 |
VOOZH | about |
MPT-30B-Instruct needs ~51.0 GB VRAM. Radeon PRO W7900 DS 48GB has 48.0 GB. With Q5_K_M quantization, expect ~16 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
3.0 GB over capacity — needs offload or smaller quantization
Fit status
Runs with offload (needs ~1.3 GB host RAM)
Decode
15.9 tok/s
TTFT
12202 ms
Safe context
8K
Memory
51.0 GB / 48.0 GB
Offload
10%
It fits through host-memory offload, and offload is the main reason performance drops.
CPU or host-memory offload is active
About 10% 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 1.3 GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | A | Runs well | 24.1 tok/s | 4387 ms | 8K |
| Coding | B | Runs with offload (needs ~1.3 GB host RAM) | 15.9 tok/s | 12202 ms | 8K |
| Agentic Coding | F | Too heavy | 7.2 tok/s | 39321 ms | 8K |
| Reasoning | B | Runs with offload (needs ~1.3 GB host RAM) | 15.9 tok/s | 14420 ms | 8K |
| RAG | F | Too heavy | 7.2 tok/s | 49152 ms | 8K |
How MPT-30B-Instruct (30B params) fits at each quantization level on Radeon PRO W7900 DS 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 11.7 GB | Low | B64 |
Q3_K_S | 3 | 14.7 GB | Low | B65 |
NVFP4 | 4 | 16.8 GB | Medium | B66 |
Q4_K_M | 4 | 18.3 GB | Medium | B66 |
Q5_K_M | 5 | 21.6 GB | High | B67 |
Q6_K | 6 | 24.6 GB | High | B68 |
Q8_0Best for your GPU | 8 | 32.1 GB | Very High | B68 |
F16 | 16 | 61.5 GB | Maximum | F0 |
Copy-paste commands to run MPT-30B-Instruct on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "mosaicml/mpt-30b-instruct" \
--hf-file "mpt-30b-instruct-Q5_K_M.gguf" \
-c 4096 -ngl 99Upgrade options
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 1635%.
~$8,000 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 231%.
~$10,000 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 1002%.
~$12,000 MSRP