Advanced Compute Extensions

AMD and Intel are steering the CPU into a more active role in artificial intelligence through Advanced Compute Extensions, or ACE. The shift does not aim to replace GPUs in large-scale AI training, but it does target a part of the market that has often been overlooked.That target includes smaller AI models, latency-sensitive workloads, and systems that do not have a GPU or do not need the overhead that comes with one. In those cases, keeping the work on the CPU can help avoid data-transfer bottlenecks between processors.A Different AI Role for the CPUThe core argument behind ACE is practical rather than dramatic. Many AI tasks now rely on fast movement between CPU and GPU, but that transfer can become a performance cost in constrained systems.For edge computing and single-user scenarios, that cost matters even more. A CPU that can handle more of the AI workload directly may offer a simpler path with lower latency.ACE is designed for x86 processors and focuses on matrix multiplication, which sits at the heart of many modern AI operations. CPUs have already handled similar math through AVX instructions, but those instructions were never built specifically for heavy matrix workloads.By preserving the AVX10 register structure with 512-bit input while adding dedicated hardware for matrix operations, ACE aims to make those workloads more efficient. That approach also helps software teams because it keeps a familiar data structure in place.Efficiency Gains Without a GPU DependencyOne technical claim tied to ACE is that it can perform up to 16 times more operations than AVX10 for certain input vector sequences. That figure does not mean every application becomes 16 times faster, but it does point to a substantial improvement in instruction-level efficiency.Better instruction efficiency can also reduce power consumption and memory bandwidth pressure. Those traits make the architecture more relevant for devices and workloads where energy use and responsiveness matter.For developers, the appeal is not just about raw speed. ACE is intended to support a more consistent x86 ecosystem, so developers should not need to constantly adapt code to different AVX implementations.That consistency could matter for popular frameworks such as PyTorch and TensorFlow. If certain AI tasks can stay on the CPU with acceptable performance, deployment becomes simpler and less dependent on accelerator-specific tuning.Why This Matters Beyond Technical SpecsThe arrival of ACE also highlights a broader issue in AI hardware: NPUs are spreading quickly, but standardization still lags. Moving workloads to an NPU can create new compatibility questions depending on the hardware in use.In that environment, a more capable CPU becomes an attractive fallback. ACE gives AMD and Intel a way to keep x86 processors relevant as AI moves into more devices and more real-time services.GPU remains the dominant platform for massive AI training, and that is not changing here. The real change is that CPU now has a clearer route into the AI tasks where efficiency, latency, and simplicity matter most.For the market, that means the contest is no longer only about maximum compute power. It is also about which processor family can do the right job with the least friction, and AMD and Intel clearly want CPUs to have a bigger say in that answer. Source: inet.detik.com

AMD and Intel Push CPU Closer to AI, Nvidia’s Dominance Faces a New Test