Welcome to the Altera Community!

The computing world is hitting a wall. As AI models grow to trillions of parameters, as in-line databases scale to massive sizes, and as high-performance computing (HPC) workloads push bandwidth and memory to their limits, the need for more efficient data movement has never been greater. Traditional approaches to scaling bandwidth and capacity can’t keep pace without unsustainable cost expenditures on power usage and infrastructure build-out. Compression offers a practical and elegant solution to this challenge. By reducing the size of data that moves across interconnects, we can stretch bandwidth, improve memory efficiency, and lower system power—all without requiring a fundamental re-architecture. The Open Compute Project (OCP) has recently recognized this reality, highlighting compression as a key enabler for modern workloads. The combination of ZeroPoint Technologies (an Altera Partner), advanced compression IP, and Altera’s CXL Type 3 IP and FPGAs results in a 2–3x increase in bandwidth, giving the industry a proven path to meet the growing demand head-on. The Problem: Data Bottlenecks in Today’s Workloads AI and LLMs Large language models are exploding in size—parameters have grown from millions to billions, and now to trillions, in just a few short years. Training and inference of these models are fundamentally constrained by memory bandwidth and capacity. Without compression, these models would require even larger amounts of data movement, which increases latency, power consumption, and cost. In-line Databases Databases are increasingly run in-line with applications, from analytics pipelines to transaction processing. These in-line databases demand high throughput and low-latency access to massive datasets. Without compression, systems are forced to overprovision bandwidth and memory resources, dramatically increasing the total cost of ownership (TCO). High-Performance Computing (HPC) From climate modeling to genomics, HPC workloads require immense amounts of parallel data movement. Without compression, HPC centers must continue scaling raw interconnect bandwidth, which is unsustainable in terms of energy and cost at exascale levels. CXL Expansion (CXL Device Type 3) CXL (Compute Express Link) has emerged as the industry-standard protocol for memory pooling and expansion. Yet, as more systems adopt CXL for disaggregated memory, the sheer volume of data moving across CXL links risks overwhelming interconnect bandwidth. Without compression, the benefits of CXL expansion hit a hard ceiling. Demo Video: ZeroPoint demonstrates 2–3x increased bandwidth using its CXL compressed memory tier solution at the Future of Memory and Storage (FMS) 2025 CXL Acceleration (CXL Device Type 2) Beyond memory expansion, CXL enables accelerators to share memory seamlessly with CPUs. But in accelerator-heavy environments, data transfer volumes explode. Lack of compression makes accelerator scaling inefficient, power-hungry, and cost-prohibitive. Contact Altera to see the demo video: 2x–6x higher QPS running a VectorDB workload using a CXL 2.0 interface. Without compression, every one of these workloads faces a bottleneck that would be extremely difficult to solve with hardware scaling alone. OCP Introduces Compression into its Specification The Open Compute Project (OCP) organization recently underscored the importance of compression by including it in its specifications. This is a landmark shift: compression is no longer viewed as optional but included as a supported feature for next-generation compute infrastructure. James Kelly, VP Market Intelligence and Innovation at the OCP Foundation, said: “Within the OCP Community, our Composable Memory Systems Project, leveraging CXL and compression technologies, is driving the development of interoperable, scalable memory architectures that empower AI workloads with unprecedented efficiency and flexibility. By enabling disaggregated memory resources to be pooled and allocated across heterogeneous systems, we’re directly supporting OCP’s Open System for AI strategic initiative, fostering open specifications and standards that accelerate innovation and accessibility in AI infrastructure.” Klas Moreau, CEO of ZeroPoint Technologies, added: “What excites us about working with Altera’s CXL Type 3 IP is not just its performance, but its flexibility. Unlike other FPGA providers, Altera’s CXL solution gives us the low-latency, high-bandwidth fabric we need to showcase the full potential of our compression IP. Together, we’re able to deliver measurable gains—up to a 2–3x effective bandwidth increase—without changing the underlying hardware footprint. That’s a game-changer for customers scaling AI, HPC, and database workloads.” The Solution: ZeroPoint Compression IP + Altera CXL Type 3 IP and FPGA-based Boards ZeroPoint Compression Technology ZeroPoint brings a powerful, low-latency, hardware-efficient compression engine designed specifically for memory and interconnect applications. Unlike general-purpose compression algorithms, ZeroPoint’s IP is optimized for inline operation at wire speed, ensuring data is compressed and decompressed seamlessly without introducing overhead. Key benefits include: High compression ratios across AI, HPC, and database workloads Ultra-low latency to avoid bottlenecks on memory paths Energy savings by reducing data movement requirements Proven scalability across CXL and memory expansion use cases Altera CXL Type 3 IP Altera’s CXL Type 3 IP provides the foundation for memory expansion and pooling. It enables compute nodes to access disaggregated memory resources efficiently and securely. By integrating ZeroPoint’s compression IP, Altera’s solution extends even further—allowing CXL links to move more effective bandwidth, reduce congestion, and scale system capacity without increasing physical resources. There is a wide variety of CXL-capable FPGA-based boards available from Altera or partners. Together: Meeting the Market Need When combined, ZeroPoint’s compression IP and Altera’s CXL Type 3 IP address the OCP-driven specification requirements and solve the core problem facing data-intensive applications, ranging from AI to databases: moving massive amounts of data efficiently. Benefits to customers include: More bandwidth without more lanes: Compression effectively multiplies CXL throughput. Boost performance, cut costs: Unleash untapped performance in your current infrastructure with minimal new investment. Future-proof compliance: Alignment with OCP specifications ensures long-term viability. This combination delivers not just a technology improvement, but a market-ready solution that meets both current and emerging requirements. Conclusion The computing industry is shifting to adjust to new demands. AI, HPC, databases, and disaggregated systems are demanding exponential growth in bandwidth and memory efficiency—growth that hardware scaling alone cannot deliver. One answer is compression. OCP’s inclusion of compression in its specifications validates this direction and creates a mandate for solutions that integrate compression seamlessly with interconnect technologies like CXL. Through the combination of ZeroPoint’s cutting-edge compression IP and Altera’s CXL Type 3 IP, customers can now confidently deploy systems that are not only faster and more efficient but also aligned with the industry’s forward-looking standards. The future of computing depends on smarter ways to move and manage data. Compression + CXL is that smarter way—and with ZeroPoint and Altera, the future is already here. Learn More Presentations or videos are available for on-demand viewing or download: FMS 2025 session (video | slides) OCP 2025 session (video | slides) Next Steps Learn more about Altera’s CXL IP core. For technical details, partnership discussions, or general inquiries, please contact: nilesh.shah@zptcorp.com — CXL compression solutions phillip.swart@altera.com — FPGA-based CXL IP and boards