L4
L4 has 24 GB of memory and 300 GB/s of memory bandwidth. 360 open models in our directory fit at 4-bit.
Planning a multi-GPU deployment? Check NVLink domains and network topology as well as memory fit. A GPU specification alone does not establish the interconnect in a server or cloud allocation.
| Category | Fits | Tight | Too big | Total |
|---|---|---|---|---|
| LLM | 183 | 16 | 81 | 280 |
| Vision | 75 | 3 | 28 | 106 |
| Image | 22 | 1 | 1 | 24 |
| Video | 22 | 0 | 0 | 22 |
| Speech | 39 | 1 | 0 | 40 |
| Embedding | 19 | 0 | 0 | 19 |
Largest open models that fit at 4-bit
See all 360 in the tableA model fits when its 4-bit memory estimate (8k context) stays within 90% of usable memory, about 21.6 GB here. Showing the 30 largest of 360.
| Model | Type | Size | Memory at 4-bit |
|---|---|---|---|
| LLM | 32.2B | 21.4 GB | |
| LLM | 31.6B | 21.3 GB | |
| LLM | 31.6B | 21.3 GB | |
| LLM | 30.1B | 20.7 GB | |
| LLM | 30.5B | 20.4 GB | |
| Vision | 29.7B | 20.1 GB | |
| Vision | 29.4B | 19.9 GB | |
| Vision | 27.4B | 19.6 GB | |
| Vision | 27.8B | 18.9 GB | |
| Vision | 27.8B | 18.9 GB | |
| Vision | 27.8B | 18.9 GB | |
| LLM | 22.1B | 17.6 GB | |
| Vision | 25.8B | 17.4 GB | |
| Vision | 25.8B | 17.4 GB | |
| LLM | 23.6B | 17 GB | |
| LLM | 11.1B | 16.2 GB | |
| LLM | 21.9B | 15 GB | |
| LLM | 21.8B | 14.9 GB | |
| LLM | 19.9B | 14.9 GB | |
| LLM | 21.5B | 14.4 GB | |
| LLM | 20.9B | 14 GB | |
| Video | 18.9B | 13.5 GB | |
| Video | 19B | 13.5 GB | |
| Vision | 16.4B | 12.8 GB | |
| Vision | 16.4B | 12.8 GB | |
| Vision | 16.4B | 12.8 GB | |
| LLM | 16B | 12.5 GB | |
| LLM | 16B | 12.5 GB | |
| LLM | 15.7B | 12.4 GB | |
| Video | 17.3B | 12.4 GB |
These are memory checks, not speed or runtime compatibility results.
Compared with
All hardware| Hardware | Total memory | Usable for models | Bandwidth |
|---|---|---|---|
L4This device | 24 GB | Up to 24 GB | 300 GB/s |
| 48 GB | Up to 48 GB | 864 GB/s | |
| 80 GB | Up to 80 GB | 2,039 GB/s | |
| 80 GB | Up to 80 GB | 3,350 GB/s |
Capacity and bandwidth answer different questions: memory decides what fits, bandwidth drives how fast it runs.
Launch dates, useful-life evidence and ownership cost scenarios