Across a single recent stretch of coverage, three unconnected stories point at one thing: the consumer GPU market is being reshaped from the bottom up by owners, not vendors. A budget salvage rig, a community bug tracker, and a modding effort that ports ringfenced DLSS features to older cards are all examples of the same behavior. Users are treating GPUs as durable, reconfigurable assets rather than disposable, vendor-capped products.
The Salvage Economy of VRAM
Tom's Hardware reported that a Reddit user paid $35 for a nine-GPU Nvidia P106-100 mining rig left in a garage, yielding 54GB of VRAM at roughly 65 cents per GB, and then ran Qwen3.6-35B-A3B on it locally. The numbers matter less than the signal. A retired crypto miner, hardware Nvidia never intended as a general-purpose compute platform, was repurposed into an inference box for a 35B-parameter model class.
That is a direct consequence of how the mainstream market prices VRAM. Consumer cards with large memory buffers carry a premium tied to AI demand, and the used channel now offers a parallel supply. The P106-100 is not a modern part, and its owners accept real tradeoffs in drivers, connectivity and efficiency. But at 65 cents per GB, the economics of local inference change. For US consumers weighing whether to run a model on their own desk or pay per token to a cloud provider, the salvage channel is now a genuine third option.
The Community Layer Around Drivers
Tom's Hardware also reported that a coder using the handle independent-arg built NvidiaWatch, a site that monitors hundreds of GPU driver bugs. This is the second example of the same pattern: the software experience around a GPU is no longer solely defined by the vendor's release notes.
Driver bugs degrade performance, break features and sometimes take months to resolve. A tracker that aggregates hundreds of them at once gives buyers and owners a view Nvidia does not publish in a single place. It is a labor of love, as the report describes it, and that framing is important. Nobody is paying for this. It exists because owners want information the vendor does not organize for them.
For US buyers, this functions as an informal reliability index. A driver branch with a long tail of open issues is a different purchase than one without, and that calculus now lives outside official channels. For Nvidia, it is a reputational mechanism it does not control.
Modders Move the Feature Line
Tom's Hardware also reported that DLSS Multi-Frame Generation and DLSS 5 Neural Rendering can now run on hardware never intended for them, at least according to Nvidia, with two of the modders foundational to these community advancements speaking out. This is the most consequential of the three, because it touches the vendor's core upsell logic.
Features like frame generation and neural rendering have been used to segment generations. Porting them to older cards undermines the assumption that a new GPU is required to access them. The report is careful to note the "according to Nvidia" qualifier, which is appropriate: running a feature on unsupported hardware does not guarantee it performs well or matches the vendor's validated experience. But the modders are not claiming parity. They are claiming access.
That distinction is what makes this a pattern rather than a stunt. The same instinct that gets a $35 mining rig running a 35B model also gets DLSS features onto cards that were never certified for them. Owners are optimizing for capability per dollar, not for the experience a vendor packaged for them.
