GPU Resource
Configuration Guide
The definitive reference connecting NVIDIA GPU specifications to verified OEM part numbers — built for ITAD operators, secondary market buyers, and IT procurement teams working with A100 and H100 hardware.
NVIDIA GPU specifications + verified OEM part numbers, in one place
Sourcing, valuing, or deploying A100 and H100 GPU hardware is complicated by a fundamental problem: NVIDIA publishes specification data, but OEM vendors ship that hardware under their own part numbers, inside their own server platforms, with their own configuration constraints. A Dell PowerEdge R760xa and an HPE Apollo XL675d can both carry an NVIDIA A100 — but they ship under completely different part numbers, with different riser configurations, different NVLink constraints, and different teardown hierarchies that determine whether the GPU has standalone value at all.
This guide reconciles those two worlds. It combines full NVIDIA GPU specifications from A100 through Blackwell with verified OEM system data across four major platforms:
NVIDIA GPU Specifications
- Ampere A100 — 40GB and 80GB, PCIe and SXM4
- Hopper H100 — 80GB PCIe Gen5, H100 NVL 94GB, SXM5
- Hopper H200 — SXM5 and NVL variants
- Blackwell — B100 and B200 SXM reference specs
- FP8, FP16, BF16, FP64 Tensor Core TFLOPS
- Memory bandwidth, TDP, NVLink generation
- MIG instance count and Transformer Engine support
- Side-by-side comparison across up to four GPUs
OEM Systems & Part Numbers
- Card-level PNs — primary and all known alternates
- System-level tied SKUs — ordering codes for quotes and invoices
- Assembly-level PNs — HGX configurations with component breakdowns
- Compatible server platforms with max GPU count
- NVLink bridge and accessory PNs per platform
- PSU minimums and cooling class (air vs. liquid)
- Non-standalone GPU flags for SXM4/SXM5 systems
- EOL and end-of-sale status by platform generation
Start from the GPU, the server, or the part number
Switch between views using the tabs at the top of the guide.
GPU Specs & OEM Systems
Browse the full GPU family, filter by generation, form factor, or OEM. Expand any card for complete NVIDIA specs alongside every OEM platform that supports it — with part numbers, GPU counts, NVLink constraints, and accessory PNs. Compare up to four GPUs side by side.
Server-First View
Pick an OEM and server model, see every GPU it supports with the complete part number picture. Built for ITAD intake: when you receive a system, you start with what you have in front of you.
PN Reverse Lookup
Paste any Dell, HPE, or Lenovo part number from an asset manifest, service tag printout, or quote sheet. Returns GPU type, memory, form factor, compatible platforms, and sourcing notes.
GPU Resource Configuration Guide — OEM Edition
NVIDIA specs · Dell · HPE · Lenovo · NVIDIA DGX · PN Reverse LookupCoverage not available in any single public source
Several areas documented here represent the data gaps most likely to cause errors in ITAD intake, secondary market valuation, and procurement. These are documented from primary OEM sources.
HGX A100 SXM4 disassembly hierarchy — Dell
The full three-tier PN structure for Dell HGX A100 4-GPU assemblies: assembly-level PNs (700T5, 6CNP5, R71T2) disassemble into baseboard FRU PNs (N4GGW 40GB, K83FW 80GB) plus individual GPU module PNs (N14YF, V40HX, GDNF4, 8393M) distinguished by die revision — PG506 vs PG510 — and SKU generation.
SXM4 GPU module vs. PCIe card — PN disambiguation
Several Dell part numbers appear in both PCIe card and SXM4 GPU module contexts. N14YF, V40HX, and GDNF4 carry explicit form factor verification flags — the same PN on a manifest could indicate a standalone PCIe card or a GPU module extracted from an HGX baseboard.
HPE SKU suffix convention — -A vs. -C
HPE ships the same GPU under different PNs for different platform families. R9S41C (H100 80GB PCIe) is for ProLiant DL and Synergy; R9S41A is for Apollo and Specialized compute. Ordering the wrong suffix for a platform is a common procurement error documented throughout the guide.
NVLink configuration constraints — by platform
NVLink support on a platform does not mean full multi-GPU NVLink fabric. Dell R760xa with 4× GPUs creates two independent NVLink pairs — not a four-GPU pool. H100 NVLink is not supported on Lenovo SR670 V2 despite H100 PCIe compatibility.
Data confidence notice: GPU specifications are sourced from NVIDIA public datasheets, and OEM primary vendor documentation. Blackwell generation data carries a lower confidence rating. For procurement decisions, always verify against current official vendor documentation.
Data sourcing
All OEM part number and configuration data is compiled from primary vendor technical documentation.
Data confidence is indicated throughout with (primary source confirmed) and (verify before use).
Need custom reports or market intelligence?
Looking for custom GPU pricing intelligence, fleet valuation, OEM configuration verification, or connections to verified buyers and sellers in the secondary market? Reach out to the GPU Resource team.