Executive Summary

  • Massive Hyperscale Win for AMD: Cloudflare’s transition to Gen 13 servers utilizing AMD EPYC 9005 “Turin” processors has yielded a 2x throughput increase and 50% better performance-per-Watt, demonstrating massive scalability and efficiency for AMD’s flagship CPU architecture.
  • NVIDIA DLSS 5 Backlash: NVIDIA’s aggressive push into “content-controlled generative AI” via DLSS 5 is facing severe pushback from the gaming community and developers alike, creating a strategic narrative opening for AMD’s FSR technology as a more artist-faithful alternative.
  • NVIDIA Fortifies AI Software Stack: NVIDIA announced OpenShell and NemoClaw, an open-source, sandbox runtime for autonomous AI agents. This signifies NVIDIA’s rapid expansion into secure, enterprise-grade AI deployment, an area where AMD’s ROCm ecosystem will need to maintain competitive parity.

🔲 AMD Hardware & Products

[2026-03-23] Cloudflare Details Their Upgrade To EPYC Turin For 2x Throughput, 50% Better Perf/Watt

Source: Phoronix

Key takeaway relevant to AMD:

  • This is a highly visible, globally recognized validation of AMD’s EPYC 9005 “Turin” architecture in top-tier enterprise environments. It proves that hyperscalers can achieve exponential performance and efficiency scaling even without relying on specialized 3D V-Cache variants.

Summary:

  • Cloudflare has deployed their “Gen 13” server architecture powered by AMD’s flagship EPYC 9965 (Turin) processors.
  • Combined with a Rust-based software rewrite (FL2), Cloudflare mitigated early latency bottlenecks to achieve monumental leaps in throughput and power efficiency over their previous Genoa-X deployment.

Details:

  • Hardware Leap: Transitioned from Gen 12 (EPYC Genoa-X) to Gen 13 utilizing the flagship EPYC 9005 series (EPYC 9965 SKU, non-3D V-Cache).
  • Performance Benchmarks: Achieved 2x the throughput, 50% better performance-per-Watt, and up to 60% higher rack throughput compared to previous generation servers.
  • Software Synergy: Initial hardware latency regressions were completely resolved by transitioning to “FL2”, a Rust-based rewrite of their core request handling layer.
  • Software Metrics: The FL2 architecture improved memory access patterns, resulting in up to 50% more requests per CPU and up to 70% lower latency.
  • Feature Utilization: Cloudflare specifically praised AMD’s Platform Quality of Service (PQOS) extensions, which allowed them fine-grained control over shared server resources like cache and memory bandwidth.
  • Platform Specs: The new servers fully utilize 100 GbE networking and PCIe 5.0 infrastructure.

🤼‍♂️ Market & Competitors

[2026-03-23] Nvidia CEO says he’s ‘empathetic’ to DLSS 5 concerns — Jensen Huang doubles down on defense while decrying ‘AI slop’

Source: Tom’s Hardware

Key takeaway relevant to AMD:

  • NVIDIA is facing a notable public relations crisis regarding DLSS 5 altering the core geometry and lighting of games without developer consent. This backlash offers AMD a prime opportunity to position FSR (specifically upcoming iterations like FSR 4.1 mentioned by the community) as the transparent, non-intrusive, and developer-friendly upscaling alternative.

Summary:

  • NVIDIA CEO Jensen Huang attempted to address the growing backlash surrounding DLSS 5, which critics have labeled as “AI slop” that overrides developer intent.
  • Despite Huang claiming the tech is “truthful to the geometry,” community members and reports indicate that DLSS 5 heavily alters character features, and critically, was allegedly implemented into demos (like Resident Evil 9) without the actual developers’ prior knowledge.

Details:

  • DLSS 5 Architecture: Huang insists DLSS 5 is not a general-purpose AI nor a simple post-processing filter, but rather “content-controlled generative AI” operating late in the rendering chain.
  • Future AI Prompts: Future updates to DLSS 5 may include text-prompt capabilities (e.g., prompting the game to look like a “toon shader”), applying localized generative styles on the fly.
  • Visual Alterations: Community analysis (via Wccftech) notes significant lighting, shading, and structural facial changes to characters (e.g., fuller/redder lips on Resident Evil’s Grace), conflicting with Huang’s “truthful to geometry” claim.
  • Developer Disconnect: A critical report highlighted that developers of Resident Evil 9 at Capcom had zero prior knowledge of DLSS 5 before NVIDIA used their game for the on-stage demonstration, contradicting NVIDIA’s claims of deep “integration with the artist.”
  • Community Sentiment: Forum users note severe skepticism regarding generative frames. Mention is made of flight simulators where “fake frames cause real problems” by blurring or misrepresenting crucial HUD gauges.

[2026-03-23] How Autonomous AI Agents Become Secure by Design With NVIDIA OpenShell

Source: NVIDIA Blog

Key takeaway relevant to AMD:

  • NVIDIA is preemptively solving the next big hurdle in AI: enterprise security for autonomous agents. To remain competitive in the enterprise software space, AMD will need to ensure ROCm and its broader software stack provide comparable, secure sandboxing runtime environments for localized AI agent workflows.

Summary:

  • NVIDIA has launched “OpenShell,” an open-source, secure-by-design runtime for autonomous AI agents (referred to as “claws”).
  • They also introduced “NemoClaw,” a reference stack allowing developers to easily spin up self-evolving AI assistants locked down by OpenShell’s security policies.

Details:

  • OpenShell Architecture: Functions like a “browser tab” model for AI. It physically separates the agent’s application-layer operations from infrastructure-layer policy enforcement via isolated sandboxes.
  • Security Mechanisms: Agents are strictly prevented from overriding system policies or leaking private credentials, even if the agent is compromised or prompted maliciously. Permissions are verified by the runtime before any action is executed.
  • NemoClaw Stack: An open-source reference stack enabling the deployment of “OpenClaw” always-on assistants using NVIDIA Nemotron models and the OpenShell runtime via a single command.
  • Hardware Compatibility: Designed to run universally across clouds, on-prem, and local hardware, explicitly supporting GeForce RTX PCs/laptops, RTX PRO workstations, DGX Station, and DGX Spark supercomputers.
  • Industry Integration: NVIDIA is not acting alone; they are building this runtime ecosystem in collaboration with major enterprise security partners including Cisco, CrowdStrike, Google Cloud, Microsoft Security, and TrendAI to ensure compliance with global standards.