NVIDIA GPU Selector
Regurarly updated
Always updated with the latest NVIDIA accelerators and information.
All info on one page
One pager information with all main parameters and AI use-cases.
Benchmarks
// choose your best accelerator
GPU selector
Quick guide
This guide will help to choose your AI accelerator quickly.
Expert guide
Performance benchmarks and technical parameters for AI experts.
Older GPUs
// NVIDIA BLACKWELL - GRACE SUPERCHIP
The most powerful GPU - NVIDIA GB200
Up to 192 GB
GPU memory per card
Compute
AI training and inferencing,
data analytics, HPC
General Purpose
Visualization, rendering, AI,
virtual workstations
High-Density VDI
Virtual applications, virtual desktops,
virtual workstations
// Quick Guide
Choose your GPU use case
NVIDIA B200
AI Inference
AI Training
HPC
NVIDIA H200
AI Training
HPC
NVIDIA L40S
AI Inference
Rendering
Virtual Desktop
Virtual Workstation
NVIDIA L4
AI Inference
Rendering
Virtual Desktop
NVIDIA A16
Virtual Desktop
NVIDIA B200
AI Inference
AI Training
HPC
NVIDIA L40S
AI Inference
Rendering
Virtual Desktop
Virtual Workstation
NVIDIA L4
AI Inference
Rendering
Virtual Desktop
NVIDIA B200
AI Inference
AI Training
HPC
NVIDIA H200
AI Training
HPC
NVIDIA B200
AI Inference
AI Training
HPC
NVIDIA H200
AI Training
HPC
NVIDIA L40S
AI Inference
Rendering
Virtual Desktop
Virtual Workstation
NVIDIA L4
AI Inference
Rendering
Virtual Desktop
NVIDIA L40S
AI Inference
Rendering
Virtual Desktop
Virtual Workstation
NVIDIA L4
AI Inference
Rendering
Virtual Desktop
NVIDIA A16
Virtual Desktop
NVIDIA L40S
AI Inference
Rendering
Virtual Desktop
Virtual Workstation
Expert guide
// GPU selector
GPU | RTX PRO 4500 Blackwell SE | H200 SXM5 | H200 NVL | RTX PRO 6000 Blackwell SE | B200 | B300 |
|---|---|---|---|---|---|---|
Architecture | Blackwell | Hopper | Hopper | Blackwell | Blackwell | Blackwell |
Card chip | GB 203 | GH100 | GH100 | GB202 | B200 | B300 |
# CUDA cores | 10 496 | 16 896 | 16 896 | 24 064 | TBA | TBA |
# Tensor cores | 328 | 528 | 528 | 752 | TBA | TBA |
GPU memory | 32 GB | 141 GB | 141 GB | 96 GB | 192 GB | 288 GB |
Memory technology | GDDR7 | HBM3e | HBM3e | GDDR7 | HBM3e | HBM3e |
Memory throughput | 896 GB/s | 4.8 TB/s | 4.8 TB/s | 1.6 TB/s | 8 TB/s | 10 TB/s |
FP64 (TFlops) | — | 34 | 30 | — | TBA | TBA |
FP64 Tensor (TFlops) | — | 67 | 60 | — | 37 | 1.2 |
FP32 (TFlops) | 51 | 67 | 60 | 126 | 75 | 72 |
TF32 Tensor (TFlops) | 203 | 989* | 835* | 251 | 2 200* | 2 200* |
FP16 Tensor (TFlops) | 406 | 1 979* | 1 671* | 503.8 | 4 500* | 4 500* |
INT8 Tensor (TOPS) | 811 | 3 958* | 3 341* | 1 007.6 | 9 000* | 280* |
FP8 Tensor (TFlops) | 811 | 3 958* | 3 341* | 2 015.2* | 9 000* | 9 000* |
FP4 Tensor (TFlops) | 1 600 | — | — | 4 030.4* | 18 000* | 18 000* |
Multi-Instance GPU | 2 instances | 7 instances | 7 instances | 4 instances | TBA | TBA |
NVENC | NVDEC | JPEG engines | 3 | 3 | 0 | 7 | 7 | 0 | 7 | 7 | 4 | 4 | 4 | TBA | TBA |
GPU link | PCIe 5.0 | NVLink 4 | NVLink 4 | PCIe 5 | NVLink 5 | NVLink 5 |
Power consumption | 165 W | 700W | 600W | 600 W | 1 000W | 1 400W |
Form factor | PCIe gen5 1-slot FHFL | SXM5 card | PCIe gen5 2-slot FHFL | PCIe gen5 2-slot FHFL | SXM5 card | SXM5 card |
Announcement | 2026 | 2023 | 2023 | 2025 | 2024 | 2025 |
1) preliminary numbers
2) the total power consumption of CPU, GPU and memory on the superchip
Availability: good (on stock or 4-6 weeks), medium (around 10 weeks), bad (15 weeks+), not available
Solving the world’s most important scientific, industrial, and business challenges with AI and HPC. Visualizing complex content to create cutting-edge products, tell immersive stories, and reimagine cities of the future. Designed for the age of elastic computing, rises to all these challenges, providing unmatched acceleration at every scale.
Benchmarks
// GPU selector
NVIDIA B200 GPUs theoretical performance in DGX systems
NVIDIA A100 vs. NVIDIA L40s application benchmarks
NVIDIA A16, A100, V100, RTX4000 Ada by CTU FEE in Prague
PyTorch training time GPU comparison
MnasNET
ResNET
DesNET
