RTX 3060 12GBGünstige Wahl12 GB VRAM (+4)360 GB/s (+136)
A
Adds memory headroom for longer context windows and future model growth.28.4 tok/s Dekodierung
Adds memory headroom for longer context windows and future model growth.
ca. $329 MSRP
Can Ornith 1.0 9B run on RTX 5050 8GB?
BARELY — Tight on Memory
B66Good
Estimated from fit model
Ornith 1.0 9B needs ~9.4 GB VRAM. RTX 5050 8GB has 8.0 GB. With Q4_K_M quantization, expect ~14 tok/s.
Runtime: llama.cppCapacity: OffloadBandwidth: Very lowStack: StandardBottleneck: Host offload
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) — 7.9 GB, 26.7 tok/s, Runs with offload
7.9 GB required8.0 GB available
99% VRAM used
Fit status
Runs with offload
Decode
26.7 tok/s
TTFT
7242 ms
Safe context
19K
Memory
7.9 GB / 8.0 GB
Memory breakdown
Weights5.7 GB
KV Cache0.5 GB
Runtime0.9 GB
Headroom0.8 GB
See how fast it feels
See how fast it feelsOrnith 1.0 9B on RTX 5050 8GB
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: 26.7 tok/s decode · 7.2s TTFT (warm) · 67 tok/s prefill
What limits this setup
It fits through host-memory offload, and offload is the main reason performance drops.
CPU or host-memory offload is active
About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.
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
Remove offload with more accelerator memory
Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Increase host RAM if you keep offloading
This setup may need roughly {ram} GB of extra host RAM just for the offloaded portion, before OS and other tools.
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | A | Runs with offload | 17.2 tok/s | 6150 ms | 5K |
| Coding | B | Very compromised | 13.7 tok/s | 14106 ms | 5K |
| Agentic Coding | F | Too heavy | 9.3 tok/s | 30172 ms | 5K |
| Reasoning | B | Very compromised | 13.7 tok/s | 16671 ms | 5K |
| RAG | F | Too heavy | 9.3 tok/s | 37715 ms | 5K |
Inference speed
Ornith 1.0 9B inference speed — tokens per second by GPU & Mac
Estimated decode speed (tokens/sec) for Ornith 1.0 9B at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~132 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 | 131.6 | Fits |
| RTX 4090 24GB | 24 GB | Q4_K_M | 131.6 | Fits |
| RTX 3090 24GB | 24 GB | Q4_K_M | 122.8 | Fits |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 114.5 | Fits |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 84.9 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 79.1 | Fits |
| RTX 4070 12GB | 12 GB | Q4_K_M | 70.9 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 70.3 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 65.9 | Fits |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 62.5 | Fits |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 53.3 | Fits |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 45.0 | Fits |
| RTX 3060 12GB | 12 GB | Q4_K_M | 44.5 | Fits |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 41.2 | Fits |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 36.3 | Fits |
| RTX 4060 8GB | 8 GB | Q4_K_M | 26.4 | Offloads |
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 Ornith 1.0 9B (9.399999618530273B params) fits at each quantization level on RTX 5050 8GB (8.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q1_0_G128 | 1.125 | 1.4 GB | Very Low | A78 |
Q2_0_G128 | 1.71 | 2.5 GB | Low | A81 |
Q2_K | 2 | 3.7 GB | Low | A81 |
Q3_K_S | 3 | 4.6 GB | Low | A81 |
NVFP4Best for your GPU | 4 | 5.3 GB | Medium | A81 |
Q4_K_M | 4 | 5.7 GB | Medium | F0 |
Q5_K_M | 5 | 6.8 GB | High | F0 |
Q6_K | 6 | 7.7 GB | High | F0 |
Q8_0 | 8 | 10.1 GB | Very High | F0 |
F16 | 16 | 19.3 GB | Maximum | F0 |
Get started
Copy-paste commands to run Ornith 1.0 9B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "deepreinforce-ai/Ornith-1.0-9B" \
--hf-file "Ornith-1.0-9B-Q4_K_M.gguf" \
-c 4096 -ngl 99Upgrade-Optionen
Hardware, die Ornith 1.0 9B gut ausführt
RTX 3060 12GBGünstige WahlRTX 5060 Ti 16GBBestes Preis-Leistungs-Verhältnis16 GB VRAM (+8)448 GB/s (+224)
A
Raises estimated decode speed by about 48%.39.5 tok/s Dekodierung
Raises estimated decode speed by about 48%.
Adds memory headroom for longer context windows and future model growth.
ca. $449 MSRP
RTX 5070 12GBNVIDIA-Upgrade12 GB VRAM (+4)672 GB/s (+448)
A
Raises estimated decode speed by about 125%.60.2 tok/s Dekodierung
Raises estimated decode speed by about 125%.
Adds memory headroom for longer context windows and future model growth.
ca. $549 MSRP
Frequently asked questions
Can RTX 5050 8GB run Ornith 1.0 9B?
Yes, RTX 5050 8GB can run Ornith 1.0 9B with a B grade (Very compromised). Expected decode speed: 13.7 tok/s.
How much VRAM does Ornith 1.0 9B need?
Ornith 1.0 9B (9.399999618530273B parameters) requires approximately 9.4 GB of memory with Q4_K_M quantization.
What is the best quantization for Ornith 1.0 9B?
The recommended quantization for Ornith 1.0 9B is Q4_K_M, which balances quality and memory efficiency.
What speed will Ornith 1.0 9B run at on RTX 5050 8GB?
On RTX 5050 8GB, Ornith 1.0 9B achieves approximately 13.7 tokens per second decode speed with a time-to-first-token of 14106ms using Q4_K_M quantization.
Can RTX 5050 8GB run Ornith 1.0 9B for coding?
For coding workloads, Ornith 1.0 9B on RTX 5050 8GB receives a B grade with 13.7 tok/s and 5K context.
What context window can Ornith 1.0 9B use on RTX 5050 8GB?
On RTX 5050 8GB, Ornith 1.0 9B can safely use up to 5K tokens of context. The model's official context limit is 262K, but available memory constrains the safe maximum.
What should I upgrade first if Ornith 1.0 9B feels slow on RTX 5050 8GB?
Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Embed this result▼
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/ornith-1.0-9b-on-rtx-5050-8gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: