Key Moments
- AOCL 6.0.0 is tailored for 6th Gen AMD EPYC™ “Venice” CPUs built on the Zen 6 architecture, targeting AI, analytics, and HPC workloads.
- Early tests show up to 1.84x faster per-core INT8 GEMM for AI inference and approximately 15% gains in memory-intensive operations versus AOCL 5.3.
- AOCL 6.0.0 introduces AOCL-OpenRNG, an open-source RNG library with 11 base generators and 17 distributions optimized for AMD processors.
Overview of AOCL 6.0.0 for Zen 6 “Venice” EPYC CPUs
AMD has introduced AOCL 6.0.0, a major update to the AMD Optimizing CPU Libraries designed specifically for the newly launched 6th Gen AMD EPYC™ server processors, code-named “Venice,” based on the Zen 6 architecture. The new release focuses on boosting performance across AI inference, numerical workloads, memory operations, and data analytics by taking advantage of Venice-specific architectural features.
According to AMD, the gains stem from more effective use of cores, cache hierarchies, memory bandwidth, and vector units in the new processors.
Benchmark Improvements Versus AOCL 5.3
Initial benchmarking of AOCL 6.0.0 against the prior AOCL 5.3 release indicates notable performance jumps across several workload categories.
| Workload Area | Improvement vs. AOCL 5.3 | Notes |
|---|---|---|
| AI inference – per-core INT8 GEMM | Up to 1.84x faster | Combines hardware and software optimizations on Venice |
| Selected mathematical functions | Up to 1.5x higher performance | Within AOCL-BLAS and related math components |
| Memory-intensive operations | Approximately 15% improvement | Driven by AOCL-LibMem enhancements |
| Analytics performance per core | 1.3x higher | On Venice systems using AOCL-DA |
| Selected nearest-neighbor analytics workloads | Up to 4x acceleration | Within AOCL-DA ANN use cases |
AI Inference and Memory Performance Tuning
The AOCL-DLP component, which targets AI inference, has been reworked for Venice to optimize INT8 and FP16 execution paths. This tuning delivers up to 1.84x faster per-core INT8 GEMM performance, reflecting a combination of software refinements and the capabilities of the underlying hardware.
Memory handling is another focus area. AOCL-LibMem incorporates updates that accelerate core memory operations critical to databases, analytics engines, and AI runtimes. The changes include:
- Up to 30% faster
memcpyperformance - Up to 20% faster
memsetoperations
HPC and Scientific Computing Enhancements
For high-performance computing and scientific workloads, AOCL-BLAS has been refined to raise DGEMM efficiency from 95.5% to 97.3%. In addition, selected mathematical functions now run up to 1.5x faster compared with AOCL 5.3.
These improvements are designed to be drop-in for existing applications, allowing users to realize performance benefits without modifying source code.
Data Analytics: New APIs and Faster Models
The AOCL-DA analytics library expands its feature set and performance profile on Venice-based systems. New APIs are available for:
- Nonnegative matrix factorization
- Kernel principal component analysis (kernel PCA)
- t-SNE
AOCL-DA also includes optimizations for commonly used machine learning models and algorithms such as linear models, support vector machines (SVMs), and decision trees. On Venice processors, AOCL-DA delivers approximately 1.3x higher per-core performance, while certain approximate nearest neighbor (ANN) workloads see speedups of up to 4x.
Launch of AOCL-OpenRNG
AOCL 6.0.0 debuts AOCL-OpenRNG, an open-source random number generation library tuned for AMD architectures. The library features:
- 11 base random number generators
- 17 statistical distributions
- AVX2-optimized kernels
AOCL-OpenRNG targets simulation, financial modeling, and AI use cases where high-throughput and high-quality random number generation is essential.
Positioning of 6th Gen EPYC for Modern Workloads
By combining these software updates within AOCL 6.0.0 with the capabilities of the Zen 6-based “Venice” EPYC processors, AMD is positioning its 6th Gen EPYC lineup as a strong platform for AI, data analytics, and HPC environments. The company highlights improved performance and efficiency as key benefits for users running modern, compute-intensive workloads on these systems.





