Here is a roundup of the latest NVIDIA GPUs.
Related entries:https://blog.qualiteg.com/nvidia-gpu-capability-level/
SM_120 (Blackwell)
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce RTX 5090 |
21,760 |
32GB GDDR7 |
2025 |
PCIe Gen5 |
2-slot (internal exhaust) |
- |
| GeForce RTX 5080 |
10,752 |
16GB GDDR7 |
2025 |
PCIe Gen5 |
2-slot (internal exhaust) |
- |
| GeForce RTX 5070 |
- |
12GB GDDR7 |
2025 |
PCIe Gen5 |
2-slot (internal exhaust) |
- |
| GeForce RTX 5060 |
- |
8GB/16GB GDDR7 |
2025 |
PCIe Gen5 |
2-slot (internal exhaust) |
- |
SM_100 (Blackwell)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| NVIDIA B100 |
16,896 |
Up to 192GB |
Announced in 2024 |
PCIe Gen6: 256GB/s |
Passive |
- |
| NVIDIA B200 |
33,792 (16,896 × 2) |
Up to 192GB |
Announced in 2024 |
PCIe Gen6: 256GB/s |
Passive |
- |
| NVIDIA GB200 Grace Blackwell Superchip |
67,584 (33,792 × 2) |
Up to 384GB |
Announced in 2024 |
- |
Passive |
- |
SM_90 (Hopper)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| NVIDIA H100 |
16,896 (SXM5) |
80GB HBM3 |
2022 |
PCIe 5.0 x16 |
Passive cooler |
SXM5: up to ~700W (PCIe: TDP ~350W) |
| NVIDIA H100 NVL |
- |
188GB HBM3 |
around 2023 |
PCIe Gen5 128GB/s |
Passive |
- |
| NVIDIA H200 |
- |
141GB HBM3e |
Announced in 2023 |
PCIe Gen5 128GB/s |
Passive |
- |
| NVIDIA GH200 Grace Hopper Superchip |
- |
96GB HBM3/144GB HBM3e |
around 2023 |
- |
Passive / liquid-cooled |
- |
SM_89 (Ada Lovelace)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| NVIDIA L4 |
7,424 |
24GB GDDR6 |
Early 2023 |
PCIe 4.0 x16 |
Passive |
TDP 72W |
| NVIDIA L40 |
18,176 |
48GB GDDR6 ECC |
Late 2022 |
PCIe 4.0 x16 |
Passive cooler |
300W (reportedly configurable up to 320W) |
| NVIDIA L40S |
18,176 |
48GB GDDR6 ECC |
around 2023 |
PCIe 4.0 x16 |
Passive |
Up to ~300-320W? |
| RTX 6000 Ada Generation |
18,176 |
48GB GDDR6 ECC |
Late 2022 |
PCIe 4.0 x16 |
Blower fan (rear exhaust) |
TDP 300W |
| RTX 5000 Ada |
12,800 |
32GB GDDR6 ECC |
2023-2024 |
PCIe 4.0 x16 |
Active (rear exhaust) |
TDP 250W |
| RTX 4500 Ada |
7,680 |
24GB GDDR6 ECC |
2023-2024 |
PCIe 4.0 x16 |
Active (rear exhaust) |
TDP 210W |
| RTX 4000 Ada |
6,144 |
20GB GDDR6 ECC |
2023-2024 |
PCIe 4.0 x16 |
Active (single slot) |
TDP 130W |
| RTX 2000 Ada |
2,816 |
16GB GDDR6 ECC |
2023-2024 |
PCIe 4.0 x16 |
Active |
TDP 70W |
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce RTX 4090 |
16,384 |
24GB GDDR6X |
Oct 2022 |
PCIe 4.0 x16 |
Dual axial fans (internal exhaust) |
TDP 450W |
| GeForce RTX 4080 |
9,728 |
16GB GDDR6X |
Nov 2022 |
PCIe 4.0 x16 |
Dual axial fans (internal exhaust) |
TDP 320W |
| GeForce RTX 4070 Ti |
7,680 |
12GB GDDR6X |
Jan 2023 |
PCIe 4.0 x16 |
Dual axial fans (internal exhaust) |
TDP 285W |
| GeForce RTX 4070 |
5,888 |
12GB GDDR6X |
Apr 2023 |
PCIe 4.0 x16 |
Dual axial fans (internal exhaust) |
TDP 200W |
| GeForce RTX 4060 Ti |
4,352 |
8GB/16GB GDDR6 |
May/Jul 2023 |
PCIe 4.0 x8 |
Dual axial fans (internal exhaust) |
TDP 160W |
| GeForce RTX 4060 |
3,072 |
8GB GDDR6 |
Jun 2023 |
PCIe 4.0 x8 |
Dual axial fans (internal exhaust) |
TDP 115W |
| GeForce RTX 4050 |
- |
6GB GDDR6 |
Not officially announced |
- |
- |
- |
SM_86 / SM_87 (Ampere)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| NVIDIA A100 |
6,912 |
40GB/80GB HBM2(e) |
2020 |
PCIe 4.0 x16 |
Passive |
SXM4 400W, PCIe 250W |
| NVIDIA A100X |
- |
80GB HBM2e |
2021-2022 |
PCIe Gen4 |
Passive |
- |
| NVIDIA A30 |
3,584 |
24GB HBM2 |
2021 |
PCIe 4.0 x16 |
Passive (2-slot) |
TDP 165W |
| NVIDIA A30X |
- |
24GB HBM2e |
2021-2022 |
PCIe Gen4 |
Passive |
- |
| NVIDIA A40 |
10,752 |
48GB GDDR6 ECC |
around 2020-2021 |
PCIe 4.0 x16 |
Blower fan / passive |
300W |
| NVIDIA A10 |
9,216 |
24GB GDDR6 |
2021 |
PCIe 4.0 x16 |
Passive / blower (rear exhaust) |
150W |
| NVIDIA A16 |
(per GPU) 1,280 x4 |
64GB GDDR6 (16GB x4) |
2021 |
PCIe 4.0 x16 |
Passive (multi-GPU board) |
250W |
| NVIDIA A2 Tensor Core |
1,280 |
16GB GDDR6 |
2021 |
PCIe 4.0 x8 |
Compact passive |
60W |
| NVIDIA A800 40GB (China market) |
6,912 |
40GB HBM2 |
around 2022 |
PCIe 4.0 x16 |
Passive |
250W (est., on par with A100 40GB) |
| NVIDIA RTX A6000 |
10,752 |
48GB GDDR6 ECC |
around 2020-2021 |
PCIe 4.0 x16 |
Active |
300W |
| NVIDIA RTX A5500 |
10,240 |
24GB GDDR6 |
2021-2022 |
PCIe 4.0 x16 |
Active |
230W |
| NVIDIA RTX A4500 |
7,168 |
20GB GDDR6 |
2021-2022 |
PCIe 4.0 x16 |
Active |
200W |
| NVIDIA RTX A4000 |
6,144 |
16GB GDDR6 ECC |
2021 |
PCIe 4.0 x16 |
Single blower fan (rear exhaust) |
140W |
| NVIDIA RTX A2000 |
3,328 |
12GB GDDR6 |
2021 |
PCIe 4.0 x16 |
Active (compact) |
70W |
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce RTX 3090 |
10,496 |
24GB GDDR6X |
Sep 2020 |
PCIe 4.0 x16 |
2 fans (internal exhaust) |
TDP 350W |
| GeForce RTX 3080 |
8,704 |
10GB/12GB GDDR6X |
Sep 2020 / Jan 2022 |
PCIe 4.0 x16 |
2 fans (internal exhaust) |
TDP 320W (12GB version: 350W) |
| GeForce RTX 3070 |
5,888 |
8GB GDDR6 |
Oct 2020 |
PCIe 4.0 x16 |
2 fans (internal exhaust) |
TDP 220W |
| GeForce RTX 3060 |
3,584 |
12GB GDDR6 |
Feb 2021 |
PCIe 4.0 x16 |
2 fans (internal exhaust) |
TDP 170W |
| GeForce RTX 3050 |
2,560 |
8GB GDDR6 |
Jan 2022 |
PCIe 4.0 x8 |
2 fans (internal exhaust) |
TDP 130W |
SM_75 (Turing)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| Quadro RTX 8000 |
4,608 |
48GB GDDR6 |
Late 2018 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 295W |
| Quadro RTX 6000 |
4,608 |
24GB GDDR6 |
Late 2018 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 250W |
| Quadro RTX 5000 |
3,072 |
16GB GDDR6 |
2019 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 265W |
| Quadro RTX 4000 |
2,304 |
8GB GDDR6 |
Late 2018 |
PCIe 3.0 x16 |
Single fan (axial) |
TDP 160W |
| T4 |
2,560 |
16GB GDDR6 |
2018 |
PCIe 3.0 x16 |
Passive (half-height) |
70W |
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce RTX 2080 Ti |
4,352 |
11GB GDDR6 |
Sep 2018 |
PCIe 3.0 x16 |
Dual fans (internal exhaust) |
TDP 250W |
| GeForce RTX 2070 |
2,304 |
8GB GDDR6 |
Oct 2018 |
PCIe 3.0 x16 |
Dual fans (internal exhaust) |
TDP 175W |
| GeForce GTX 1660 Ti |
1,536 |
6GB GDDR6 |
Feb 2019 |
PCIe 3.0 x16 |
(No FE; AIB only) |
TDP 120W |
SM_70 / SM_72 (Volta)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| Tesla V100 |
5,120 |
16GB/32GB HBM2 |
2017 |
PCIe 3.0 x16 |
Passive (for servers) |
SXM2 300W, PCIe 250W |
| Quadro GV100 |
5,120 |
32GB HBM2 |
2018 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
250W |
| Titan V |
5,120 |
12GB HBM2 |
2017 |
PCIe 3.0 x16 |
Blower (rear exhaust) |
250W |
| Xavier NX (SoC) |
384 |
8GB LPDDR4x |
around 2019-2020 |
- (SoC) |
- (embedded) |
10-15W (15W max mode) |
SM_60 / SM_61 / SM_62 (Pascal)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| Tesla P100 |
3,584 |
12GB/16GB HBM2 |
2016 |
PCIe 3.0 x16 |
Passive (for servers) |
SXM2 300W, PCIe 250W |
| Quadro GP100 |
3,584 |
16GB HBM2 |
2017 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
235W |
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce GTX 1080 |
2,560 |
8GB GDDR5X |
May 2016 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 180W |
| GeForce GTX 1070 |
1,920 |
8GB GDDR5 |
Jun 2016 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 150W |
| GeForce GTX 1060 (6GB) |
1,280 (6GB) / 1,152 (3GB) |
3GB/6GB GDDR5 |
Jul-Aug 2016 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 120W (6GB) |
| GeForce GTX 1050 |
640 |
2GB/3GB GDDR5 |
Oct 2016 |
PCIe 3.0 x16 |
(No FE; AIB partners) |
~75W (TDP around 75W) |
| GeForce GTX 1030 |
384 |
2GB GDDR5 |
May 2017 |
PCIe 3.0 x4/x8 |
Tiny fan / fanless |
TDP 30W |
| GeForce GT 1010 |
256 |
2GB GDDR5 |
around 2021-2022 |
PCIe 3.0 x4? |
(OEM) |
- |
| Titan Xp |
3,840 |
12GB GDDR5X |
Apr 2017 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 250W |
SM_50 / SM_52 / SM_53 (Maxwell)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| Tesla M40 |
3,072 (GM200) |
12GB GDDR5 |
2015 |
PCIe 3.0 x16 |
Passive (for servers) |
250W |
| Tesla M60 (Dual) |
(per GPU) 2,048 x2 |
16GB total (8GB x2) |
2015 |
PCIe 3.0 x16 |
Passive (dual-GPU board) |
300W |
| Tesla M4 |
1,024 (GM206) |
4GB GDDR5 |
2016 |
PCIe 3.0 x8 |
Passive (low profile) |
50-75W (depending on config) |
| Quadro M6000 |
3,072 (GM200) |
12GB/24GB GDDR5 |
2015 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
250W |
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GeForce GTX 980 Ti |
2,816 |
6GB GDDR5 |
Jun 2015 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 250W |
| GeForce GTX Titan X (original Maxwell) |
3,072 |
12GB GDDR5 |
Mar 2015 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 250W |
| GeForce GTX 980 |
2,048 |
4GB GDDR5 |
Sep 2014 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 165W |
| GeForce GTX 970 |
1,664 |
4GB GDDR5(3.5+0.5GB) |
Sep 2014 |
PCIe 3.0 x16 |
Blower fan (rear exhaust) |
TDP 145W |
SM_35 / SM_37 (Kepler)
Data Center / Professional
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| Tesla K40 |
2,880 (GK110B) |
12GB GDDR5 |
2013 |
PCIe 3.0 x16 |
Passive (for servers) |
TDP 235W |
| Tesla K80 |
(per GPU) 2,496 x2 (GK210) |
24GB total (12GB x2) GDDR5 |
2014 |
PCIe 3.0 x16 |
Passive (dual-GPU board) |
300W |
SM_20 (Fermi)
GeForce
| Product Name |
CUDA Cores |
VRAM |
Release Year |
PCIe Version |
Cooling |
Power (Max/TDP) |
| GTX 480 (GF100) |
480 |
1.5GB GDDR5 |
2010 |
PCIe 2.0 x16 |
Blower fan (rear exhaust) |
TDP 250W |
| GTX 460 (GF104) |
336 |
768MB/1GB |
2010 |
PCIe 2.0 x16 |
Blower fan (rear exhaust) |
TDP 150-160W |
| GTX 580 (GF110) |
512 |
1.5GB |
2011 |
PCIe 2.0 x16 |
Blower fan (rear exhaust) |
TDP 244W |
| GT 630 (Fermi) |
96 (GF108) |
1-2GB |
2012 |
PCIe 2.0 x16 |
Compact / fanless, etc. |
TDP ~65W (OEM) |
Qualiteg Technology Consulting
GPU selection advice grounded in real operational experience.
When it comes to GPUs and inference, the challenges never end—from getting things running to operating them afterward.
We operate our own GPU clusters and develop LLM products. From GPU selection to inference optimization (quantization, vLLM) and distributed configurations, we offer practical advice grounded in operational experience.
Explore our LLM infrastructure services →
