pre-built dpus
Off-the-Shelf DPU Acceleration Cards
For customers seeking immediate deployment, Kalray offers pre-designed, production-ready DPU solutions based on the MPPA architecture. These DPUs target well-understood workloads and can be used as standalone accelerators or integrated into larger systems.
Networking
AI
Security
Compute
Storage
K200-LP Acceleration Card
Kalray’s K200 is a family of data-centric acceleration cards. The K200-LP storage acceleration cards offer an unprecedented level of performance and programmability, a game changer solution in terms of performance per Watt and per dollar. Kalray K200-LP cards are easy to configure with the Easy Programming and Open software environment.
PROCESSOR
1x MPPA®3-80 v1.2 @ 1Ghz
NUMBER OF CORES
80 cores
MEMORY
8GB DDR4 @ 3200MT/s
SSD DEVICE SUPPORT
Western Digital, Micron, Intel, Kioxia, Hynix, Seagate, Samsung, Solidigm
DEVICE CAPACITY
Up to 24x 30TB NVMe SSD
HOST INTERFACE
PCIe Gen4 16-lane
Security
Compute
Storage
TC4 Acceleration Card
Kalray’s TC4 is designed to ensure customers can merge classical and AI-based processing technologies and create superior, efficiency-driven systems for the most processing-intensive AI applications.
TC4 offers complementary architecture to GPUs, allowing the processing of a large number of different operations in parallel, asynchronously.
PROCESSOR
4x MPPA®3-80 v2
NUMBER OF CORES
320 cores
MEMORY
32GB DDR4 @ 3200MT/s
PERFORMANCES
INT8: 174 TOPS
FP16: 92 TFLOPS
FP16: 92 TFLOPS
SUPPORTED FORMATS
INT8/16/32/64, FP16/32/64
HOST INTERFACE
PCIe Gen4 16-lanes
Security
Compute
Storage
K300 Acceleration Card
Kalray’s K300 cards deliver exceptional performance, and energy efficiency. Designed for versatility, K300 supports a wide range of use cases, including AI, Smart vision, networking, security, and telecoms. It integrates seamlessly with Kalray’s Easy Programming and Open software environment for fast and flexible deployment.
PROCESSOR
1x MPPA®3-80 v2
NUMBER OF CORES
80 cores
MEMORY
16GB DDR4 @ 3200MT/s
PERFORMANCES
INT8: 45 TOPS
FP16: 22.5 TFLOP
FP16: 22.5 TFLOP
ETHERNET
4x 10/25G SFP28 interfaces
HOST INTERFACE
PCIe Gen4 16-lanes
AI INFERENCE
Computer Vision
Industry 4.0
Turbo Server
The Kalray TS1 is a plug&play hyper dense server solution targeting data-intensive smart vision type of applications requiring to maximize performances per server for pure AI inference or combination of AI and non-AI workloads.
PROCESSOR
2 AMD EPYC 9004 CPU (168 Cores)
NUMBER OF ACCELERATION CARDS
8 Kalray TC4 PCIe Cards
Server Performances
FP16 INT8 670 TFLOPS 1300 TOPS
MISC
- 2U Format
- 24 slots of 12-channel RDIMM DDR5
- 8 PCIe Gen5 x16 FHFL slots
- Networking Card (Ethernet or Infiniband)
Software
- Delivered with Kalray SDK including 80+ optimized AI network and compute librairies
- Docker Container Support
- Triton Inference Server API
TDP
3000W typical
All Technical Specifications
Kalray’s MPPA: The architecture behind smarter DPU acceleration.
Kalray’s response to this change is its Massively Parallel Processing Architecture (MPPA) - a scalable, modular architecture designed to power the next generation of customized DPU accelerators.
Customizable
Purpose-built for customization, allowing domain-specific optimizations without starting from scratch.
Modular
Allowing different compute, memory, and I/O blocks to be combined and optimized for each use case.
Programmable
Enabling developers to tune and adapt acceleration behavior through standard software environments.
Scalable
Supporting deployments from low-power edge devices to high-throughput data center systems.
Kalray is recognized by industry analysts and certified by leading quality and interoperability organizations.
Kalray is recognized by industry analysts and certified by leading quality and interoperability organizations.
Analyst Recognition
GigaOm recognized Kalray as a Leader and Fast Mover in the 2023 Data Processing Units (DPU) Sonar Report — highlighting our innovative architecture and strong performance across AI and data workloads.
Quality Certifications
Kalray’s ISO 9001 certification and interoperability validations demonstrate our commitment to high-quality standards and sustainable performance in every solution we build.
