8) · 80 TFLOPS (FP16)
Turbo Server (TS1)
Plug-and-play, hyper-dense server solution for data-intensive smart-vision applications, combining two AMD EPYC 9004 CPUs with eight Kalray TC4 PCIe cards. Maximizes performance per server for AI inference and combined AI and non-AI workloads.
PROCESSOR
2× AMD EPYC 9004 (168 cores)
ACCELERATION CARDS
8× Kalray TC4
Server Performances
1280 TOPS (INT8) · 640 TFLOPS (FP16)
MEMORY
24 slots of 12-channel RDIMM DDR5
EXPANSION
8× PCIe Gen5 ×16 FHFL slots
Software
Kalray SDK with 80+ AI libraries, Docker, Triton Inference Server API
All Technical Specifications