 |
VOOZH | about |
StarCoder2 15B needs ~14.1 GB VRAM. RTX 5070 12GB has 12.0 GB. With Q5_K_M quantization, expect ~22 tok/s.
Operating mode
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
2.1 GB over capacity — needs offload or smaller quantization
Fit status
Very compromised (needs ~1.6 GB host RAM)
Decode
22.6 tok/s
TTFT
8556 ms
Safe context
4K
Memory
14.1 GB / 12.0 GB
Offload
20%
It fits through host-memory offload, and offload is the main reason performance drops.
CPU or host-memory offload is active
About 20% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Remove offload with more accelerator memory
Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Increase host RAM if you keep offloading
This setup may need roughly {ram} GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Very compromised | 24.0 tok/s | 4399 ms | 4K |
| Coding | D | Very compromised | 21.9 tok/s | 8825 ms | 4K |
| Agentic Coding | F | Too heavy | 18.5 tok/s | 15202 ms | 4K |
| Reasoning | D | Very compromised | 21.9 tok/s | 10429 ms | 4K |
| RAG | F | Too heavy | 18.5 tok/s | 19002 ms | 4K |
How StarCoder2 15B (15B params) fits at each quantization level on RTX 5070 12GB (12.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 5.9 GB | Low | C54 |
Q3_K_S | 3 | 7.4 GB | Low | C53 |
NVFP4Best for your GPU |
Copy-paste commands to run StarCoder2 15B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "bigcode/starcoder2-15b" \
--hf-file "starcoder2-15b-Q5_K_M.gguf" \
-c 4096 -ngl 99Upgrade options
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Adds memory headroom for longer context windows and future model growth.
~$449 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Adds memory headroom for longer context windows and future model growth.
~$499 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Adds memory headroom for longer context windows and future model growth.
~$625 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 258%.
~$1,599 MSRP
| 4 |
8.4 GB |
| Medium |
| C53 |
Q4_K_M | 4 | 9.2 GB | Medium | F0 |
Q5_K_M | 5 | 10.8 GB | High | F0 |
Q6_K | 6 | 12.3 GB | High | F0 |
Q8_0 | 8 | 16.1 GB | Very High | F0 |
F16 | 16 | 30.7 GB | Maximum | F0 |
Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
