GPU Types (Brands, Memory, Speeds, Capacities,
Uses)
1. NVIDIA GPUs
• Families:
o GeForce (GTX / RTX) → Gaming & creators.
o RTX A-Series (ex-Quadro) → Workstations (CAD, 3D, design).
o Data Center (Tesla → A100, H100, B200) → AI, HPC.
• Memory Types: GDDR5, GDDR6, GDDR6X, GDDR7, HBM2/3.
• Clock Speeds: ~1.2 – 2.8 GHz.
• VRAM Capacities:
o GeForce: 4 – 24 GB (RTX 4090 = 24 GB GDDR6X; RTX 5090 = 32 GB
GDDR7).
o Workstation: 6 – 48 GB.
o Datacenter: 40 – 192 GB HBM3 (e.g., H100, B200 Blackwell).
• Where used: Gaming PCs, content creation, CAD, AI training, datacenters.
2. AMD GPUs
• Families:
o Radeon RX → Gaming.
o Radeon Pro → Workstations.
o Instinct (MI Series) → AI, HPC, supercomputers.
• Memory Types: GDDR5, GDDR6, HBM2, HBM3.
• Clock Speeds: ~1.5 – 3.0 GHz.
• VRAM Capacities:
o RX: 4 – 24 GB (RX 7900 XTX = 24 GB GDDR6).
o Pro: 8 – 32 GB.
o Instinct: 32 – 192 GB (MI300X = 192 GB HBM3).
• Where used: Gaming, CAD, servers, Frontier supercomputer.
3. Intel GPUs
• Families:
o Intel Arc → Gaming/consumer discrete.
o Intel Iris Xe → Integrated (ultrabooks, CPUs).
o Data Center GPUs (Ponte Vecchio, Rialto Bridge, Falcon Shores)
→ HPC/AI.
• Memory Types: GDDR6, HBM2e.
• Clock Speeds: ~1.5 – 2.5 GHz.
• VRAM Capacities:
o Arc: 6 – 16 GB GDDR6.
o Iris Xe: shared system RAM.
o Data center: up to 128 GB HBM2e.
• Where used: Gaming PCs, laptops, AI servers.
4. Apple GPUs
• Type: Integrated in Apple Silicon (M1, M2, M3).
• Memory: Unified LPDDR5/5X shared with CPU.
• Clock Speeds: ~1.5 – 2.5 GHz.
• Capacities: 8 – 128 GB unified memory.
• Where used: MacBooks, iPads, professional media workloads.
5. Other Mobile/Embedded GPUs
• ARM Mali, Qualcomm Adreno, Imagination PowerVR.
• Memory: Shared system RAM (LPDDR4/5).
• Speeds: 50 – 200 GB/s bandwidth.
• Where used: Smartphones, tablets, AR/VR, IoT.
GDDR7 (RTX 50-Series Update)
• Standard: JEDEC JESD239 (2024).
• Products: NVIDIA RTX 50-series (Blackwell, e.g., 5090, 5080, 5070 Ti).
• Speeds:
o Initial: 32 Gb/s per pin (~1.5 TB/s total bandwidth on 384-bit bus).
o Roadmap: 36 – 48 Gb/s in future.
• VRAM Capacities (RTX 50 examples):
o RTX 5090 → 32 GB GDDR7.
o RTX 5080 → 16 GB GDDR7.
o RTX 5070 → 12 GB GDDR7.
• Why important: ~60% higher bandwidth vs GDDR6X at same/lower power.
Perfect for AI-heavy gaming and ray tracing.
Summary:
• Gaming (2025): NVIDIA RTX 50-series (GDDR7) and AMD RX 7000
(GDDR6).
• Workstations: NVIDIA RTX A-series, AMD Radeon Pro.
• AI/HPC: NVIDIA H100/B200 (HBM3e), AMD MI300 (HBM3), Intel Data Center
GPUs.
• Laptops/Ultrabooks: Intel Iris Xe, Apple M-series, AMD mobile GPUs.
Mobile: ARM Mali, Adreno, PowerVR.
Comparison Table (2025 GPUs)
Brand GPU Family VRAM Memory Type Bandwidth (approx) Use Case
NVIDIA GeForce RTX 40 8–24 GB GDDR6X 500–1000 GB/s Gaming, creators
NVIDIA GeForce RTX 50 12–32 GB GDDR7 672 GB/s – 1.5 TB/s Next-gen gaming, AI
NVIDIA RTX A / Quadro 6–48 GB GDDR6 300–1000 GB/s CAD, workstations
NVIDIA H100 / B200 80–192 GB HBM3e 2–5 TB/s AI, datacenters
AMD Radeon RX 7000 8–24 GB GDDR6 400–1000 GB/s Gaming PCs
AMD Radeon Pro 8–32 GB GDDR6 300–800 GB/s CAD, design
AMD Instinct MI300 128–192 GB HBM3 2–5 TB/s AI supercomputers
Intel Arc 6–16 GB GDDR6 ~300–600 GB/s Gaming (entry/mid)
Intel Iris Xe Shared LPDDR4/5 ~100 GB/s Laptops, ultrabooks
Intel Ponte Vecchio Up to 128 GB HBM2e 1–2 TB/s HPC/AI
Apple M3 GPU 8–128 GB unified LPDDR5/5X ~200–800 GB/s Macs, iPads
Mobile (Adreno/Mali) Shared LPDDR4/5 50–200 GB/s Smartphones, VR
Comprehensive NVIDIA GeForce RTX Overview — All Generations
RTX 2000 Series (Turing, ~2018–2019)
Model VRAM Memory Type Approx. Bandwidth Notes / Use Case
RTX 2060 6 GB GDDR6 ~336 GB/s Mid-range gaming
RTX 2070 8 GB GDDR6 ~448 GB/s Upper mid-range gaming
RTX 2080 8 GB GDDR6 ~448 GB/s High-end 1440p / 4K gaming
RTX 2080 Ti 11 GB GDDR6 ~616 GB/s Enthusiast-class gaming
RTX 3000 Series (Ampere, ~2020–2022)
Model VRAM Memory Type Approx. Bandwidth Notes / Use Case
RTX 3060 12 GB GDDR6 ~360 GB/s Mid-range gaming
RTX 3070 8 GB GDDR6 ~448 GB/s 1440p / 4K gaming
RTX 3080 10–12 GB GDDR6X ~760–912 GB/s High-end 4K gaming
RTX 3090 24 GB GDDR6X ~936 GB/s 8K gaming / content creation
RTX 3090 Ti 24 GB GDDR6X ~1008 GB/s Top-tier performance
RTX 4000 Series (Ada Lovelace, ~2022–2024)
Model VRAM Memory Type Approx. Bandwidth Notes / Use Case
RTX 4070 12 GB GDDR6X ~672 GB/s 1440p / 4K efficient gaming
RTX 4080 16 GB GDDR6X ~716 GB/s High-end 4K gaming
RTX 4090 24 GB GDDR6X ~1008 GB/s Flagship: 4K/8K and AI workloads
RTX 5000 Series (Blackwell, 2025)
Model VRAM Memory Type Approx. Bandwidth Notes / Use Case
RTX 5070 12 GB GDDR7 ~672 GB/s Next-gen gaming & AI features like DLSS 4
RTX 5080 16 GB GDDR7 ~896 GB/s (est) High-end gaming & frame gen tasks
RTX 5090 32 GB GDDR7 ~1.536 TB/s Ulti-high bandwidth for AI and 8K gaming
Complete NVIDIA GeForce Comparison
VRAM Approx.
Series / Generation Models Included Memory Type Typical Use Case
Range Bandwidth
2060, 2070, 2080, Entry to high-end gaming (2018
RTX 2000 (Turing) 6–11 GB GDDR6 ~336 – 616 GB/s
2080 Ti era)
VRAM Approx.
Series / Generation Models Included Memory Type Typical Use Case
Range Bandwidth
GDDR6 / 1440p → 8K gaming, content
RTX 3000 (Ampere) 3060 → 3090 Ti 8–24 GB ~360 – 1008 GB/s
GDDR6X creation
RTX 4000 (Ada High-end modern gaming, AI-
4070, 4080, 4090 12–24 GB GDDR6X ~672 – 1008 GB/s
Lovelace) enhanced tasks
~672 GB/s → ~1.5 Next-gen gaming, AI frame gen,
RTX 5000 (Blackwell) 5070, 5080, 5090 12–32 GB GDDR7
TB/s 8K
Integrated Overview with All GPU Brands
Brand GPU Family / Series VRAM (GB) Memory Type Bandwidth Use Case Context
GeForce RTX 2000– GDDR6 / GDDR6X / ~336 GB/s → 1.5 Gaming: entry to next-gen, AI
NVIDIA 6–32 GB
5000 GDDR7 TB/s features
RTX A (Workstation) 6–48 GB GDDR6 ~300–1000 GB/s CAD, 3D modeling
H100 / B200 (Data
80–192 GB HBM3e ~2–5 TB/s AI, HPC, datacenters
Center)
AMD Radeon RX 7000 8–24 GB GDDR6 ~400–1000 GB/s Modern gaming
Radeon Pro
8–32 GB GDDR6 ~300–800 GB/s Professional visual workloads
(Workstation)
Brand GPU Family / Series VRAM (GB) Memory Type Bandwidth Use Case Context
AI, supercomputing (e.g.
Instinct MI (HPC/AI) 128–192 GB HBM3 ~2–5 TB/s
Frontier)
Intel Arc (Gaming) 6–16 GB GDDR6 ~300–600 GB/s Entry/mid-tier gaming PCs
Iris Xe (Integrated) Shared RAM LPDDR4/5 ~100 GB/s Laptops, office ultrabooks
Ponte Vecchio (Data High-performance AI/HPC
Up to 128 GB HBM2e ~1–2 TB/s
Center) servers
M-series Integrated Shared (8–128 GB
Apple LPDDR5/5X ~200–800 GB/s Macs/iPads, media production
GPU unified)
Mobile Adreno, Mali, PowerVR Shared system RAM LPDDR4/5 ~50–200 GB/s Smartphones, tablets, AR/VR
