Mac Studio M3 Ultra 96GBBudget pick
96 GB Unified (+48)
C
Adds memory headroom for longer context windows and future model growth.91.3 tok/s decode
Adds memory headroom for longer context windows and future model growth.
~$3,999 MSRP
Can HelpingAI2.5 10B i1 run on Radeon Pro W7900 48GB?
YES — Runs Great
C47Usable
○Estimated from fit model
HelpingAI2.5 10B i1 needs ~13.0 GB VRAM. Radeon Pro W7900 48GB has 48.0 GB. With Q4_K_M quantization, expect ~84 tok/s.
Runtime: llama.cppCapacity: RoomyBandwidth: HighStack: StandardBottleneck: Balanced
Share:
Operating mode
Choose the run profile you care about
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
Q4_K_M (Medium quality) — 13.0 GB, 83.6 tok/s, Runs well
13.0 GB required48.0 GB available
27% VRAM used
Fit status
Runs well
Decode
83.6 tok/s
TTFT
2317 ms
Safe context
494K
Memory
13.0 GB / 48.0 GB
Memory breakdown
Weights6.1 GB
KV Cache1.2 GB
Runtime0.9 GB
Headroom4.8 GB
See how fast it feels
See how fast it feelsHelpingAI2.5 10B i1 on Radeon Pro W7900 48GB
1st promptCold start — includes initialization
>What is local AI inference?
Local AI inference means running an AI language model directly on your own hardware — your laptop, desktop, or server — instead of sending requests to a remote cloud API.
When you run inference locally the model weights are loaded into your GPU or unified memory. Each token you generate requires reading those weights from memory, so memory bandwidth is the main bottleneck for decode speed.
Key benefits of running locally:
- Full privacy: your prompts never leave your machine
- No per-token cost or rate limits
- Works offline once the model is downloaded
- Latency depends only on your hardware
2nd promptWarm — model ready, no init overhead
>How much VRAM do I need?
It depends on the model size and quantization level. A rough rule of thumb:
Model size Q4 (4-bit) Q8 (8-bit) FP16
7B params ~4.3 GB ~7.5 GB ~14 GB
13B params ~7.9 GB ~13.9 GB ~26 GB
70B params ~42.7 GB ~74.9 GB ~140 GB
Most people use 4-bit quantization (Q4_K_M) which gives 90-95% of full quality at a fraction of the memory. A 24 GB GPU can comfortably run most 7B-13B models.
Estimated: 83.6 tok/s decode · 2.3s TTFT (warm) · 209 tok/s prefill
What limits this setup
This setup is broadly balanced for this model.
No major red flags
This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.
Best improvement path
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 83.6 tok/s | 1264 ms | 494K |
| Coding | C | Runs well | 83.6 tok/s | 2317 ms | 494K |
| Agentic Coding | C | Runs well | 83.6 tok/s | 3370 ms | 494K |
| Reasoning | C | Runs well | 83.6 tok/s | 2738 ms | 494K |
| RAG | C | Runs well | 83.6 tok/s | 4212 ms | 494K |
Quantization options
How HelpingAI2.5 10B i1 (10B params) fits at each quantization level on Radeon Pro W7900 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 3.9 GB | Low | C41 |
Q3_K_S | 3 | 4.9 GB | Low | C41 |
NVFP4 | 4 | 5.6 GB | Medium | C41 |
Q4_K_M | 4 | 6.1 GB | Medium | C41 |
Q5_K_M | 5 | 7.2 GB | High | C42 |
Q6_K | 6 | 8.2 GB | High | C42 |
Q8_0 | 8 | 10.7 GB | Very High | C42 |
F16Best for your GPU | 16 | 20.5 GB | Maximum | C46 |
Get started
Copy-paste commands to run HelpingAI2.5 10B i1 on your machine.
Run
lms load hf-mradermacher--helpingai2-5-10b-i1-gguf && lms server startUpgrade options