Skip to content
Bitpute

GPU database

Every card, judged by what it actually runs

By Bitpute · Updated 24 July 2026 · How we estimate · Sources · Editorial policy · Version history · Report an error

Spec sheets tell you the memory and the clocks. These pages tell you what that means: which of the 12 models in our engine each card fits at Q4_K_M and Q8_0, how fast its bandwidth can possibly go, what it costs in electricity, and when renting beats buying. Prefer to start from the model instead? Use the model guides.

RTX 3060 12GB

12 GB · 360 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

RTX 4060 Ti 16GB

16 GB · 288 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

RTX 4070 Ti SUPER

16 GB · 672 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

RTX 3090

24 GB · 936 GB/s
Runs 10 of 12 models at Q4 · largest: Qwen2.5 32B

Full guide →

RTX 4090

24 GB · 1008 GB/s
Runs 10 of 12 models at Q4 · largest: Qwen2.5 32B

Full guide →

RTX 5090

32 GB · 1792 GB/s
Runs 10 of 12 models at Q4 · largest: Qwen2.5 32B

Full guide →

RTX A6000

48 GB · 768 GB/s
Runs 11 of 12 models at Q4 · largest: Llama 3.3 70B

Full guide →

RX 7600 XT

16 GB · 288 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

RX 7900 XT

20 GB · 800 GB/s
Runs 8 of 12 models at Q4 · largest: Gemma 2 27B

Full guide →

RX 7900 XTX

24 GB · 960 GB/s
Runs 10 of 12 models at Q4 · largest: DeepSeek-R1 32B

Full guide →

Intel Arc B580

12 GB · 456 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

Intel Arc A770 16GB

16 GB · 560 GB/s
Runs 7 of 12 models at Q4 · largest: Qwen2.5 14B

Full guide →

L40S

48 GB · 864 GB/s
Runs 11 of 12 models at Q4 · largest: Llama 3.3 70B

Full guide →

A100 40GB

40 GB · 1555 GB/s
Runs 10 of 12 models at Q4 · largest: Qwen2.5 32B

Full guide →

A100 80GB

80 GB · 2039 GB/s
Runs 12 of 12 models at Q4 · largest: Qwen2.5 72B

Full guide →

H100 80GB

80 GB · 3350 GB/s
Runs 12 of 12 models at Q4 · largest: Qwen2.5 72B

Full guide →

H200 141GB

141 GB · 4800 GB/s
Runs 12 of 12 models at Q4 · largest: Qwen2.5 72B

Full guide →

New to this? Start with why VRAM decides what you can run.

Browse by tier

The three product segments behave differently on memory, interconnect and cooling — which is usually what decides the choice before price does.

Consumer

Gaming cards — the cheapest route to VRAM. No ECC memory, and on NVIDIA no NVLink from Ada onward.

RTX 3060 12GB12 GBRTX 4060 Ti 16GB16 GBRTX 4070 Ti SUPER16 GBRTX 309024 GBRTX 409024 GBRTX 509032 GBIntel Arc A770 16GBIntel Arc B580RX 7600 XTRX 7900 XTRX 7900 XTX

Workstation

Professional cards: ECC memory and blower cooling for multi-card desktops.

RTX A600048 GB

Data centre

Server GPUs: HBM bandwidth, NVLink, MIG partitioning, rack thermals.

A100 40GB40 GBL40S48 GBA100 80GB80 GBH100 80GB80 GBH200 141GB141 GB