Nvidia GeForce RTX 2080 Ti with 22GB Memory Appears on eBay for USD500
An eBay listing for a modified Nvidia GeForce RTX 2080 Ti upgraded to 22GB of VRAM for around $500 has sparked interest as a budget-friendly way to run larger AI models and extend the life of older hardware. Commenters weigh this against buying used RTX 3090s, and use the example to highlight Nvidia’s tight control over hardware configurations, aggressive market segmentation, and the difficulty of choosing GPUs for machine learning amid opaque product lines. The thread also branches into broader concerns about soldered memory, future CPU/GPU designs, and how crypto and AI demand have driven gaming GPU prices far beyond what many players can justify.
VRAM, sockets, and future memory architectures
- Pluggable GPU VRAM is viewed as technically unrealistic due to tight signaling/timing; trend is toward soldered/on‑package memory, not modules.
- Similar shift predicted for CPUs: consumer parts with RAM-on-package and no external memory interface; workstations/servers likely to use hybrids (on‑package HBM + external DDR).
- Some expect this to enable price segmentation: buyers must pay upfront for max memory, with no upgrade path.
- Debate over socketed CPUs: some argue almost no one upgrades CPUs; others cite long-lived AM4 as evidence many do.
- Chiplets and large caches already blur the line between “RAM” and “cache”; extra on‑package memory may just appear as bigger cache levels.
2080 Ti 22GB mods and technical doubts
- Modded 2080 Ti cards with 22GB (and even 44GB) VRAM have existed for a while, mostly from Chinese workshops; current “news” is US resale on eBay.
- Unclear how seamlessly VRAM doubling works and whether VBIOS/driver modifications are needed.
- Some fear Nvidia might push driver updates that disable or brick such cards; others think the PR/legal risk is too high.
- Suggestions include freezing driver versions, but this is seen as fragile given fast-moving ML software.
GPU market segmentation, VRAM scarcity, and buying pain
- Many complain Nvidia withholds high VRAM from “affordable” cards to force ML users into expensive professional SKUs.
- Historically, board partners could vary VRAM and layouts; now Nvidia tightly controls designs, limiting such differentiation.
- People report confusion navigating Nvidia’s fragmented product lines and marketing, plus opaque availability and pricing.
- Used high-end consumer cards (e.g., 3090) are often preferred over data-center GPUs on price/performance and VRAM.
- Lack of midrange, high‑VRAM options pushes tinkerers/students toward mods like the 22GB 2080 Ti.
Gaming vs AI/crypto and GPU pricing
- Some argue the “dream of game graphics” has been derailed by crypto and AI, with gaming no longer the primary driver of high-end GPUs.
- Others say graphics have been “good enough” for years; modern cards already exceed what most games need, except at the ultra‑high end.
- There’s frustration over midrange and enthusiast GPU prices rising dramatically and games increasingly targeting 3090/4090-class hardware.
Industrial policy and geopolitics
- One camp wants an “Airbus for GPUs” in Europe to counter Nvidia and build strategic autonomy, though skepticism exists about lagging performance.
- US export bans on high-end Nvidia AI GPUs to China are criticized as shortsighted and harmful to global cooperation.