AMD MI355X narrows inference gap for DeepSeek V4 Pro

// 1h agoBENCHMARK RESULT

AMD MI355X narrows inference gap for DeepSeek V4 Pro

On September 15, 2026, SemiAnalysis published updated performance data for DeepSeek V4 Pro on InferenceX running on AMD's MI355X hardware. The data indicates a clear trend of the MI355X catching up in agentic inference scenarios, a gain attributed to significant software stack optimizations that are altering the competitive landscape.

// ANALYSIS

Hardware is only half the battle; software optimization is where the real gains are made in AI infrastructure. AMD's software improvements are successfully closing the performance gap in complex, agentic inference tasks. Improved efficiency for models like DeepSeek V4 Pro on the MI355X could disrupt the current AI hardware market dynamics. Better software support for AMD hardware will likely lead to more competitive pricing and choices for organizations running AI inference.

// TAGS

amd-mi355xinferencedeepseeksemianalysishardwareai

DISCOVERED

1h ago

2026-09-16

PUBLISHED

1h ago

2026-09-16

RELEVANCE

8/ 10

AUTHOR

StragglerLiu

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS1h ago

AI Field Notes highlights Claude cuts, Gemini 3.8

The 106th edition of AI Field Notes covers significant industry developments, notably Anthropic reducing Claude Code's weekly limit by 17% while framing it as a 25% increase, which prompted developers to quickly adopt cheaper alternatives. Additionally, the newsletter highlights Google's launch of Gemini 3.8 Live, featuring API voice models capable of seamlessly switching between 97 languages mid-call, along with an update from a Shanghai lab.

NEWS2h ago

Savante outlines 12-stage software supply chain

AI development studio Savante has published an operational breakdown of its 12-stage software supply chain, illustrating how client requests transition from chat discussions into tested production software. The blueprint delegates architecture, coding, testing, and staging to AI agents while maintaining human oversight for approvals, environment provisioning, and production releases.

UPDATE4h ago

Codex adds cross-thread referencing, agent orchestration

Developers can reference one Codex thread from another using @-mentions or drag-and-drop to monitor and message between sessions. Pairing inter-thread communication with schedules allows automated coordination across entire fleets of coding agents.