1344 GB/s (+544)
C
Raises estimated decode speed by about 143%.88.1 tok/s decode
Raises estimated decode speed by about 143%.
~$4,999 MSRP
Can Codestral 21B Pruned i1 run on Mac Studio M2 Ultra 64GB?
YES — Runs Great
C49Usable
○Estimated from fit model
Codestral 21B Pruned i1 needs ~23.1 GB VRAM. Mac Studio M2 Ultra 64GB has 46.1 GB. With Q4_K_M quantization, expect ~36 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) — 23.1 GB, 36.2 tok/s, Runs well
23.1 GB required46.1 GB available
50% VRAM used
Fit status
Runs well
Decode
36.2 tok/s
TTFT
5345 ms
Safe context
166K
Memory
23.1 GB / 46.1 GB
Memory breakdown
Weights12.8 GB
KV Cache2.5 GB
Runtime0.9 GB
Headroom6.9 GB
See how fast it feels
See how fast it feelsCodestral 21B Pruned i1 on Mac Studio M2 Ultra 64GB
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: 36.2 tok/s decode · 5.3s TTFT (warm) · 91 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 | 36.2 tok/s | 2915 ms | 166K |
| Coding | C | Runs well | 36.2 tok/s | 5345 ms | 166K |
| Agentic Coding | C | Runs well | 36.2 tok/s | 7774 ms | 166K |
| Reasoning | C | Runs well | 36.2 tok/s | 6317 ms | 166K |
| RAG | C | Runs well | 36.2 tok/s | 9718 ms | 166K |
Quantization options
How Codestral 21B Pruned i1 (21B params) fits at each quantization level on Mac Studio M2 Ultra 64GB (46.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.2 GB | Low | C42 |
Q3_K_S | 3 | 10.3 GB | Low | C43 |
NVFP4 | 4 |
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 startUpgrade options