MacBook Pro M4 Max 96GBBudget pick
96 GB Unified (+48)
C
Adds memory headroom for longer context windows and future model growth.4.9 tok/s decode
Adds memory headroom for longer context windows and future model growth.
~$2,499 MSRP
LFM2.5 350M needs ~6.1 GB VRAM. NVIDIA L40 48GB has 48.0 GB. With Q4_K_M quantization, expect ~6 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
Fit status
Runs well
Decode
5.6 tok/s
TTFT
34571 ms
Safe context
128K
Memory
6.1 GB / 48.0 GB
This model fits, but memory bandwidth is the part holding decode speed back.
Throughput will feel slow
Estimated decode speed is only 5.6 tok/s, so this is more of a technical fit than a comfortable daily-driver setup.
Prioritize bandwidth, not only capacity
If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 5.6 tok/s | 18857 ms | 128K |
| Coding | C | Runs well | 5.6 tok/s | 34571 ms | 128K |
| Agentic Coding | C | Runs well | 5.6 tok/s | 50286 ms | 128K |
| Reasoning | C | Runs well | 5.6 tok/s | 40857 ms | 128K |
| RAG | C | Runs well | 5.6 tok/s | 62857 ms | 128K |
Inference speed
Estimated decode speed (tokens/sec) for LFM2.5 350M at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~7 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 | 6.6 | Fits |
| RTX 4090 24GB | 24 GB | Q4_K_M | 5.6 | Fits |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 5.6 | Fits |
| RTX 4070 12GB | 12 GB | Q4_K_M | 5.6 | Fits |
| RTX 4060 8GB | 8 GB | Q4_K_M | 5.6 | Fits |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 4.9 | Fits |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 4.9 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 4.9 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 4.9 | Fits |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 4.9 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 4.9 | Fits |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 4.9 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 4.9 | Fits |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 4.9 | Fits |
| RTX 3090 24GB | 24 GB | Q4_K_M | 4.2 | Fits |
| RTX 3060 12GB | 12 GB | Q4_K_M | 4.2 | Fits |
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 LFM2.5 350M (0.3499999940395355B params) fits at each quantization level on NVIDIA L40 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 0.1 GB | Low | C53 |
Q3_K_S | 3 | 0.2 GB | Low | C53 |
NVFP4 | 4 | 0.2 GB | Medium | C53 |
Q4_K_M | 4 | 0.2 GB | Medium | C53 |
Q5_K_M | 5 | 0.3 GB | High | C53 |
Q6_K | 6 | 0.3 GB | High | C53 |
Q8_0 | 8 | 0.4 GB | Very High | C53 |
F16Best for your GPU | 16 | 0.7 GB | Maximum | C53 |
Copy-paste commands to run LFM2.5 350M on your machine.
Run
lms load LFM2.5-350M && lms server startUpgrade options
Adds memory headroom for longer context windows and future model growth.
~$2,499 MSRP
Adds memory headroom for longer context windows and future model growth.
~$2,499 MSRP
Yes, NVIDIA L40 48GB can run LFM2.5 350M with a C grade (Runs well). Expected decode speed: 5.6 tok/s.
LFM2.5 350M (0.3499999940395355B parameters) requires approximately 6.1 GB of memory with Q4_K_M quantization.
The recommended quantization for LFM2.5 350M is Q4_K_M, which balances quality and memory efficiency.
On NVIDIA L40 48GB, LFM2.5 350M achieves approximately 5.6 tokens per second decode speed with a time-to-first-token of 34571ms using Q4_K_M quantization.
For coding workloads, LFM2.5 350M on NVIDIA L40 48GB receives a C grade with 5.6 tok/s and 128K context.
On NVIDIA L40 48GB, LFM2.5 350M can safely use up to 128K tokens of context. The model's official context limit is 128K, but available memory constrains the safe maximum.
Prioritize bandwidth, not only capacity. If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/lfm2.5-350m-on-l40-48gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: