RTX 3090 24GBコスパ重視24 GB VRAM (+4)936 GB/s (+576)
B
Raises estimated decode speed by about 133%.52.5 tok/sデコード
Raises estimated decode speed by about 133%.
〜$1,499 MSRP
Can Codestral 22B run on RTX 4000 Ada 20GB?
YES — With Offload
B59Good
Estimated from fit model
Codestral 22B needs ~19.1 GB VRAM. RTX 4000 Ada 20GB has 20.0 GB. With Q4_K_M quantization, expect ~23 tok/s.
Runtime: OllamaCapacity: OffloadBandwidth: LowStack: BasicBottleneck: 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) — 19.1 GB, 22.5 tok/s, Runs with offload
19.1 GB required20.0 GB available
96% VRAM used
Fit status
Runs with offload
Decode
22.5 tok/s
TTFT
8607 ms
Safe context
22K
Memory
19.1 GB / 20.0 GB
Memory breakdown
Weights13.4 GB
KV Cache2.4 GB
Runtime1.2 GB
Headroom2.0 GB
See how fast it feels
See how fast it feelsCodestral 22B on RTX 4000 Ada 20GB
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: 22.5 tok/s decode · 8.6s TTFT (warm) · 56 tok/s prefill
What limits this setup
This setup is broadly balanced for this model.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Best improvement path
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | B | Tight fit | 22.5 tok/s | 4695 ms | 22K |
| Coding | B | Runs with offload | 22.5 tok/s | 8607 ms | 22K |
| Agentic Coding | C | Runs with offload (needs ~0.9 GB host RAM) | 14.5 tok/s | 19443 ms | 22K |
| Reasoning | B | Runs with offload | 22.5 tok/s | 10172 ms | 22K |
| RAG | C | Runs with offload (needs ~0.9 GB host RAM) | 14.5 tok/s | 24303 ms | 22K |
Inference speed
Codestral 22B inference speed — tokens per second by GPU & Mac
Estimated decode speed (tokens/sec) for Codestral 22B at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~96 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 | 96.2 | Fits |
| RTX 4090 24GB | 24 GB | Q4_K_M | 61.4 | Fits |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 55.4 | Fits |
| RTX 3090 24GB | 24 GB | Q4_K_M | 52.5 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 44.6 | Fits |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 37.4 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 37.4 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 37.2 | Fits |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 35.2 | Fits |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 26.5 | Heavy offload |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 23.6 | Fits |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 19.2 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 17.6 | Fits |
| RTX 4070 12GB | 12 GB | Q4_K_M | 9.4 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 5.9 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.4 | Too big |
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.
Quantization options
How Codestral 22B (22B params) fits at each quantization level on RTX 4000 Ada 20GB (20.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | B60 |
Q3_K_S | 3 | 10.8 GB | Low | B61 |
NVFP4 | 4 | 12.3 GB | Medium | B60 |
Q4_K_MBest for your GPU | 4 | 13.4 GB | Medium | B60 |
Q5_K_M | 5 | 15.8 GB | High | F0 |
Q6_K | 6 | 18.0 GB | High | F0 |
Q8_0 | 8 | 23.5 GB | Very High | F0 |
F16 | 16 | 45.1 GB | Maximum | F0 |
Get started
Copy-paste commands to run Codestral 22B on your machine.
Run
ollama run codestralアップグレードオプション
Codestral 22Bを快適に動かすハードウェア
RTX 3090 24GBコスパ重視RTX 4090 24GBコスパ最良24 GB VRAM (+4)1008 GB/s (+648)
B
Raises estimated decode speed by about 173%.61.4 tok/sデコード
Raises estimated decode speed by about 173%.
〜$1,599 MSRP
RTX PRO 4000 Blackwell 24GBNVIDIAアップグレード24 GB VRAM (+4)672 GB/s (+312)
B
Raises estimated decode speed by about 101%.45.2 tok/sデコード
Raises estimated decode speed by about 101%.
〜$1,599 MSRP
Frequently asked questions
Can RTX 4000 Ada 20GB run Codestral 22B?
Yes, RTX 4000 Ada 20GB can run Codestral 22B with a B grade (Runs with offload). Expected decode speed: 22.5 tok/s.
How much VRAM does Codestral 22B need?
Codestral 22B (22B parameters) requires approximately 19.1 GB of memory with Q4_K_M quantization.
What is the best quantization for Codestral 22B?
The recommended quantization for Codestral 22B is Q4_K_M, which balances quality and memory efficiency.
What speed will Codestral 22B run at on RTX 4000 Ada 20GB?
On RTX 4000 Ada 20GB, Codestral 22B achieves approximately 22.5 tokens per second decode speed with a time-to-first-token of 8607ms using Q4_K_M quantization.
Can RTX 4000 Ada 20GB run Codestral 22B for coding?
For coding workloads, Codestral 22B on RTX 4000 Ada 20GB receives a B grade with 22.5 tok/s and 22K context.
What context window can Codestral 22B use on RTX 4000 Ada 20GB?
On RTX 4000 Ada 20GB, Codestral 22B can safely use up to 22K tokens of context. The model's official context limit is 33K, but available memory constrains the safe maximum.
What should I upgrade first if Codestral 22B feels slow on RTX 4000 Ada 20GB?
Buy headroom, not only minimum fit. A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Embed this result▼
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/codestral-22b-on-rtx-4000-ada-20gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: