Local coding on Linux x64 with 8–32 GB, no GPU
The tested setup is llamacpp/qwen3-1.7b on llamacpp: Q4_K_M weights, a 32,768-token context, and a 1.1 GB download. On a Linux x64 container with 7 GB of RAM, the probe took 157.3 s cold and 119.2 s warm, with a peak of 5.82 GB of RAM. 7 GB is below the 8 GB floor the catalog sets for this class.
Run it
Install the CLI. Start this exact setup. Open the harness. The first pull creates a free handle.
An explicit setup runs on the OS its catalog entry declares, after that entry's RAM, VRAM, and driver checks. The no-arg bootstrap is macOS on Apple silicon only.
curl -fsSL https://oloop.run/install.sh | sh
oloop host start --setup linux-x64-cpu@1.2.2
oloop host openWhy this setup without a GPU
llamacpp/qwen3-1.7b is a 1.7B model at Q4_K_M, run by llamacpp on the CPU. The download is 1.1 GB and the measured peak was 5.82 GB of system RAM, so it runs on a small box with no discrete card.
The trade is capability. This is a smaller model, and it fits where an 8B model would not. With a GPU and more memory, the L4 page is the measured step up.
| Fact | Value |
|---|---|
| OS / arch | linux / x64 |
| Memory kind | cpu |
| Host | llamacpp@b11292 |
| Harness | opencode@1.18.21 |
| Model | llamacpp/qwen3-1.7b |
| Model revision | unsloth/Qwen3-1.7B-GGUF@d7f544eead698dbd1f15126ef60b45a1e1933222 |
| Quantization | q4_k_m |
| Context window | 32,768 tokens |
| Download | 1.1 GB |
Source: /setups/evidence/linux-x64-cpu@1.2.2.json (qualified 2026-10-05, engine 0.1.0).
Measured results
The timer measures one probe. The prompt asks for the word OK in answer.txt, and the run is capped at 3 turns. Cold is the first run, which starts the host and loads the model; warm repeats the same probe on the same machine. It is not coding throughput and not a tokens-per-second figure.
Warm was 38.1 s faster than cold here. One probe pair gives a direction, not a rate.
| Run | Cold | Warm | Peak RAM | Peak VRAM |
|---|---|---|---|---|
| Probe | 157.3 s | 119.2 s | 5.82 GB | none (no discrete GPU) |
Measured on Linux x64, 7 GB RAM on 2026-10-05. Source: /setups/evidence/linux-x64-cpu@1.2.2.json.
Tested on, listed for
Tested on: Linux x64, 7 GB RAM.
Listed for: Linux x64, 8–32 GB RAM.
The catalog lists a range, and the record covers one machine in it. The times and memory numbers belong to that machine.
What the run proves
One pinned setup passed one probe on the stated hardware and date, using the checksummed bundle from the catalog.
- The run is not a security review. It checks the named file; it does not audit the model, the harness, or their permissions.
- A pass is not a quality score. It means the configured check passed, not that every answer is correct.
- The model license belongs to the model publisher. Read it before you ship the output.
- The bundle is pinned to one version. A newer catalog version is a different record.
- The tested box had 7 GB, below the catalog's 8 GB floor. Its times are not a 32 GB machine's times.
Questions
Does this need a GPU?
llamacpp/qwen3-1.7b runs on the CPU at Q4_K_M. The record lists no discrete card.
Will it run in 8 GB of RAM?
The measured peak was 5.82 GB on a 7 GB box, so an 8 GB machine has more headroom than the test box did. The catalog lists 8–32 GB, and we measured only the 7 GB box.
Why such a small model?
A 1.7B model is what fits and starts on a small CPU box. It is a coding model, and it is not the same capability as an 8B model on a GPU.
What is the step up from here?
The NVIDIA L4 class is the measured GPU step. More RAM on the same CPU is a different configuration for a larger model, and it is unmeasured.
Can I use a larger model on a CPU box?
Only as a new qualification. A 7B model needs more memory and a longer download, and we have not measured one on a CPU box.
Related
Evidence
The record for linux-x64-cpu@1.2.2 is where every number comes from: the hardware, the timings, the peaks, and the date.