5s of video, generated in 1.3s!
Nunchux brings @MiniMax’s MiniMax-H3 to @AMD MI355X with up to 26.7× faster inference than SGLang on 8 GPUs @AMDServer
And with streaming generation, you can change the prompt as the video plays, steering what happens next.
One stack, optimized across hardware.
Free access to MiniMax-H3 is coming soon! Join the waitlist now at nunchux.ai.
Our blog: nunchux.ai/blog/video-genera…#AI#VideoGeneration#MiniMaxH3
AI infrastructure will change. The question is how much has to change with it.
Compute, networking, servers and other layers evolve on different timelines. Designing those layers to evolve independently can give organizations greater flexibility to adapt without rebuilding everything around them.
Explore how infrastructure designed for flexibility can evolve one layer at a time: bit.ly/4xtDkek
Agentic AI brings multiple CPU roles into one evolving workflow. 6th Gen AMD EPYC helps customers optimize for each task, with flexibility for what comes next.
See how "Venice" supports the diverse performance demands of agentic AI, from latency-sensitive work to highly concurrent workloads: bit.ly/4jbuxdo
Windows Server 2025 delivers modern security, but many features depend on the hardware beneath it.
AMD EPYC 4005 Server CPUs help small businesses and remote offices/branch offices get the most from their upgrade, with security from silicon to OS. Read the blog to see why. bit.ly/4gwgu0l
ALT Digital poster featuring floating code and circuit diagrams. Text reads: "Modern Security from Silicon to OS: Why SMBs Should Move Beyond Windows Server 2016, Read the News." AMD logo at the bottom.
The biggest AI infrastructure challenge is not predicting what comes next but being ready to adapt as workloads evolve.
AMD Executive Vice President & GM of the Data Center Solutions Group Forrest Norrod shares why a broad compute portfolio and common software foundation can help organizations match workloads to the right compute today, then reshape their deployments as requirements change.
Learn how to build AI infrastructure ready to adapt to whatever comes next: bit.ly/4xtDkek
ALT A server rack filled with computing hardware labeled with AMD's logo. Text on the image reads: "Read the news. Building infrastructure for an AI world in motion."
Inference is moving fast, and AMD is providing the infrastructure needed to deliver at scale.
Setting a new MLPerf record with 5.75M tokens/sec at 512-GPU scale, our broadest submission yet provides another proof point for performance, scale, software maturity and reproducibility.
Read more about the results: bit.ly/4hdWnDr
AI infrastructure economics start with the cost of compute.
@Signal_65 testing hosted via @tensorwave found AMD Instinct MI355X offered a 39.6% GPU-hour cost advantage based on blended on-demand cloud pricing. bit.ly/45hAvl3
ALT Advertisement comparing token costs per dollar. MI355X/ GPU-H offers $4.71 versus B200/ GPU-HR at $7.80, showing a 39.6% lower cost for MI355X. Source: Signal65 research for AMD, July 2026.
Get more from every AI infrastructure dollar.
In @Signal_65 testing hosted via @tensorwave, AMD Instinct MI355X delivered more tokens per dollar across all three models tested, with up to 2.15x the price-performance. bit.ly/45hAvl3
ALT The image is an advertisement by AMD promoting increased AI inference efficiency. It states "Get more AI inference for every dollar" and highlights a performance of "up to 2.15X more tokens per dollar across all three models tested." The source is Signal65 research commissioned by AMD in July 2026. The background features abstract blue wave patterns.
AI inference is about more than peak throughput.
Independent @Signal_65 testing shows AMD Instinct MI355X led 2 of 3 models tested and delivered 1.42×–2.15× tok/$ across all 3, even on Kimi-K2.6, where the competition led median throughput. Testing was hosted via @tensorwave. Read more ↓ bit.ly/45hAvl3
ALT Promotional image featuring the text "AMD Instinct MI355X delivers competitive throughput with leading perf/$." The AMD logo and tagline "together we advance" are at the bottom. The background has dynamic blue wave patterns.
ALT The image is a promotional graphic for a financial analysis of GPT-OSS-120B. It highlights a 1.3 times higher throughput and 53% lower cost per document. The source is noted as Signal65, July 2026.
ALT A promotional graphic titled "Long Report Generation KIMI-K2.6" features bold text indicating "0.96X Competitive Throughput" and "1.42X More Tokens/$." The source cited is "Signal65, July 2026." The design includes dynamic blue and turquoise digital waves on a black background.
ALT The image is a promotional graphic for "Interactive RAG Chat QWEN3-NEXT-80B." It highlights two key features: "1.2x higher throughput" and "2x more users at SLA." The source is Signal65, dated July 2026. The design includes a black background with blue digital wave patterns.
Easy to integrate. Open by design.
Alex Carter, CEO and co-founder of XPerf AI, shares how AMD ROCm gives developers access to the tools they need, helping save development time.
Getting more from AI infrastructure starts with making better use of the hardware you already have.
Hear how XPerf AI works with AMD to boost hardware utilization, reliability, and the return on every GPU.
You can't optimize what you can't see.
As AI infrastructure grows more complex, IDC highlights observability and operational efficiency as critical to maintaining performance and resiliency.
Learn more: bit.ly/4quKrkB
ALT The image is a digital graphic featuring the text "You can't optimize what you can't see. Visibility drives operational excellence." The AMD logo and slogan "together we advance_" are also displayed. The design includes dynamic, wave-like blue lines on a dark background.
MetaRoCE is more than a new transport protocol.
It demonstrates how programmable AI networking can help infrastructure evolve at the pace of AI. Developed by Meta and implemented on the AMD Pensando Pollara 400 AI NIC, with plans to be deployed on the AMD Pensando Vulcano 800 AI NIC, the protocol carries forward on a shared programmable foundation, without need to start over with each hardware generation.
Read more: amd.com/en/blogs/2026/metaro…
ALT The image contains text stating "Networking That Keeps Pace" and "A programmable approach to evolving AI requirements." It features the AMD logo with the tagline "together we advance_." The background has a digital network pattern with blue and orange lights.
The most expensive GPU is the one sitting idle.
IDC forecasts the AI scale-out networking market will reach $79B by 2030, reflecting networking's growing role in keeping AI infrastructure running efficiently at scale.
Read the IDC Spotlight: bit.ly/4quKrkB
ALT Graphic featuring modern server racks with glowing lights. The text reads: "KEEP AI RUNNING AT SCALE - Connected infrastructure enables AI performance." The AMD logo and tagline "together we advance_" are displayed.
AI is no longer just a compute problem.
IDC forecasts AI infrastructure spending will grow from $486B in 2025 to $1.38T by 2030, highlighting the growing importance of networking, software, and systems design. Learn more: bit.ly/4quKrkB
ALT The image features text over a digital-themed background. The prominent text reads, "AI IS NO LONGER JUST A COMPUTE PROBLEM." Below, it says, "The future of AI depends on connected systems." The bottom left includes the AMD logo and the phrase "together we advance.
The network has to evolve as fast as AI does. The AMD Pensando Vulcano 800 AI NIC is fully programmable, built for scale-out and scale-across, keeping every GPU productive at scale.
Read more: bit.ly/4zhajET
ALT A promotional image featuring the AMD Pensando Vulcano 800 AI NIC, with a digital wave design in the background. The AMD logo and tagline "together we advance_" are displayed at the bottom.
Agentic AI demands more from CPUs. AMD EPYC 9006 Series Server CPUs offer:
✅ Up to 512 threads for agent execution
✅ Up to 1.6 TB/s memory bandwidth and up to 5 GHz frequency to help keep GPUs utilized
✅ Leadership performance and performance per watt for general-purpose workloads
Learn more about AMD EPYC 9006 Series Server CPUs for agentic AI: bit.ly/4xrkp4d
ALT ALT Text: Image of an AMD EPYC processor on a black background with blue streaks. Text reads: "The Server CPU Portfolio for the Agentic AI Era. AMD EPYC 9006 Series Server CPUs." AMD logo with "together we advance_AI" is at the bottom.
What makes AMD EPYC 9006 Series Server CPUs the world’s best server CPU for cloud, enterprise, and AI?
✅ Leadership performance and performance per watt
✅ Four purpose-built CPU families
✅ Agent sandbox, AI host node, and general-purpose CPUs for agentic AI
See 5 reasons to choose AMD EPYC 9006 Series Server CPUs: bit.ly/4xrkp4d
ALT Graphic highlighting AMD's EPYC 9006 Series Server CPUs. The text reads, "World's Best Server CPUs for Cloud, Enterprise, and AI." The AMD logo and the phrase "together we advance_AI" are displayed at the bottom. The background features stylized digital lines.