Ian Ye
Is memory the moat?
Running Kimi K3 at ~952 tok/s/node, AMD continues to prove its case as the winner in performance per dollar.
- Rishiraj Dutta Gupta and Wafer Team
Wafer integration with TrueFoundry AI Gateway
How Wafer's fast, OpenAI-compatible serverless inference integrates with TrueFoundry AI Gateway for unified routing, observability, and zero data retention.
- Ian Ye
Performance per dollar is getting faster and cheaper
How we served GLM5.2 on AMD MI355X at 2626 tok/s/node and 213 tok/s single stream at over 2x lower cost than Blackwell.
- Balaji Varadarajan and Wafer Team
The Inference Alpha: Maximizing Frontier Models on AMD
How DigitalOcean and Wafer unlock order-of-magnitude inference speedups on AMD GPUs for Kimi 2.5, DeepSeek V3.2, and GLM-5 through deep kernel and systems engineering.
- Ian Ye
Achieving Heterogeneous Compute One Kernel at a Time
How custom kernels pushed our AMD MI355X deployment from a tuned baseline to leading Qwen3.5 397B throughput.
- John Hahn
Quantizing Kimi K2.6 to NVFP4 for Blackwell Inference
Wafer and Parasail released wafer-ai/Kimi-K2.6-NVFP4: Blackwell NVFP4 weights for production Kimi K2.6 inference.
- Danny Willow Liu
Qwen3.6-35B-A3B on AMD MI355X: Being The Fastest With ATOM on AMD
ATOM on 8×MI355X leads public Qwen3.6-35B-A3B on Artificial Analysis decode and sustains ~15k tok/s per node at production latency.
- Emilio Andere
Announcing our Seed Round
Wafer has raised $4 million in seed funding led by Fifty Years to build AI that optimizes AI infrastructure.
- Steven Arellano
Where Did My Microseconds Go?
In our NVFP4 KernelArena suite, cpp_extension.load() hid a blind spot: JIT and CPU migration inflated kernel launch times.
- Emilio Andere
A Field Guide to Reward Hacking in AI Kernel Generation
Ten ways LLMs game GPU kernel benchmarks: timer tricks, garbage reads, caching, and the checks that catch them.
- Steven Arellano
Introducing KernelArena
Benchmark AI-generated GPU kernels, with first results on WaferBench NVFP4 (B200) and KernelBench HIP (MI300X).
- Wafer Team
Trace Compare: Compare vLLM traces across platforms
1:1 kernel mappings across providers. Diff huge vLLM traces fast with clean prefill vs decode.
- Wafer Team
Workspaces: GPU Compute for Your Coding Agent
Give your AI coding assistant direct GPUs, with no manual SSH, Docker, or infra babysitting.
- Wafer Team
Cloud Compiler Analyzer (PTX/SASS) Inside Your IDE
Cloud CUDA builds with PTX/SASS, PyTorch headers, and VS Code integration. No local CUDA install.
- Wafer Team
Nordlys Labs: 8x Faster Routing with Wafer-Guided Kernel Optimization
A non-kernel expert hit 8× on latency-critical CUDA clustering with Wafer profile-guided optimization.
- Wafer Team
Profile-Guided GPU Kernel Optimization
Profiling in our CLI broke a theory-only plateau: 11.65× on the Kimi Delta Attention kernel.
- Wafer Team
The Year of the LLM GPU Kernel Engineer
Our agent optimized AMD's topk_sigmoid kernel: 9× over PyTorch, step by step.
- Wafer Team
Case Study: A 104x (?) Speedup on KernelBench
A fused kernel claimed 104× speedup while reading garbage, and passed checks until we added a determinism guard.
- Wafer Team
Which models are the most HIP?
Frontier models wrote HIP kernels for KernelBench on MI300X. We measured which ones were correct and how fast.
- Wafer Team
wafer-ai CLI: GPU Superpowers for Your Coding Agent
GPU docs, trace analysis, and remote kernel eval for your coding agent via the wafer-ai CLI.
- Wafer Team
GPU Docs: Now Available on the Web
The GPU docs tool from our IDE extension, now a standalone web app.
- Wafer Team
Introducing ROCprofiler Compute: AMD GPU Profiling in Your IDE
AMD profiling in VS Code and Cursor: metrics, roofline, and kernel stats without leaving the editor.
- Wafer Team
Introducing Wafer's Built-in Perfetto Trace Viewer
Open Chrome trace JSON in your IDE with Perfetto: timeline, flamegraphs, SQL, metrics.
- Wafer Team
Introducing the Wafer Extension for VS Code and Cursor
Profiling (NCU), compiler explorer, and GPU docs: your GPU stack inside the editor.
- Wafer Team
Introducing Chip Benchmark: Hardware-Centric Performance Insights for AI Workloads
Chip Benchmark is our open suite for open-weight LLMs across accelerators, so you can pick hardware with real numbers.
- Wafer Team
Unlocking AMD MI300X for High-Throughput, Low-Cost LLM Inference
MI300X packs 192GB HBM3 and 5.3 TB/s bandwidth, often skipped for LLM inference. We show quantization and tuning.