The Speedster7t FPGA is highly optimized for AI/ML, compute, and is at the heart of VectorPath accelerator cards.
Achieve an industry-leading tokens per second output with open source LLMs, accelerating text generation, chatbot responses, and AI-powered content creation while delivering significantly lower latency for AI workloads leveraging various size foundational or fine-tuned and task specific LLMs vs. traditional GPU and CPU solutions.
Transcribe up to 2,400 simultaneous audio
streams with multiple language support,
latency under 25ms, and an industry-leading
word error rate (WER) of less than 3%, ensuring real-time, low-latency speech-to-text conversion for real-time agent assist in contact centers, voice assistants, transcription services, and real-time conversational AI applications.
Wire-rate 400 GbE capture, real-time traffic analysis, packet steering and more are all possible. With 16 channels of GDDR6, each operating at 256 Gbps, combined with 16x 100 GbE creates a platform that is uniquely constructed to enable a whole new generation of 400 GbE appliances.
FPGA by
Brought to market in partnership with Achronix, the VP815 gives you fast GDDR6 memory, 400G networking interfaces (two QSDP-DDs), and PCIe Gen5 x16. Of course, it’s all driven by the Speedster7t FPGA with advanced 2D NoC, machine learning processors (MLPs), and programmable logic.
Generation 5.0
VectorPath 815 leverages the Speedster®7t FPGA architecture, optimized for AI/ML and high-performance computing workloads. Its Machine Learning Processor (MLP) blocks, 2D Network-on-Chip (NoC), GDDR6 memory, and 112G SerDes deliver exceptional AI inferencing performance.
5x Faster Large Memory
Using high-bandwidth GDDR6 memory, the VP815 gives your application a large memory resource of 32 Gigabytes, but at more than 5 times greater bandwidth. Plus with the 2D NoC, the GDDR6 is available for read/write from the host over PCIe without using FPGA fabric resources.
The VP815 card offers a range of network interfaces connected to the Speedster7t FPGA fabric. The card supports 112G
PAM4, with hard IP MAC and FEC support. On-board jitter cleaners are available for synchronous ethernet (SyncE).
QSFP-DD Interfaces
Two QSFP-DD interfaces with up to 400 Gbps per port. A range of other options including 16x 100G (total) are available using breakout cables.
r0 v1
Get extended warranty support and save time with a pre-integrated solution!
Our technical sales team is ready to provide availability and configuration information, or answer your technical questions.
BittWare On-Demand Webinar Using Intel® oneAPI™ to Achieve High-Performance Compute Acceleration with FPGAs Join BittWare and Intel as we look at oneAPI™ with a focus
Efficient Sharing of FPGA Resources in oneAPI Building a Butterfly Crossbar Switch to Solve Resource Sharing in FPGAs The Shared Resource Problem FPGA cards usually
Go Back to IP & Solutions Dynamic Neural Accelerator ML Framework EdgeCortix Dynamic Neural Accelerator (DNA), is a flexible IP core for deep learning inference
Panel Discussion How Today’s FPGAs are Taming the Data Deluge Problem From Gen5 to AI, NOCs to RF at the Edge Watch the recording for
