Do you sell SXM GPUs as standalone cards?

No — SXM and OAM modules are baseboard-bound for HGX / UBB platforms. For standard servers we quote PCIe variants (H100 PCIe, L40S, MI210) instead.

How are used data-center GPUs tested?

Per batch we record GPU hours, ECC error history and thermals; units are stress-tested on OEM platforms before dispatch and the report ships with your order.

Can you supply complete 8-GPU HGX nodes?

We kit complete HGX baseboards with CPUs, memory and NICs as a single BOM — send the node specification for a consolidated quote.

Product line 06 / 06 — Live stock list

Graphics Card

Wholesale data-center GPUs — NVIDIA Blackwell and Hopper (B200, H200, H100), Ada Lovelace inference cards (L40S, L40, L4) and AMD Instinct MI300-series accelerators, new and tested pulls, with platform compatibility guidance.

30SKUs in stock list

12New condition

18Pulled & tested

24hQuote turnaround

PDF

Graphics Card Stock List — SEP 2026 (PDF)Full part-number list with condition & remarks · updated monthly

Download PDF

Featured models

In-demand part numbers

A selection from the current stock list — filter by model or paste a part number. Can't find yours? Download the full list or send a BOM; we source hard-to-find parts.

AI Training — New

Blackwell & Hopper · SXM / PCIe · HBM3e

2 featured −

Model	VRAM	TDP	Form	Part No.	Cond
NVIDIA B200192GB SXM Blackwell · Flagship training	HBM3e 192GB	~1000W	GB100 SXM	On board label	NEW	Quote →
NVIDIA H200141GB SXM5 Hopper · HGX H200	HBM3e 141GB	700W	GH100 SXM5	900-21010-0200-000	NEW	Quote →

AI / HPC — Refurbished

H100 & A100 pulls · HGX and PCIe variants · tested

2 featured +

Model	VRAM	TDP	Form	Part No.	Cond
NVIDIA H10080GB PCIe Hopper GH100 · OEM tray avail.	HBM3 80GB	350W	PCIe Gen5	900-21010-0900-000	USED	Quote →
NVIDIA A10080GB PCIe Ampere GA100	HBM2e 80GB	300W	PCIe Gen4	900-21001-0014-000	USED	Quote →

Inference & VDI — New

Ada Lovelace · L40S / L4 / A10 · MIG & vGPU

3 featured +

Model	VRAM	TDP	Form	Part No.	Cond
NVIDIA L40S48GB Ada Lovelace AD102 · Inference flagship	GDDR6 48GB	350W	PCIe Gen4	900-2G133-0080-000	NEW	Quote →
NVIDIA L424GB Ada Lovelace AD104 · Low profile · 72W	GDDR6 24GB	72W	PCIe Gen4	900-2G193-0000-000	NEW	Quote →
NVIDIA A1024GB Ampere GA102 · vGPU / VDI	GDDR6 24GB	150W	PCIe Gen4	900-2G133-0020-000	NEW	Quote →

AMD Instinct — New

CDNA 3 accelerators · ROCm stack · OAM

1 featured +

Model	VRAM	TDP	Form	Part No.	Cond
AMD InstinctMI300X 192GB CDNA · Flagship OAM	HBM3 192GB	750W	3 OAM	On board label	NEW	Quote →

Before you order

SXM / OAM modules are baseboard-bound (HGX / UBB) and are not standalone PCIe cards — verify tray and baseboard compatibility before quoting.
Used units: GPU hours, ECC error counts and thermal service history are recorded per batch and disclosed on request.
Multiple board variants exist per model (CEC / non-CEC, OEM-specific) — confirm the exact 900-number on the board label. Quotes valid ≤ 7 days.

FAQ

Graphics Card — answers before you ask

The questions procurement teams ask first. Anything else — WhatsApp is fastest.

—— Request for quotation

Send the BOM.
We quote in 24h.

Availability, pricing and lead time from a named engineer — inside one business day.

Request a Quote All product lines