Used / secondary market: NVIDIA Tesla P40 24GB (mining-era / server pull)
various server liquidators and refurbished parts resellers on Newegg · US / global (Newegg refurbished/marketplace, USD listings)
Fields
id
supply/used-nvidia-tesla-p40-marketKind
used-market
Vendor
various server liquidators and refurbished parts resellers on Newegg
Region
US / global (Newegg refurbished/marketplace, USD listings)
Channels
marketplace
Price (USD)
397.99
Price basis
USED / Server pull / Refurbished bare passive PCIe accelerator card, no fan/shroud included unless stated. Observed 2026-10-11 on Newegg marketplace. $397.99 is the low-end quote across multiple liquidators (HP 870919-001 OEM and retail TCSP40M-24GB-PB at $397.99, other sellers at $413.99-$425.99). PERISHABLE-FIGURE STAMP: OBSERVED 2026-10-11. RE-VERIFY BY: 2027-01-11.
Availability
in-stock
Lead time source
structural-na
Lead time basis
lead_time_weeks is null because inventory is in-stock at surplus/refurb liquidators.
Export controlled
no
Export control regime
EAR 3A001 or EAR99 (legacy Pascal architecture below 3A090.a thresholds)
Export control note
Legacy Pascal enterprise accelerator below 3A090.a thresholds. General commercial export rules apply.
notes
OBSERVATION DATE 2026-10-11. Source sup-used-newegg-tesla-p40. The Tesla P40 (24GB GDDR5, Pascal GP102) remains a classic budget homelab LLM card due to its native INT8 DP4A support (47 TOPS INT8) and 24GB VRAM for under $400 ($16.58 per GB of VRAM). Key gotchas: Pascal lacks FP16 tensor cores (FP16 runs at 1/64 FP32 speed, so FP16 inference is unusable; GGUF/llama.cpp INT8/Q4_0 DP4A or Q8 quantization is required); passive cooling requires 3D-printed fan shrouds and high-static-pressure fans for desktop workstations; requires 'Above 4G Decoding' and motherboard PCIe BAR resizing. The accelerator record nvidia-tesla-p40 now exists and is wired into accelerator_ids. Note: eBay queries returned HTTP 403 Forbidden on 2026-10-11.
Notes
OBSERVATION DATE 2026-10-11. Source sup-used-newegg-tesla-p40. The Tesla P40 (24GB GDDR5, Pascal GP102) remains a classic budget homelab LLM card due to its native INT8 DP4A support (47 TOPS INT8) and 24GB VRAM for under $400 ($16.58 per GB of VRAM). Key gotchas: Pascal lacks FP16 tensor cores (FP16 runs at 1/64 FP32 speed, so FP16 inference is unusable; GGUF/llama.cpp INT8/Q4_0 DP4A or Q8 quantization is required); passive cooling requires 3D-printed fan shrouds and high-static-pressure fans for desktop workstations; requires 'Above 4G Decoding' and motherboard PCIe BAR resizing. The accelerator record nvidia-tesla-p40 now exists and is wired into accelerator_ids. Note: eBay queries returned HTTP 403 Forbidden on 2026-10-11.