Follow @AMDServer for the latest AMD Data Center news, including networking, AMD EPYC processors, AMD Instinct accelerators, AI infrastructure, and more.
Santa Clara, CA
Joined January 2017
- Tweets579
- Following126
- Followers1.5K
- Likes881
Networking is an essential part of the modern data center, and now it's part of a bigger conversation.
Follow @AMDServer for the latest across the AMD Data Center portfolio, including networking, AMD EPYC processors, AMD Instinct accelerators, AI infrastructure, and more.
AI clusters move at the speed of their network.
AMD Pensando Vulcano 800 AI NIC delivers up to 2.4 Tbps per GPU and up to 13% faster AI job completion—all in a unique 3 NICs-per-GPU architecture built for AMD Helios.
Learn more: amd.com/en/solutions/ai/netw…
Scaling AI infrastructure isn't just a compute problem.
7 questions show why the network is where deployments can break down — or break through.
Learn more: bit.ly/4euVaWE
What's a DPU?
While CPUs handle apps, DPUs tackle networking, storage and security. The result is efficiency and infrastructure that scales.
Learn how CPUs, GPUs, DPUs and AI NICs work together: bit.ly/4eqI3av
Your GPUs are only as fast as the data feeding them.
As AI clusters scale, networking becomes critical to performance. AI NICs are built to keep data moving efficiently, helping maximize throughput and minimize bottlenecks ↓
Scaling AI is not just about adding GPUs.
AMD Pensando Pollara 400 AI NICs give Crusoe's AMD Instinct MI355X GPU instances the high-bandwidth networking needed to support demanding multi-node AI workloads.
Read more ⬇️
We were one of the first cloud providers to virtualize @AMD Instinct™ MI355X GPUs. Here's what it actually took:
❌ RCCL routing through the wrong interface
❌ GPU memory reg failing (ib_peermem is gone in kernel 6.8+)
❌ PCIe topology flattened by Cloud Hypervisor
✅ 368 GB/s all-reduce. Zero errors. Nice work @shubh_chakrawar and Andrew Carp.
Full debug walkthrough: bit.ly/4tQhorV
#GPUCompute #CloudInfrastructure #AMD #ROCm
As AI scales across hundreds of GPUs, communication becomes critical to performance. Enter AI networking.
In our new blog “What is AI Networking?”, we break down how it enables efficient, scalable AI infrastructure.
Learn more: amd.com/en/blogs/what-is-ai-…
As AI clusters scale, network congestion and tail latency limit efficiency.
Learn how the AMD Pensando Pollara 400 AI NIC overcomes these limitations with higher cluster uptime and better RCCL performance.
Read the product guide now: bit.ly/4cNPqIa
Networking is the backbone of scalable AI.
@Dell + AMD Pensando Pollara 400 AI NIC deliver a programmable Ethernet solution that accelerates AI workloads and adapts to evolving standards.
Explore the full ecosystem: bit.ly/3PjqUFk
In the modern data center, CPUs and GPUs have been getting all the attention. Discover how Data Processing Units (DPUs) are reshaping modern infrastructure in our blog: bit.ly/4qXj4OK
“Together, AMD and @Cisco can drive superior performance across frontend and backend environments—proving the power of an open Ethernet for AI.”
Cisco + AMD Pensando Pollara 400 AI NIC deliver intelligent, programmable solutions for complex AI tasks: bit.ly/4t4F4JS
NICs, such as the AMD Pensando Pollara 400 AI NIC, empower efficient scale-out fabrics with intelligent load balancing and congestion management.
Explore more on the role of NICs in AI infrastructure in @esg_global’s newest analyst paper: bit.ly/4bWOO2L
For 10+ years, AMD Solarflare™ solutions have set the standard for ultra-low latency networking.
The AMD Solarflare™ X4 Series pushes performance for HFT and beyond — when every sub micro-second matters: amd.com/en/products/ethernet…
AI workloads are massive and traditional Ethernet can’t keep up.
AI NICs enable faster, more reliable GPU-to-GPU communication, reducing latency and improving scalability for generative AI and LLM training. amd.com/en/blogs/2025/what-i…
Introducing AMD Pensando Pollara 400 AI NIC–Ready Server Platforms, built with leading OEM/ODM partners.
Together, we’re enabling high-performance AI networking for training & inference at scale. Learn more: amd.com/en/blogs/2025/pollar…
The AMD Pensando Pollara 400 AI NIC keeps data flowing efficiently by intelligently retransmitting only lost or corrupted packets to save time, help reduce bandwidth usage, and speed up job completion.
It is built for demanding AI and ML workloads: bit.ly/3LPXvko