OLMo 2 32B needs ~27.0 GB VRAM. RTX 3090 24GB has 24.0 GB. With Q4_K_M quantization, expect ~21 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
Very compromised (needs ~2.2 GB host RAM)
Decode
21.2 tok/s
TTFT
9143 ms
Safe context
4K
Memory
27.0 GB / 24.0 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 2.2 GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | A | Runs with offload (needs ~0.8 GB host RAM) | 24.8 tok/s | 4259 ms | 4K |
| Coding | A | Very compromised (needs ~2.2 GB host RAM) | 21.2 tok/s | 9143 ms | 4K |
| Agentic Coding | F | Too heavy | 15.9 tok/s | 17670 ms | 4K |
| Reasoning | A | Very compromised (needs ~2.2 GB host RAM) | 21.2 tok/s | 10805 ms | 4K |
| RAG | F | Too heavy | 15.9 tok/s | 22087 ms | 4K |
Inference speed
Estimated decode speed (tokens/sec) for OLMo 2 32B at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~66 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.
| GPU / Mac | Memory | Quant | Speed (tok/s) | Fits? |
|---|---|---|---|---|
 RTX 5090 32GB | 32 GB | Q4_K_M | 66.4 | Tight |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 33.2 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 33.2 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 30.8 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 25.7 | Fits |
| RTX 4090 24GB | 24 GB | Q4_K_M | 24.8 | Heavy offload |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 24.3 | Fits |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 22.9 | Heavy offload |
| RTX 3090 24GB | 24 GB | Q4_K_M | 21.2 | Heavy offload |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 20.9 | Tight |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 13.3 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 12.2 | Fits |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 9.0 | Too big |
| RTX 4070 12GB | 12 GB | Q4_K_M | 3.1 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.0 | Too big |
Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.
How OLMo 2 32B (32B params) fits at each quantization level on RTX 3090 24GB (24.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 12.5 GB | Low | A83 |
Q3_K_S | 3 | 15.7 GB | Low | A82 |
NVFP4Best for your GPU | 4 | 17.9 GB | Medium | A82 |
Q4_K_M | 4 | 19.5 GB | Medium | F0 |
Q5_K_M | 5 | 23.0 GB | High | F0 |
Q6_K | 6 | 26.2 GB | High | F0 |
Q8_0 | 8 | 34.2 GB | Very High | F0 |
F16 | 16 | 65.6 GB | Maximum | F0 |
Copy-paste commands to run OLMo 2 32B on your machine.
Run
lms load OLMo-2-0325-32B-Instruct && lms server startYour hardware
| Model | Params | Grade | Decode | Capabilities |
|---|---|---|---|---|
 Qwen 3.5 35B A3B | 35B | A | 55.5 tok/s |
Yes, RTX 3090 24GB can run OLMo 2 32B with a A grade (Very compromised (needs ~2.2 GB host RAM)). Expected decode speed: 21.2 tok/s.
OLMo 2 32B (32B parameters) requires approximately 27.0 GB of memory with Q4_K_M quantization.
The recommended quantization for OLMo 2 32B is Q4_K_M, which balances quality and memory efficiency.
On RTX 3090 24GB, OLMo 2 32B achieves approximately 21.2 tokens per second decode speed with a time-to-first-token of 9143ms using Q4_K_M quantization.
For coding workloads, OLMo 2 32B on RTX 3090 24GB receives a A grade with 21.2 tok/s and 4K context.
On RTX 3090 24GB, OLMo 2 32B can safely use up to 4K tokens of context. The model's official context limit is 4K, but available memory constrains the safe maximum.
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.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/olmo-2-32b-on-rtx-3090-24gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: