AMD RDNA 4 – AMD Radeon RX 9000 Series Graphics Cards

Original Article ↗ Hacker News Discussion ↗

VRAM, AI Workloads, and Product Positioning

  • Big argument over 16GB in 2025: many say it’s fine for a mid‑range gaming card at ~$600, others think it’s undersized for AI/LLM and longevity.
  • Some argue AMD is intentionally avoiding high‑VRAM consumer cards to protect datacenter products, mirroring Nvidia’s segmentation.
  • Counterpoint: AMD could differentiate by offering more VRAM to attract hobbyist AI users and researchers who can’t afford H100‑class hardware.

Pricing, Value, and Market Dynamics

  • Broad agreement that RX 9070 XT/9070 pricing is attractive versus Nvidia’s 50‑series MSRPs, especially given Nvidia’s effective street prices and low availability.
  • Skepticism that MSRPs matter when actual prices and stock are driven by scalpers and constrained supply.
  • Some feel the non‑XT 9070 is a classic “decoy tier” whose main role is to make the XT look better.

Linux Support and Driver Experiences

  • Long thread on AMD vs Nvidia drivers under Linux: experiences are heavily mixed.
  • Many say current Radeon cards “just work” on modern distros (especially with open drivers), and prefer AMD/Intel for openness, Wayland support, and long‑term maintenance.
  • Others report Nvidia being rock‑solid for decades when installed correctly, claiming the “bad Nvidia on Linux” narrative is exaggerated, especially on X11.
  • Wayland, laptop thermals, multi‑monitor VRR, and kernel/DRM integration are common pain points for Nvidia; meanwhile, AMD historically had serious issues too and improved a lot with Valve’s involvement.
  • Consensus: brand‑new AMD GPUs often need newer kernels/Mesa than stable distros ship, making early adoption painful.

Upscaling, Ray Tracing, and Gaming Features

  • Some hope RDNA 4 finally matches Nvidia’s hardware BVH raytracing rather than shader hacks.
  • Mixed views on FSR vs DLSS: one camp says AMD is far behind DLSS in quality, another notes FSR adoption is helped by consoles all using AMD.
  • Frame‑generation (“fake frames”) divides opinion: some love the FPS boost, others dislike added latency.

ROCm, CUDA, and AI Ecosystem

  • Lack of ROCm support at launch is heavily criticized, especially since AMD’s own slides push “AI performance.”
  • Comparison to Nvidia: every GeForce has usable CUDA on day one, which helped cement CUDA as the default AI stack.
  • Official ROCm support matrix for consumer RDNA is narrow; many cards work only unofficially or via distro patches.
  • Some users already have ROCm running on 9070 XT from source, but this is seen as inadequate versus plug‑and‑play CUDA.

Form Factor, VRAM Tiers, and Segmentation

  • Complaints that board partners aren’t offering compact 2‑slot designs despite the 9070’s power budget suiting small builds.
  • Strong demand for a 32GB consumer card; expectation is that such configurations will be reserved for expensive workstation SKUs (48GB/80GB) well above $1,000.
  • Several argue that even if AMD shipped a cheap 32GB card, market scarcity would quickly push prices up to parity with other high‑VRAM options.

Branding and Naming Confusion

  • Many find AMD’s product naming a mess compared to Nvidia’s relatively consistent series.
  • Confusion over skipped or reused number ranges (e.g., previous 8000‑series, mobile vs desktop, “AI Max+” branding) and partial realignment to Ryzen 9000.
  • Some welcome the 9070/9070 XT naming as closer to Nvidia’s scheme; others see it as late, half‑hearted, and likely to change again.

Overall Sentiment

  • Hardware itself is viewed as welcome and competitively priced, especially if AMD can ship real volume at MSRP.
  • Enthusiasm from Linux gamers and anti‑Nvidia users is tempered by frustration over ROCm, AI tooling, and branding chaos.
  • Many see RDNA 4 as a solid gaming option but still not the obvious choice for AI developers or those needing >16GB VRAM.