Top AI Hardware & GPUs
GPUs and AI accelerators, with full specs, from the companies building them.
NVIDIA Blackwell flagship datacenter GPU for frontier model training
Rack-scale Grace Blackwell superchip for trillion-parameter models
Rack-scale Blackwell Ultra superchip with expanded memory
Flagship consumer GPU for gaming and local AI workloads
NVIDIA flagship Blackwell consumer GPU
PCIe-form-factor Hopper datacenter GPU
Hopper-architecture datacenter GPU for large-scale AI training
High-memory Hopper refresh for large-model inference
Successor to Blackwell built for agentic-scale AI
CDNA 2 datacenter accelerator used in Frontier supercomputer
AMD flagship datacenter accelerator for large-model AI training
High-memory CDNA 3 refresh for large-model inference
AMD's latest flagship AI training and inference accelerator
CDNA 5 flagship of the MI400 series, built for the Helios rack
Google's eighth-generation TPU, purpose-built for inference
Google's eighth-generation TPU, split off to focus purely on training
Intel flagship datacenter GPU, used in the Aurora supercomputer
Original 40GB Ampere datacenter GPU
Ampere-generation datacenter GPU, still widely deployed
Drop-in Blackwell successor to H100 for HGX systems
Flagship Ampere consumer GPU
APU combining CDNA 3 GPU and Zen 4 CPU cores
AMD flagship RDNA 3 consumer GPU
AMD flagship RDNA 4 consumer GPU
Google's inference-optimized seventh-generation TPU
Google custom AI accelerator for training large models on Cloud TPU pods
Fourfold performance-per-chip jump over TPU v5e
Flagship Ampere consumer GPU
High-end Ada Lovelace consumer GPU
High-end Ada Lovelace consumer GPU
Refreshed high-end Ada consumer GPU
High-end Blackwell consumer GPU
China-market Hopper GPU with reduced NVLink bandwidth
Ada Lovelace datacenter GPU for inference and graphics workloads
NVIDIA flagship Ada Lovelace workstation GPU
Flagship Blackwell workstation GPU with 96GB VRAM
Rack-scale inference accelerator built on the Hexagon NPU
AMD flagship RDNA 3 workstation GPU
AWS's second-generation custom training chip
Largest chip ever built, with 4 trillion transistors
Cost-efficient TPU generation for training and inference
Language Processing Unit purpose-built for low-latency LLM inference
Second-generation Habana Labs training accelerator
Intel third-generation AI training and inference accelerator
China-market Ampere GPU with reduced NVLink bandwidth
Flagship Turing consumer GPU
High-end Ampere consumer GPU
High-end Ampere consumer GPU
Refreshed midrange Ada consumer GPU
Upper-midrange Ada consumer GPU
Upper-mid Ada Lovelace consumer GPU
Midrange Blackwell consumer GPU
Upper-midrange Blackwell consumer GPU
Top-end mobile Blackwell GPU
Export-compliant Hopper GPU for the China market
Low-power Ada Lovelace inference GPU for cloud servers
High-end Ada Lovelace workstation GPU
Ampere-generation workstation GPU
Ubiquitous low-power inference GPU
First-generation Tensor Core datacenter GPU
Single-die PCIe CDNA 2 accelerator
Dual-die CDNA 2 accelerator, cut-down MI250X
High-end RDNA 3 workstation GPU
High-end RDNA 2 consumer GPU
Flagship RDNA 2 consumer GPU
Refreshed flagship RDNA 2 consumer GPU
Upper-mid RDNA 3 consumer GPU
Upper-high-end RDNA 3 consumer GPU
High-end RDNA 4 consumer GPU
Dual-die Apple Silicon flagship with up to 512GB unified memory
Apple's highest-end SoC GPU, popular for local LLM inference
Second-generation wafer-scale AI chip
Optically-reconfigurable TPU pod generation
China's leading domestic AI training accelerator
Intel flagship consumer GPU from its first Arc generation
Ampere GPU for mixed graphics and AI inference
AWS-favored Ampere GPU for graphics and inference
Ampere data center GPU for graphics and inference
Flagship Pascal-era consumer GPU
Refreshed midrange Turing consumer GPU
Refreshed high-end Turing consumer GPU
Mainstream Ampere consumer GPU
Popular midrange Ampere consumer GPU
Midrange Ampere consumer GPU
Midrange Ampere consumer GPU
High-end mobile Ampere GPU
Mainstream Ada Lovelace consumer GPU
Mainstream Ada Lovelace consumer GPU
Midrange Ada Lovelace consumer GPU
Midrange mobile Ada Lovelace GPU
High-end mobile Ada Lovelace GPU
Top-tier Ada Lovelace mobile GPU for laptops
Mainstream Blackwell consumer GPU
Entry-flagship Blackwell consumer GPU
High-end mobile Blackwell GPU
Flagship edge AI module for robotics and autonomous machines
Flagship Turing-era workstation GPU
Midrange Ada Lovelace workstation GPU
Upper-midrange Ada Lovelace workstation GPU
High-end Ampere workstation GPU
Prosumer Turing card with 24GB VRAM
PCIe-form-factor Volta datacenter GPU
First-generation CDNA datacenter accelerator
High-end RDNA 2 workstation GPU
Midrange RDNA 3 workstation GPU
First-generation RDNA flagship consumer GPU
Midrange RDNA 2 consumer GPU
High-end RDNA 2 consumer GPU
Mainstream RDNA 3 consumer GPU
Midrange RDNA 3 consumer GPU
Mainstream RDNA 4 consumer GPU
First consumer 7nm GPU, with 16GB HBM2
Dual-die Apple Silicon chip with up to 192GB unified memory
AWS's second-generation inference chip
AWS's first purpose-built training chip
Deterministic single-core language processing unit card
Midrange Alchemist consumer GPU
Entry Battlemage consumer GPU
Intel second-generation Battlemage consumer GPU
PCIe variant of Intel's Ponte Vecchio datacenter GPU
Meta's in-house recommendation and ranking inference chip
Microsoft's first custom AI accelerator for Azure
Mainstream Ampere inference and analytics GPU
One of the most widely used Pascal-era GPUs
Popular midrange Pascal-era GPU
Upper-midrange Pascal-era GPU
High-end Pascal-era consumer GPU
Budget Turing GPU without RT/Tensor cores
Mainstream Turing consumer GPU
Midrange Turing consumer GPU
Midrange Turing consumer GPU
Entry-level Ampere consumer GPU
Midrange mobile Ampere GPU
Mainstream mobile Ada Lovelace GPU
Compact edge AI module for makers and light robotics
High-end Turing-era workstation GPU
Compact entry Ada Lovelace workstation GPU
Compact entry Ampere workstation GPU
Midrange Ampere workstation GPU
Upper-midrange Ampere workstation GPU
First HBM2 datacenter GPU, pre-dates Tensor Cores
Prosumer Volta card, first consumer GPU with Tensor Cores
Reconfigurable dataflow chip for trillion-parameter inference
Vega-based datacenter accelerator, AMD's first 7nm GPU
Compact midrange RDNA 2 workstation GPU
Midrange first-generation RDNA GPU
Mainstream RDNA 2 consumer GPU
Mainstream RDNA 2 consumer GPU
Flagship Vega-generation consumer GPU with HBM2
AWS's first custom inference chip
Chinese chiplet-based general-purpose GPU for AI training
ASIC hardwired to run only transformer models
Liquid-cooled third-generation TPU
First wafer-on-wafer AI processor
Second-generation Intelligence Processing Unit
Huawei's first Ascend AI training chip
Value Alchemist consumer GPU
Media transcode and inference GPU for cloud workloads
Quad-GPU Ampere card for virtual desktops
Budget Pascal-era GPU that needs no power connector
Entry-level Turing GPU
Budget Turing GPU refresh
Flagship Maxwell-era consumer GPU
Mainstream mobile Ampere GPU
Entry mobile Ada Lovelace GPU
Midrange Pascal-era workstation GPU
Flagship Pascal-era workstation GPU
Pascal-era inference GPU, still popular for cheap local LLM VRAM
Prosumer flagship of the Pascal generation
Power-efficient datacenter inference card
High-memory inference card for large generative models
Tenstorrent's third-generation RISC-V AI accelerator
RISC-V-based scalable AI accelerator card
Alibaba's first in-house AI inference chip
AMD's first Instinct-branded datacenter accelerator
Entry-level RDNA 2 consumer GPU
Second-generation Baidu cloud AI chip
Half-size sibling of the BR100 for inference workloads
Dual-die AI training and inference card
Cambricon's flagship large-model training chip
In-memory-compute chiplet card for low-latency LLM inference
Power-efficient second-generation inference chip
Tiny USB and M.2 inference accelerator for embedded devices
First TPU generation to support training, not just inference
Low-power edge inference chip
Entry-level Alchemist consumer GPU
Low-power dual-GPU media and inference card
Habana Labs' first-generation training accelerator
Chinese full-function GPU for AI training and inference
Entry mobile Ampere GPU
Dual-GPU Kepler-era compute accelerator
Low-power Pascal-era inference GPU
Second-generation reconfigurable dataflow chip
Samsung's first in-house AI inference chip
Tenstorrent's first commercial AI inference card
FuriosaAI's first commercial inference chip
Google's first-generation inference-only TPU, never sold externally
Entry HTPC and media transcode GPU
Dual-GPU Maxwell-era virtualization card