Our partnership with Blackcore® Technologies continues to push the boundaries of low-latency AI inference and real-time performance. Since our initial collaboration, we’ve combined Xelera’s Silva AI acceleration framework with Blackcore’s high-performance server platforms to deliver fully optimized systems for trading and high-performance computing workloads where every microsecond counts.
Over the course of this partnership, we’ve released three joint benchmark reports, each one validating the integration between Silva and Blackcore’s ACE server line-up. Our latest results mark a new standard.
🚀 Xelera Silva on the Blackcore® ACE 3100-RZ
The new benchmark demonstrates that Xelera Silva running on the Blackcore® ACE 3100-RZ with an AMD Alveo U50 accelerator sets a new standard for high-frequency trading environments, where speed and consistency directly impact outcomes. The overclocked ACE 3100-RZ achieves ultra-low PCIe latency, allowing the FPGA to receive instructions and return results with minimal delay — a crucial advantage when nanoseconds can separate wins from losses. Combined with Silva’s tightly bounded inference latencies, even under parallel workloads, the platform removes bottlenecks that typically occur between hardware acceleration and software execution.
The benchmark results highlight:
✅ 44× faster inference compared to CPU-based software
✅ Stable and predictable latency, with 99th percentile below 1.6 µs
✅ Concurrent inferences handled without performance degradation
For trading firms, this means deterministic performance, seamless hardware–software integration, and scalable real-time decision-making — enabling ML-driven strategies to execute without compromise.
Together, Xelera and Blackcore continue to demonstrate that AI inference can deliver both ultra-low latency and uncompromising reliability, setting a new benchmark for performance in latency-critical computing.
👉 Read the full benchmark report here!
👉 Read the blackcore press release here!
%201.svg){kind=icon}