Mac Studio M3 Ultra 256GBBudget pick
256 GB Unified (+128)
C
Adds memory headroom for longer context windows and future model growth.112 tok/s decode
Adds memory headroom for longer context windows and future model growth.
~$6,999 MSRP
Can Dolphin 2.9 8B run on Gaudi 3 128GB?
YES — Runs Great
C46Usable
○Estimated from fit model
Dolphin 2.9 8B needs ~20.5 GB VRAM. Gaudi 3 128GB has 128.0 GB. With Q4_K_M quantization, expect ~112 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) — 20.5 GB, 112.0 tok/s, Runs well
20.5 GB required128.0 GB available
16% VRAM used
Fit status
Runs well
Decode
112.0 tok/s
TTFT
1729 ms
Safe context
33K
Memory
20.5 GB / 128.0 GB
Memory breakdown
Weights4.9 GB
KV Cache2.0 GB
Runtime0.9 GB
Headroom12.8 GB
See how fast it feels
See how fast it feelsDolphin 2.9 8B on Gaudi 3 128GB
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: 112.0 tok/s decode · 1.7s TTFT (warm) · 280 tok/s prefill
What limits this setup
The raw memory story may look fine, but the software ecosystem is still a constraint here.
Runtime ecosystem is narrower than CUDA
Intel GPUs can look attractive on memory per dollar, but local AI tooling, kernels, and model coverage are still broader and easier on CUDA today.
Best improvement path
Prefer CUDA if you want the path of least resistance
If your goal is maximum runtime coverage, easier troubleshooting, and better support for new local AI releases, CUDA is usually still the safer upgrade path.
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 112.0 tok/s | 943 ms | 33K |
| Coding | C | Runs well | 112.0 tok/s | 1729 ms | 33K |
| Agentic Coding | C | Runs well | 112.0 tok/s | 2514 ms | 33K |
| Reasoning | C | Runs well | 112.0 tok/s | 2043 ms | 33K |
| RAG | C | Runs well | 112.0 tok/s | 3143 ms | 33K |
Quantization options
How Dolphin 2.9 8B (8B params) fits at each quantization level on Gaudi 3 128GB (128.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 3.1 GB | Low | D39 |
Q3_K_S | 3 | 3.9 GB | Low | D39 |
NVFP4 | 4 |
Get started
Copy-paste commands to run Dolphin 2.9 8B on your machine.
Run
ollama run dolphin-llama3Upgrade options