All models › deepseek-v4 vs gemma-4-E4B-it
deepseek-v4 vs gemma-4-E4B-it — hardware requirements compared (measured)
Both sides use real GGUF file sizes from Hugging Face — not params×bits formulas. Updated daily.
| deepseek-v4 | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 403B | 8B |
| Downloads (30d) | 1956k | 1509k |
| Quantizations published | 6 | 3 |
| Smallest build needs | 118.6 GB (Q2KDown-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-fixed-0731) | 7.0 GB (Q4_0) |
| Recommended quant needs | 118.6 GB (Q2KDown-AProjQ8-SExpQ8-OutQ8) | 11.1 GB (Q8_0) |
| Lossless build needs | 968.5 GB | 19.6 GB |
Verdict: deepseek-v4 needs more memory at the recommended quant. Capability is a different question (see benchmark leaderboards) — this page answers only “which one fits my machine, and at what quality cost?”
Will it run on your GPU?
At the recommended quant, 8K context:
| Hardware | VRAM/RAM | deepseek-v4 | gemma-4-E4B-it |
|---|---|---|---|
| RTX 3060 12GB | 12 GB | no | tight |
| RTX 4070 12GB | 12 GB | no | tight |
| RTX 4090 24GB | 24 GB | no | fits |
| RTX 5090 32GB | 32 GB | no | fits |
| Mac 24GB | 18 GB | no | fits |
| Mac 64GB | 48 GB | no | fits |