granite-3.3-8b-instruct VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (IBM): 5.98 GiB–6.53 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
This result is an estimate, not a measured trace.Not verifiedFile-backed weights
Publisher and provenance
- PublisherIBM
- Hugging Face
ibm-granite/granite-3.3-8b-instruct - Revision
51dd4bc2ade4059a6bd87649d68aa11e4fb2529b - Total parameters8,170,864,640
- Layers40
- Query heads32
- KV heads8 (gqa)
- Head dimension128
- Recorded context capability131,072
- Completenesscomplete · full
Memory layers in this estimate
Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.
- Weight artifact
- 4.60 GiB (4,942,873,344 bytes)
- Raw KV cache
- 1.25 GiB (1,342,177,280 bytes)
- Engine extras / workspace
- 0.13 GiB–0.68 GiB
- Total estimated demand
- 5.98 GiB–6.53 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| BF16 safetensors bf16 | model-00001-of-00004.safetensors, model-00002-of-00004.safetensors, model-00003-of-00004.safetensors, model-00004-of-00004.safetensors | 16,341,729,280 (15.22 GiB / 16.34 GB) | vllm, sglang, unspecified | file-metadata |
| GGUF Q4_K_M (IBM) gguf-q4_k_m | granite-3.3-8b-instruct-Q4_K_M.gguf | 4,942,873,344 (4.60 GiB / 4.94 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q5_K_M (IBM) gguf-q5_k_m | granite-3.3-8b-instruct-Q5_K_M.gguf | 5,797,462,784 (5.40 GiB / 5.80 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q6_K (IBM) gguf-q6_k | granite-3.3-8b-instruct-Q6_K.gguf | 6,705,464,064 (6.24 GiB / 6.71 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q8_0 (IBM) gguf-q8_0 | granite-3.3-8b-instruct-Q8_0.gguf | 8,684,264,992 (8.09 GiB / 8.68 GB) | llama.cpp, ollama, unspecified | file-metadata |
| INT4 theoretical | No published file | 4,085,432,320 (theoretical) | — | theoretical |
Memory by quantization and context
Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.
| Quantization | 2,048 tokens | 8,192 tokens | 32,768 tokens |
|---|---|---|---|
| BF16 safetensors | 16.16 GiB–18.04 GiB | 17.09 GiB–19.01 GiB | 20.84 GiB–22.87 GiB |
| GGUF Q4_K_M (IBM) | 5.04 GiB–5.51 GiB | 5.98 GiB–6.53 GiB | 9.73 GiB–10.55 GiB |
| GGUF Q5_K_M (IBM) | 5.84 GiB–6.33 GiB | 6.77 GiB–7.34 GiB | 10.52 GiB–11.36 GiB |
| GGUF Q6_K (IBM) | 6.68 GiB–7.19 GiB | 7.62 GiB–8.21 GiB | 11.37 GiB–12.22 GiB |
| GGUF Q8_0 (IBM) | 8.53 GiB–9.07 GiB | 9.46 GiB–10.09 GiB | 13.21 GiB–14.10 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 4.24 GiB–5.84 GiB | 5.18 GiB–7.01 GiB | 8.93 GiB–11.63 GiB |
Raw KV contribution
Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (8), not query heads (32). Using query heads would overstate this cache by 4×.
- 8,192 tokens, 1 sequence, FP16: 1,342,177,280 bytes (1.2500 GiB)
- 32,768 tokens: 5,368,709,120 bytes (5.00 GiB)
Related pages
Sources
- granite-3.3-8b-instruct model card
- config.json — revision 51dd4bc2ade4059a6bd87649d68aa11e4fb2529b
- safetensors index — metadata.total_size=16341729280
- Artifact repo ibm-granite/granite-3.3-8b-instruct-GGUF
Assumptions that affect this result
- Requested context is within the recorded native window of 131072 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Counted files: granite-3.3-8b-instruct-Q4_K_M.gguf.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
- Concurrent sequences share one model replica; weights are not multiplied by sequence count.
- KV dtype is independent of weight quantization unless you change it.
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.
Search or pick a catalog checkpoint to see memory demand.