Universal data center GPU
NVIDIA L40S
FP3291.6 TFLOPS
FP81 466 TFLOPS*
Power350 W
WorkloadsAI + graphics
Universal PCIe GPU for inference, rendering, Omniverse and mixed workloads.
PCIe GPU / AI / HPC / inference
NVIDIA data center GPU accelerators
A procurement page for NVIDIA server GPUs and accelerators: PCIe cards for standard servers, AI inference, HPC and graphics workloads.
Related sections
NVIDIA catalog navigation
For mixed requests such as GPU + ConnectX + cables, use the common RFQ and attach the BOM.
NVIDIA GeForce RTX graphics cards
B2B sourcing of NVIDIA GeForce RTX 40 and RTX 50 graphics cards: RTX 5090, 5080, 5070 Ti, 5070, 5060 Ti, 5060, 5050, RTX 4090, 4080 Super, 4070 Ti Super, 4070 Super, 4070, 4060 Ti and 4060. We verify SKU, vendor, dimensions, power and warranty route.
Open sectionNVIDIA RTX professional workstation GPUs
B2B sourcing of NVIDIA RTX PRO and RTX Ada workstation GPUs: RTX PRO 6000 Blackwell, RTX 6000 Ada, RTX 5000 Ada, RTX 4500 Ada and RTX 4000 Ada. We verify ECC, DisplayPort, form factor and certifications.
Open sectionNVIDIA Networking equipment
B2B sourcing of NVIDIA Networking products: Quantum InfiniBand, Spectrum Ethernet, ConnectX adapters, BlueField DPU and LinkX cables. We verify ports, speed, latency, firmware and datasheets.
Open sectionNVIDIA data center GPU accelerators
B2B sourcing of NVIDIA data center GPUs: L40S, L4, H100 NVL, H100, H200 and B200. We verify form factor, cooling, power, TFLOPS, server compatibility and datasheets.
Open sectionNVIDIA Jetson for robotics and Edge AI
B2B sourcing of NVIDIA Jetson modules and developer kits: Jetson AGX Thor, AGX Orin, Orin NX and Orin Nano. We verify TOPS, power, dimensions, camera interfaces, I/O and JetPack.
Open sectionCatalog
Product and specification cards
Cards do not replace datasheets; they help structure the initial procurement request and identify what must be verified.
Low-profile inference GPU
NVIDIA L4
Form factorlow-profile PCIe
Use casevideo + inference
Key checkserver compatibility
Documentdatasheet
Compact accelerator for inference, video analytics and edge/data-center servers.
Hopper
NVIDIA H100 / H100 NVL
Use caseLLM / HPC
Form factorPCIe / NVL depending on SKU
Key checkcooling
Verificationserver QVL
High-performance segment for training, inference and supported server platforms.
Hopper
NVIDIA H200
Use caseLLM memory-bound
Key checkplatform
Verificationdatasheet + QVL
Supplyproject-based supply
Server, power, cooling and platform compatibility must be verified before quoting.
Blackwell
NVIDIA B200 / GB200
Use caseAI factory
Formatplatform / system
Key checkrack design
Supplyproject-based
Blackwell is usually sourced as part of a system or rack-level configuration, not as a simple standalone card.
Blackwell Ultra
NVIDIA B300 / GB300
Memory288 GB HBM3e per GPU
FormatGB300 NVL72
Key checkliquid-cooled rack
Supplyproject-based
Blackwell Ultra for AI reasoning factories; quote requires rack, network, power and cooling checks.
Enterprise AI server
RTX PRO 6000 Server
Memory96 GB GDDR7 ECC
Powerserver SKU
Use caseAI + industrial
Key checkOEM server
Enterprise AI and industrial workload option checked against the OEM server BOM.
DGX system
NVIDIA DGX H100 / H200
GPU8x H100/H200
Form factorrack system
Key checkpower + cooling
Supplyproject-based
Integrated DGX platform for AI training and inference that requires data-center power and cooling readiness.
DGX system
NVIDIA DGX B200
GPU8x Blackwell
GPU memory1 440 GB total
Power~14.3 kW max
Form factor10 RU
Blackwell DGX platform for large AI projects where quoting starts with site readiness and infrastructure checks.
What we verify before pricing
- Data center GPU quotes cannot be based on the chip name alone: server model, QVL, slot, cooling, power budget, firmware, GPU count and destination region are required.
- For B200, H200 or H100 projects, we first collect the node specification and service requirements, then check channel availability.
What to send
Model, SKU, quantity, acceptable brands, delivery country, warranty, compatibility, timeline and document requirements. For networking and data center projects, attach the BOM early.
Specification sources
Parameters are checked against official NVIDIA pages and datasheets. For AIB/OEM models, final dimensions, clocks, weight and part number are confirmed with the vendor.
RFQ
Send an NVIDIA request
Attach a model list or BOM for a faster response. If you do not have an exact SKU, describe the use case: gaming, workstation, AI server, networking fabric or Jetson prototype.