Radeon RX 7900 XTX
AMD
CompatibleRuns Gemma 4 31B IT Q4_K_M entirely in memory
Model · Google
Gemma 4 31B IT: VRAM and hardware requirements
At Q4_K_M with an 8K context, Gemma 4 31B IT needs about 21.3 GB: a GPU with 24 GB or more runs it entirely in VRAM.
- Parameters31.3B
- Max context256K
- LicenseApache-2.0
Specifications
- Organization
- Architecture
- gemma4
- Released
- March 2026
- Layers
- 60
- Hidden size
- 5376
- Attention heads
- 32
- KV heads
- 16
- Head dimension
- 256
- Vocabulary
- 262,144
- KV cache per 1K tokens (f16)
- 0.859 GB
- Repository
- google/gemma-4-31B-it
Official specData updated todaySource: Hugging Face config.json + model card
Quantizations and memory
| Quantization | File size | Memory needed | Minimum GPU memory | Quality |
|---|---|---|---|---|
| BF16 | 57.2 GB | ~63.8 GB | 80 GB | Reference (unquantized) |
| Q8_0 | 30.4 GB | ~35.1 GB | 48 GB | Near-lossless |
| Q6_K | 23.5 GB | ~27.7 GB | 32 GB | Near-lossless |
| Q5_K_M | 20.2 GB | ~24.4 GB | 32 GB | Very good |
| Q4_K_MMost used | 17.1 GB | ~21.3 GB | 24 GB | Good — the usual choice |
| IQ4_XS | 15.2 GB | ~19.6 GB | 20 GB | Good — the usual choice |
| Q3_K_M | 13.7 GB | ~17.9 GB | 20 GB | Noticeable loss |
| Q2_K | Not available | ~14.7 GB | 16 GB | Significant loss |
File sizes: not available until imported from the model repository. Memory = weights (estimated from bits per weight) + KV cache + runtime overhead + safety margin. (BF16 ≈ 16 bpw, Q8_0 ≈ 8.5 bpw, Q6_K ≈ 6.57 bpw, Q5_K_M ≈ 5.7 bpw, Q4_K_M ≈ 4.9 bpw, IQ4_XS ≈ 4.46 bpw, Q3_K_M ≈ 4 bpw, Q2_K ≈ 3.17 bpw)
Q4, Q5 or Q8: how to choose a quantization
GPUs that run Gemma 4 31B IT
| RTX 509032 GB · 64 GB RAM~20.5 GB · 58–83 tok/s estimate | Excellent |
|---|---|
| RTX 409024 GB · 64 GB RAM~20.5 GB · 36–49 tok/s estimate | Compatible |
| RTX 309024 GB · 64 GB RAM~20.5 GB · 33–46 tok/s estimate | Compatible |
| RX 7900 XTX24 GB · 64 GB RAM~20.5 GB · 29–43 tok/s estimate | Compatible |
| RTX 5060 Ti 16 GB16 GB · 64 GB RAM~20.5 GB · 3.5–6.5 tok/s estimate | CPU/GPU offload |
| RTX 5070 Ti16 GB · 64 GB RAM~20.5 GB · 4–7.5 tok/s estimate | CPU/GPU offload |
| RTX 508016 GB · 64 GB RAM~20.5 GB · 4–7.5 tok/s estimate | CPU/GPU offload |
| RTX 4060 Ti 16 GB16 GB · 64 GB RAM~20.5 GB · 3–6 tok/s estimate | CPU/GPU offload |
| RTX 4070 Ti SUPER16 GB · 64 GB RAM~20.5 GB · 3.5–7 tok/s estimate | CPU/GPU offload |
| RX 9070 XT16 GB · 64 GB RAM~20.5 GB · 3.5–7 tok/s estimate | CPU/GPU offload |
| RTX 3060 12 GB12 GB · 64 GB RAM~20.5 GB · 2–4 tok/s estimate | Possible but slow |
| RTX 407012 GB · 64 GB RAM~20.5 GB · 2–4 tok/s estimate | Possible but slow |
| RTX 507012 GB · 64 GB RAM~20.5 GB · 2–4.5 tok/s estimate | Possible but slow |
| RTX 40608 GB · 64 GB RAM~20.5 GB · 1.5–3 tok/s estimate | Possible but slow |
Macs that run Gemma 4 31B IT
| M3 Ultra (80-core GPU) · 512 GBMac Studio~20.3 GB · 24–40 tok/s estimate | Excellent |
|---|---|
| M4 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~20.3 GB · 17–26 tok/s estimate | Excellent |
| M3 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~20.3 GB · 13–20 tok/s estimate | Excellent |
| M4 Max (32-core GPU) · 36 GBMacBook Pro 14, MacBook Pro 16~20.3 GB · 13–20 tok/s estimate | Excellent |
| M4 Pro (20-core GPU) · 64 GBMac mini, MacBook Pro 14~20.3 GB · 9–13 tok/s estimate | Compatible |
| M4 · 32 GBMacBook Air, MacBook Pro 14~20.3 GB · 4–6 tok/s estimate | Compatible |
| M5 · 32 GBMacBook Pro 14~20.3 GB · 5–7.5 tok/s estimate | Compatible |
| M3 · 24 GBMacBook Air, MacBook Pro 14~20.3 GB · No data | Incompatible |
EstimateData updated 40 days agoSource: Run LLM Locally engine
Limits in plain words
The KV cache grows with context: at 32K it adds about 3.3 GB on top of the weights.
With under 8 GB of VRAM, use a smaller quantization or a smaller model of the same family.
Where to buy
Prices and links follow the market you choose.
AMD Ryzen AI Max+ 395
Unified memory PC
CompatibleRuns Gemma 4 31B IT Q4_K_M entirely in memory
GeForce RTX 2080 Ti
Lowest memory tier
Compatible with limitsRuns Gemma 4 31B IT Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
How much VRAM does Gemma 4 31B IT need?
About 21.3 GB at Q4_K_M with an 8K context, including the KV cache and runtime overhead. At Q8_0 it needs about 35.1 GB. A longer context adds more: about 3.3 GB of KV cache at 32K.
Can Gemma 4 31B IT run on a Mac?
Yes on a Mac whose GPU can use at least 20.3 GB, e.g. M4 32 GB. macOS lets the GPU use about two thirds to three quarters of unified memory by default.
How fast is Gemma 4 31B IT?
On a RTX 4090, we estimate 36–49 tok/s at Q4_K_M. This is an estimate from memory bandwidth (confidence low) until benchmarks are imported.