In a bid to circumvent the escalating cost of high-VRAM consumer graphics cards, computing enthusiast Oscar Molnar successfully repurposed an enterprise-grade Nvidia Tesla V100 SXM2 GPU to double his system memory capacity for local large language model inference. By combining a secondhand Tesla V100 with 16GB of HBM2 memory and an existing RTX 4080 rig, Molnar achieved a 32GB total VRAM configuration for approximately $266, a fraction of mainstream retail pricing. Integrating the SXM2-form-factor card into a standard desktop motherboard required an aftermarket adapter, but the primary engineering challenge proved to be thermal management. The stock enterprise cooling solution operates at roughly 82 decibels, producing noise comparable to industrial machinery. Molnar resolved this through a straightforward PWM modification, rerouting the fan cables directly to a motherboard header. Operating the fan at just ten percent duty cycle maintains the GPU at under 50 degrees Celsius under full load, effectively neutralizing the acoustic output while preserving adequate thermal headroom. Software configuration leveraged NixOS paired with a legacy Nvidia driver stack compatible with both Volta and Ada architectures. Once integrated, the hybrid system demonstrated robust performance for generative AI workloads. Testing a 27-billion parameter language model yielded an output rate of 32 tokens per second, a pace characterized as fully adequate for interactive deployment and notably faster than several contemporary cloud-based inference APIs. This build highlights a pragmatic workaround amid the ongoing RAMpocalypse, proving that decommissioned datacenter hardware can be adapted for consumer AI applications. While native 32GB variants of the Tesla V100 exist, they command double the market price, making the dual-16GB configuration a compelling alternative for budget-conscious developers. The project underscores a growing trend of hardware optimization to democratize local large language model deployment, bypassing traditional consumer hardware bottlenecks through strategic enterprise component repurposing and driver-level bridging.