NVIDIA A100 80GB经济之选2039 GB/s (+1220)
C
Raises estimated decode speed by about 207%.133.7 tok/s 解码
Raises estimated decode speed by about 207%.
~$15,000 MSRP
Can Codestral 21B Pruned i1 run on Mac Studio M3 Ultra 96GB?
YES — Runs Great
C47Usable
Estimated from fit model
Codestral 21B Pruned i1 needs ~26.5 GB VRAM. Mac Studio M3 Ultra 96GB has 69.1 GB. With Q4_K_M quantization, expect ~44 tok/s.
Runtime: llama.cppCapacity: RoomyBandwidth: HighStack: StandardBottleneck: Balanced
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) — 26.5 GB, 43.5 tok/s, Runs well
26.5 GB required69.1 GB available
38% VRAM used
Fit status
Runs well
Decode
43.5 tok/s
TTFT
4453 ms
Safe context
293K
Memory
26.5 GB / 69.1 GB
Memory breakdown
Weights12.8 GB
KV Cache2.5 GB
Runtime0.9 GB
Headroom10.4 GB
See how fast it feels
See how fast it feelsCodestral 21B Pruned i1 on Mac Studio M3 Ultra 96GB
1st promptCold start — includes initialization
>Write a Python function to sort a list
Here's a Python function that sorts a list using quicksort:
def quicksort(arr):
if len(arr) <= 1:
return arr
pivot = arr[len(arr) // 2]
left = [x for x in arr if x < pivot]
middle = [x for x in arr if x == pivot]
right = [x for x in arr if x > pivot]
return quicksort(left) + middle + quicksort(right)
# Example usage
numbers = [3, 6, 8, 10, 1, 2, 1]
print(quicksort(numbers)) # [1, 1, 2, 3, 6, 8, 10]
The algorithm picks a pivot, partitions the list into smaller, equal, and larger elements, then recursively sorts each partition and concatenates the results.
2nd promptWarm — model ready, no init overhead
>Now add type hints and a docstring
def quicksort(arr: list[int]) -> list[int]:
"""Sort a list using the quicksort algorithm.
Args:
arr: List of integers to sort.
Returns:
A new sorted list.
"""
if len(arr) <= 1:
return arr
pivot = arr[len(arr) // 2]
left = [x for x in arr if x < pivot]
middle = [x for x in arr if x == pivot]
right = [x for x in arr if x > pivot]
return quicksort(left) + middle + quicksort(right)
Added type hints for the input and return type, plus a Google-style docstring.
Estimated: 43.5 tok/s decode · 4.5s TTFT (warm) · 109 tok/s prefill
What limits this setup
This setup is broadly balanced for this model.
Shared-memory contention still exists
The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.
Best improvement path
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 43.5 tok/s | 2429 ms | 293K |
| Coding | C | Runs well | 43.5 tok/s | 4453 ms | 293K |
| Agentic Coding | C | Runs well | 43.5 tok/s | 6477 ms | 293K |
| Reasoning | C | Runs well | 43.5 tok/s | 5263 ms | 293K |
| RAG | C | Runs well | 43.5 tok/s | 8097 ms | 293K |
Quantization options
How Codestral 21B Pruned i1 (21B params) fits at each quantization level on Mac Studio M3 Ultra 96GB (69.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.2 GB | Low | C40 |
Q3_K_S | 3 | 10.3 GB | Low | C40 |
NVFP4 | 4 | 11.8 GB | Medium | C41 |
Q4_K_M | 4 | 12.8 GB | Medium | C41 |
Q5_K_M | 5 | 15.1 GB | High | C41 |
Q6_K | 6 | 17.2 GB | High | C42 |
Q8_0 | 8 | 22.5 GB | Very High | C43 |
F16Best for your GPU | 16 | 43.1 GB | Maximum | C47 |
Get started
Copy-paste commands to run Codestral 21B Pruned i1 on your machine.
Run
lms load hf-mradermacher--codestral-21b-pruned-i1-gguf && lms server start升级选项
能流畅运行 Codestral 21B Pruned i1 的硬件
Frequently asked questions
Can Mac Studio M3 Ultra 96GB run Codestral 21B Pruned i1?
Yes, Mac Studio M3 Ultra 96GB can run Codestral 21B Pruned i1 with a C grade (Runs well). Expected decode speed: 43.5 tok/s.
How much VRAM does Codestral 21B Pruned i1 need?
Codestral 21B Pruned i1 (21B parameters) requires approximately 26.5 GB of memory with Q4_K_M quantization.
What is the best quantization for Codestral 21B Pruned i1?
The recommended quantization for Codestral 21B Pruned i1 is Q4_K_M, which balances quality and memory efficiency.
What speed will Codestral 21B Pruned i1 run at on Mac Studio M3 Ultra 96GB?
On Mac Studio M3 Ultra 96GB, Codestral 21B Pruned i1 achieves approximately 43.5 tokens per second decode speed with a time-to-first-token of 4453ms using Q4_K_M quantization.
Can Mac Studio M3 Ultra 96GB run Codestral 21B Pruned i1 for coding?
For coding workloads, Codestral 21B Pruned i1 on Mac Studio M3 Ultra 96GB receives a C grade with 43.5 tok/s and 293K context.
What context window can Codestral 21B Pruned i1 use on Mac Studio M3 Ultra 96GB?
On Mac Studio M3 Ultra 96GB, Codestral 21B Pruned i1 can safely use up to 293K tokens of context. The model's official context limit is —, but available memory constrains the safe maximum.
Is unified memory on Mac Studio M3 Ultra 96GB as fast as VRAM for Codestral 21B Pruned i1?
Not always. Mac Studio M3 Ultra 96GB can often fit larger models thanks to unified memory, but a discrete GPU with dedicated high-bandwidth VRAM may still decode faster once the model fits. For this combination, the important distinction is capacity versus sustained throughput.
Embed this result▼
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/hf-mradermacher--codestral-21b-pruned-i1-gguf-on-m3-ultra-96gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: