Advancing AI innovation together. Built with devs, for devs. Supported through an open ecosystem. Powered by AMD. #TogetherWeAdvance
Joined November 2016
- Tweets2K
- Following116
- Followers59.1K
- Likes1.4K
New from the PyTorch community: FlyDSL helps accelerate GEMM, attention, RMSNorm, TopK and more on AMD GPUs through familiar PyTorch workflows. 🚀
FlyDSL, an open source Python kernel DSL, now plugs into PyTorch on @AMD MI350-series GPUs via torch.compile and eager mode.
Up to 1.68x geomean scaled GEMM kernel speedups and up to 3.01x end-to-end #vLLM serving speedups.
Read more: bit.ly/3W1zwUv
Big research ideas need the compute to match.
The AMD AI and HPC Cluster offers academic researchers and educators remote access to AMD Instinct GPUs and EPYC CPUs to support AI and high-performance computing projects.
We invite you to join our AMD University Program to learn more about cluster access: amd.com/en/corporate/univers…
AMD now has MonoKernel decode support on ROCm™, fusing layer-level decode operations into fewer persistent GPU kernel launches to improve responsiveness for latency-sensitive/agentic AI workloads. This development brings community-driven inference optimization into upstream vLLM.
How do you make speculative decoding fast at any batch size? Let the drafter measure the hardware.
Working with @EagleCorp, we've brought EAGLE-4 to AMD Instinct MI355X GPUs on day 0. EAGLE-Harness profiles drafting and verification costs at startup, then dynamically selects the optimal guess tree for the workload.
On Qwen3.8-27B (BF16, single GPU):
⚡ ~300 tok/s single request
⚡ 1.6× faster than built-in MTP
⚡ Performance holds from 1-32 concurrent conversations
⚡ Lossless vs. plain decoding, including Gated DeltaNet layers via FlyDSL tree verification
Details + launch scripts for SGLang and ATOM: amd.com/en/developer/resourc…
What would you build with up to 192GB of unified memory?
Meet AMD Ryzen AI Max+ PRO 495 processors. More room to experiment, create, and take on demanding AI workloads locally.
The next build is yours. Learn more about AMD Ryzen AI Max+ PRO 495: amd.com/en/blogs/2026/meet-t…
Shobha joined the AMD AI Developer Program to put her belief in open infrastructure for the agentic AI era into practice.
By building with AMD Instinct GPUs and the ROCm software stack, she's exploring how multi-agent AI workloads can be orchestrated, optimized, and made more resilient in real-world environments.
Explore Shobha's project and how she's putting AMD technology into practice: lnkd.in/eafr8DGR
What is she excited to build next? Shobha is exploring AI-assisted ROCm troubleshooting, telemetry-driven observability, intelligent inference routing, and self-healing AI agents that can detect issues and trigger automated remediation. Her goal is to help make AI infrastructure more performant, resilient, and operationally efficient on AMD GPUs.
Want your project featured? Share what you're building with the community and submit your project: lnkd.in/eCaRHtHG
ALT Testimonial featuring a person named Shobha Ranganathan describing their positive experience with the AMD AI Developer Program. The testimonial highlights hands-on work with AMD Instinct GPUs and ROCm software, along with advanced topics like AI-driven observability and self-healing infrastructure. AMD AI Developer Program logos are present.
Muse Spark Open 🤝 AMD Instinct GPUs.
Day 0 support is here for the first open weights model: Muse Spark Open.
Start building 👇
amd.com/en/developer/resourc…
Don’t miss AMD at #PyTorchCon 2026.
Learn how developers are optimizing AI workloads with @PyTorch and ROCm through technical sessions, hands-on labs, and conversations with AMD engineers.
events.linuxfoundation.org/p…
#AMDevs
ALT Promotional graphic for the PyTorch Conference North America 2023. AMD and PyTorch logos are featured. Text reads: "Build. Optimize. Scale. with AMD and PyTorch. Training at Scale, Inference Optimization, PyTorch Performance, ROCm & Open Source AI. San Jose • October 20-21." Background shows a digital visualization with glowing dots and lines.
Through the AMD University Program, we connect educators, researchers, and students worldwide with the technologies, tools, and resources to advance learning and accelerate research.
From hands-on AI and high-performance computing experiences to hardware donations and research collaboration, discover how AMD is helping shape the next generation of innovators.
Learn more: amd.com/en/corporate/univers…
We’re excited to attend #PyTorchCon North America 2026 in San Jose!
Visit the AMD booth to explore how developers are building, optimizing, and deploying AI with PyTorch and ROCm across cloud and client environments.
Stop by for hands-on demos & labs, technical sessions, research posters, and opportunities to connect with AMD engineers advancing open-source AI development.
Explore the schedule: events.linuxfoundation.org/p…
Build practical skills. Validate your expertise on AMD GPUs.
The ROCm Certified Associate program provides hands-on training in HIP programming, workload profiling and performance optimization on AMD Instinct GPUs.
Join the AMD AI Developer Program to get ROCm certified: developer.amd.com/rocm-certi…
AI at AMD retweeted
Meet EmbeddingGemma 2, our first natively multimodal open model for on-device embeddings.
It expands beyond text to unify code, images, audio, and video in a shared space. 🧵
ROCm 10.1 is here, delivering on our commitment to keep moving the AMD software stack forward.
Faster storage and I/O. New programming models. Expanded virtualization. More developer tools. And continued improvements across compilers, libraries and AI frameworks.
See what’s new for #AMDevs: rocm.blogs.amd.com/ecosystem…
Coming soon to a city near you: AMD AI DevDay World Tour. 🌏 🌍 🌎
AI developers can build, optimize, and get ROCm certified, all in one day.
See where DevDay is headed next: amd.cventevents.com/event/8c…
AI at AMD retweeted
AMD MI355X posts the best margin on every model in SemiAnalysis InferenceX agentic profit estimates:
Kimi K3: 45.7% vs 43.8% (GB300)
GLM 5.3: 63.4% vs 63.2% (GB200)
MiniMax M3: 51.5% vs 48.4% (B300)
DeepSeek V4 Pro: 87.9% (no green data)
inferencex.semianalysis.com
AMD AI DevDay is going global! First stop, our own backyard. 📍 Austin, TX
Join AMD engineers and AI developers for hands-on workshops, technical talks, in person ROCm certification, and local networking. Build practical skills for your AI projects.
Register Today! amd.cventevents.com/event/3f…
ALT Poster for AMD AI DevDay World Tour. Features a digital globe with network connections. Text reads: "Austin, Texas, AT&T Hotel and Conference Center, Oct 29, 2026."