NVIDIA
Tesla Referral V100 SXM2 16GB HBM2 GPU – Tested & Fully Functional
Condition: Used – Tested Working
Genuine NVIDIA Tesla Referral V100 SXM2 16GB GPU accelerator, pulled from my personal Linux AI inference server as part of an upgrade to a 32GB module.
This GPU was actively used for local AI/LLM inference workloads and was fully operational when removed. Before removal, I performed hardware diagnostics, ECC memory health checks, and sustained GPU load testing to verify its NVIDIA
Model: Tesla Referral V100 SXM2 16GB
GPU Architecture: Volta (GV100)
Memory: 16GB HBM2 (16,384 MiB)
Interface: SXM2 (not standard PCIe)
Board Part Number / MPN: 900-2G503-0410-100
GPU Part Number: 1DB1-895-A1
VBIOS Version: 88.00.13.00.02
ECC Memory: Supported and enabled
Diagnostic & Memory Health Results
Tested using NVIDIA drivers and CUDA under Ubuntu Linux.
All reported ECC and memory health checks were clean:
Single-bit ECC errors: 0
Double-bit ECC errors: 0
Aggregate ECC errors: 0
Retired memory pages: 0
Pending page blacklisting: None
PCIe link: Gen 3 x16, negotiated successfully
No GPU recovery action indicated
No thermal slowdown recorded in NVIDIA diagnostic counters
ECC diagnostics were collected both before and after load testing, with no new errors reported.
GPU Load Testing
The GPU was tested under a sustained CUDA-based AI inference workload using llama.cpp.
Recorded test results:
GPU utilization: 99–100% under sustained load
Typical power consumption: 210–220 watts
Maximum observed GPU temperature: 71°C
Maximum observed HBM2 temperature: 76°C
GPU clock speeds during sustained load: Approximately 1,380–1,440 MHz, with no thermal slowdown indicated by NVIDIA diagnostics.
GPU memory allocated during testing: Approximately 11.5 GB
The GPU operated under heavy computational load, with stable clock speeds and temperatures below the reported operating limits.
Diagnostic results available upon request.
GPU Power Limit: The GPU was configured with a 220W software power limit in my inference server for power efficiency. NVIDIA diagnostics confirm that the module supports an adjustable power limit of 100W–300W, with a factory default of 300W. The 220W setting was a configuration choice, not a hardware limitation. All load testing was performed with the 220W limit in place.
Additional Information
Removed from a working system solely to upgrade GPU memory capacity.
Used in a personal AI/LLM inference server.
NVIDIA hardware identification and ECC diagnostics available.
Actual item Compatibility Notice: This is an SXM2 GPU module, NOT a standard PCIe graphics card. It requires compatible SXM2 hardware, appropriate cooling, and system support. Please verify compatibility before purchasing.
The item is used and may show normal signs of handling or installation. Sold as the pictured GPU module; no other accessories are included unless explicitly shown.
Carefully packaged in antistatic protection for shipping.
