Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

Original Article ↗ Hacker News Discussion ↗

Nvidia’s announcement of its RTX 5090 and 5080 GPUs is seen as a pivot further toward AI and upscaling tricks rather than big gains in raw raster performance, with the flagship 5090 offering 32GB of GDDR7, very high power draw, and a $2,000 price tag. Commenters argue that DLSS 4 and multi‑frame generation account for much of the advertised speedups, raising questions about latency, visual artifacts, and how “real” the performance gains are for competitive or 4K gaming. Many expect the 5090 to be snapped up for local AI and LLM workloads despite its relatively modest VRAM for that use, while gamers debate whether to stick with older high‑end cards, wait for mid‑range 50‑series models, or consider AMD/Intel alternatives amid concerns over cost, power, and future VRAM needs.

Pricing, Positioning, and Product Segmentation

  • RTX 5090 at ~$2,000 and 5080 at $999 are seen as cementing a split: 5070/5080 as “real gaming” cards and 5090 as a prosumer / entry‑level AI card.
  • Several argue the xx90 line has effectively replaced the old Titan series; others frame 5080 as the true “high‑end gamer” SKU.
  • Many expect severe availability issues and scalping, especially for the 5090, with comparisons to crypto-era shortages.
  • Some see the 5090 price as a “wealth tax” on enthusiasts; others note PCB complexity, die size, VRAM bus width and layers as genuine cost drivers.

Performance, DLSS4, and Frame Generation

  • Nvidia’s big “2×” claims are mostly tied to DLSS4 Multi‑Frame Generation and AI upscaling, not raw raster performance.
  • Multiple commenters estimate non‑DLSS raster gains at only ~10–30% vs 40‑series in early marketing graphs.
  • Strong skepticism about frame generation: perceived visual artifacts, “fake FPS,” and added latency, especially harmful for fast competitive games.
  • Others are enthusiastic, arguing that if 40 → 120 FPS “looks and feels good,” users won’t care how frames are produced.

VRAM, Memory Bandwidth, and AI Workloads

  • 32GB on 5090 is called “way too little” by people wanting to run larger local LLMs; many hoped for 48–64GB.
  • 16GB on the 5080 and 12GB on lower SKUs are widely viewed as stingy for expensive cards and future AAA titles.
  • Bandwidth is seen as crucial for token generation; several compare 5090 vs Apple Silicon, Ampere Altra, Epyc, and Nvidia’s new “Project Digits” 128GB AI desktop box.
  • Some argue Nvidia deliberately caps VRAM on gaming cards to push buyers to higher-margin pro/AI products.

Power, Thermals, and Form Factor

  • 5090’s 575W TDP is a major concern: heat, noise, breaker limits, and the need for huge PSUs.
  • Enthusiasts note you can heavily power‑limit high‑end cards with modest performance loss.
  • Excitement around the 5090 FE being nominally 2‑slot and “SFF‑ready,” tempered by doubts about cooling 575W in small cases.

Gaming Use Cases: 4K/8K, RT, and VR

  • Debate over whether 4K and ray tracing are “necessary”: some prioritize gameplay and dislike RT/TAA/DLSS artifacts; others care deeply about visual realism.
  • 4K adoption is still relatively low; some say that’s because of GPU cost, not desire.
  • VR and flight sims are called out as uniquely demanding; even 40‑series struggles at high refresh rates.

Market Dynamics and Alternatives

  • Many lament the death of the “$300–$400 high‑end” era and say consoles or used 30‑series/40‑series now offer better value.
  • AMD is perceived as having ceded the ultra‑high‑end to Nvidia and focusing on midrange; Intel is cautiously mentioned as a long‑term disruptor, especially in budget GPUs.