Ornith 1.0 35B VRAM requirements
Ornith 1.0 35B has 35.1B parameters, of which about 3B are used per token; all 256 experts still have to be in memory. With an 8K-token context and one request it needs about 22.4 GB of GPU memory at Q4_K_M, 36.6 GB at FP8 and 72.6 GB at FP16/BF16. The published weights take 65.4 GB (BF16). The smallest setup here that holds it at Q4_K_M with an 8K context is the RTX 3090 (24 GB), which runs it with up to 57K tokens of context.
22.4 GB at Q4_K_M, 8K context, one request
- FP8
- 36.6 GB
- FP16 / BF16
- 72.6 GB
- Published weights
- 65.4 GB
- Smallest setup, Q4_K_M
- RTX 3090 (24 GB)
Worked example: Ornith 1.0 35B with 32K tokens
Inputs: Ornith 1.0 35B · Q4_K_M weights · 32,768 tokens of context · one request · FP16 KV cache
- Weights (35.1B at Q4_K_M)
- 19.8 GB
- KV cache (20 KB per token)
- 640 MB
- Buffers and runtime (0.5 GB + 10%)
- 2.54 GB
- Total
- 22.9 GB
The smallest setup here that holds it is the RTX 3090 (24 GB), with about 1.05 GB to spare. Change the inputs in the calculator
What makes Ornith 1.0 35B's memory use different
30 of its 40 layers keep a fixed-size linear-attention state, so the cache grows by 20 KB per token where caching every layer would take 80 KB.
Per token of context it adds 20 KB of FP16 cache; Granite 4.2 30B (29.3B), the nearest-sized model here with plain full attention, adds 256 KB, so Ornith 1.0 35B needs 8% as much.
As a mixture-of-experts model it reads about 1.69 GB of its 19.8 GB Q4_K_M weights per generated token (9%), so it writes like a much smaller model while needing memory for all of them.
How much VRAM does Ornith 1.0 35B need?
Ornith 1.0 35B needs 22.4 GB at Q4_K_M, 38.9 GB at Q8_0 and 72.6 GB at FP16 with 8,192 tokens of context and one request. Each total below is the weights plus the FP16 KV cache and the runtime overhead (0.5 GB plus 10%), and each row opens the calculator with that setting. What the GGUF names mean.
| Precision | Weights | Total | Smallest setup |
|---|---|---|---|
| As published (BF16) | 65.4 GB | 72.6 GB | A100 80GB |
| FP16 / BF16 | 65.4 GB | 72.6 GB | A100 80GB |
| FP8 / INT8 | 32.7 GB | 36.6 GB | M4 Pro Mac (64 GB, 48 GB usable) |
| INT4 (AWQ / GPTQ) | 17.4 GB | 19.8 GB | RTX 3090 (24 GB) |
| GGUF Q8_0 | 34.7 GB | 38.9 GB | M4 Pro Mac (64 GB, 48 GB usable) |
| GGUF Q6_K | 26.8 GB | 30.2 GB | RTX 5090 (32 GB) |
| GGUF Q5_K_M | 23.2 GB | 26.2 GB | RTX 5090 (32 GB) |
| GGUF Q4_K_M | 19.8 GB | 22.4 GB | RTX 3090 (24 GB) |
| GGUF IQ4_XS | 17.8 GB | 20.2 GB | RTX 3090 (24 GB) |
| GGUF Q3_K_M | 16.0 GB | 18.3 GB | RTX 3090 (24 GB) |
| GGUF IQ3_XXS | 13.5 GB | 15.5 GB | RTX 4060 Ti 16GB |
| GGUF Q2_K | 13.7 GB | 15.7 GB | RTX 4060 Ti 16GB |
Fine-tuning Ornith 1.0 35B? Ornith 1.0 35B VRAM for LoRA, QLoRA and full training.
Ornith 1.0 35B GGUF files on Hugging Face
The 8 GGUF files llama.cpp picks for Ornith 1.0 35B with -hf <repo>:<QUANT>, with the bytes the Hugging Face file list reported on 2026-09-29. Each total is that file plus the FP16 KV cache and 0.5 GB + 10%, one request. "vs estimate" compares the file with the bits-per-weight size the table above uses.
| Quant | File | Size | vs estimate | 8K | 32K | 128K |
|---|---|---|---|---|---|---|
| Q2_K | bartowski/ | 11.8 GB12,617,078,848 bytes | −14% | 13.6 GB | 14.1 GB | 16.2 GB |
| Q3_K_M | bartowski/ | 15.1 GB16,226,547,776 bytes | −5% | 17.3 GB | 17.8 GB | 19.9 GB |
| IQ4_XS | bartowski/ | 17.5 GB18,806,446,144 bytes | −1% | 19.9 GB | 20.5 GB | 22.5 GB |
| Q4_K_M | bartowski/ | 19.9 GB21,391,448,128 bytes | +1% | 22.6 GB | 23.1 GB | 25.2 GB |
| Q5_K_M | bartowski/ | 23.3 GB25,017,456,704 bytes | +1% | 26.3 GB | 26.8 GB | 28.9 GB |
| Q6_K | bartowski/ | 28.0 GB30,053,414,976 bytes | +4% | 31.5 GB | 32.0 GB | 34.0 GB |
| Q8_0 | bartowski/ | 34.4 GB36,914,690,112 bytes | −1% | 38.5 GB | 39.0 GB | 41.1 GB |
| BF16 | bartowski/ | 64.6 GB69,376,636,896 bytes | −1% | 71.7 GB | 72.3 GB | 74.3 GB |
GB here is GiB (1024³ bytes), as everywhere on this site; Hugging Face shows the Q4_K_M file as 21.4 GB, in 10⁹ bytes.
Which Ornith 1.0 35B GGUF fits a 16, 24, 32, 48 or 80 GB GPU or a Mac
From the file sizes above: the largest file that fits each machine with an 8K context and 0.5 GB left free, the longest context it then has room for, and the longest context for Q4_K_M (19.9 GB). One request, FP16 KV cache.
| Hardware | Largest GGUF, 8K | Its longest context | Q4_K_M, longest context |
|---|---|---|---|
| RTX 5060 Ti 16GB | Q2_K, 11.8 GB | 96K | No |
| RTX 4090 (24 GB) | Q4_K_M, 19.9 GB | 50K | 50K |
| RTX 5090 (32 GB) | Q6_K, 28.0 GB | 9K | 256K (full) |
| L40S (48 GB) | Q8_0, 34.4 GB | 256K (full) | 256K (full) |
| H100 SXM (80 GB) | BF16, 64.6 GB | 256K (full) | 256K (full) |
| M4 Pro Mac (64 GB, 48 GB usable) | Q8_0, 34.4 GB | 256K (full) | 256K (full) |
| M4 Max Mac (128 GB, 96 GB usable) | BF16, 64.6 GB | 256K (full) | 256K (full) |
Run Ornith 1.0 35B with llama.cpp
llama-server -hf bartowski/deepreinforce-ai_Ornith-1.0-35B-GGUF:Q4_K_M -c 51200deepreinforce-ai_Ornith-1.0-35B-Q4_K_M.gguf, 21.4 GB, from bartowski/
Ornith 1.0 35B at 8K, 32K, 128K and 256K (full) tokens of context
10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache. Each extra token of context adds 20 KB of FP16 cache per request. The last column is the smallest setup here, one card or a group, that holds it at Q4_K_M. How the KV cache works.
| Context | KV cache, FP16 | Total at Q4_K_M | Total at Q8_0 | Smallest setup, Q4_K_M |
|---|---|---|---|---|
| 8K tokens | 160 MB | 22.4 GB | 38.9 GB | RTX 3090 (24 GB) |
| 32K tokens | 640 MB | 22.9 GB | 39.4 GB | RTX 3090 (24 GB) |
| 128K tokens | 2.50 GB | 25.0 GB | 41.5 GB | RTX 5090 (32 GB) |
| 256K tokens | 5.00 GB | 27.8 GB | 44.2 GB | RTX 5090 (32 GB) |
How fast Ornith 1.0 35B writes
Tokens per second for one request with 8,192 tokens of context, estimated from memory bandwidth and the parameters read per token. A dash means it does not fit on one card. Try other settings in the speed calculator.
| Hardware | Bandwidth | Q4_K_M | FP8 |
|---|---|---|---|
| RTX 3060 12GB | 360 GB/s | — | — |
| RTX 4090 | 1,008 GB/s | 124–226 | — |
| RTX 5090 | 1,792 GB/s | 193–369 | — |
| M4 Max Mac (128 GB) | 546 GB/s | 74–129 | 48–83 |
| M3 Ultra Mac Studio (512 GB) | 819 GB/s | 104–187 | 69–121 |
| H100 SXM | 3,350 GB/s | 288–594 | 215–418 |
| H200 | 4,800 GB/s | 348–754 | 270–549 |
Longest context on one GPU
How many tokens of context Ornith 1.0 35B fits on a single card with one request, an FP16 KV cache and 0.5 GB left free. "Full" means the model's whole context window fits.
| GPU | Memory | Q4_K_M | Q8_0 | FP8 |
|---|---|---|---|---|
| RTX 3060 | 12 GB | No | No | No |
| RTX 4060 Ti 16GB | 16 GB | No | No | No |
| RTX 3090 / 4090 | 24 GB | 57K | No | No |
| RTX 5090 | 32 GB | 256K (full) | No | No |
| A100 40GB | 40 GB | 256K (full) | 36K | 141K |
| Mac, 64 GB unified memory | 48 GB | 256K (full) | 256K (full) | 256K (full) |
| L40S / RTX 6000 Ada | 48 GB | 256K (full) | 256K (full) | 256K (full) |
| A100 / H100 80GB | 80 GB | 256K (full) | 256K (full) | 256K (full) |
| Mac, 128 GB unified memory | 96 GB | 256K (full) | 256K (full) | 256K (full) |
| H200 | 141 GB | 256K (full) | 256K (full) | 256K (full) |
| B200 | 180 GB | 256K (full) | 256K (full) | 256K (full) |
Model details
- Parameters
- 35.1B (35,107,181,936)
- Experts
- 256 routed experts, all loaded
- Active per token
- 3B (estimated from the config)
- Layers
- 10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache
- Attention cache
- 2 KV heads × 256
- Context length
- 262,144 tokens
- Published weights
- 65.4 GB (BF16)
- On Hugging Face
- ornith-ai/
Ornith-1.0-35B
Why this estimate looks this way
At Q4_K_M and an 8K-token context, Ornith 1.0 35B uses 19.8 GB for weights, 160 MB for its FP16 KV cache and 2.49 GB for estimated runtime overhead, totaling 22.4 GB. The overhead is a 0.5 GB base plus 10% of weights and cache.
This is a mixture-of-experts model: all 256 experts contribute to the 35.1B parameters held in memory, even though only about 3B parameters run per token. Using only active parameters would understate VRAM.
Its attention layout matters for long context: 10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache. The FP16 cache grows by about 20 KB per additional token per request.
The architecture and published checkpoint size come from the model's config.json and weight files. These are estimates rather than measured peak memory; inference engines can reserve extra buffers or preallocate the full configured cache.
Ornith 1.0 35B next to similar-sized models
The three models here closest to it in parameter count, at Q4_K_M with 8,192 tokens of context.
| Model | Parameters | Total | KV per token | Smallest setup |
|---|---|---|---|---|
| Ornith 1.0 35B | 35.1B, 3B active | 22.4 GB | 20 KB | RTX 3090 (24 GB) |
| Ornith 1.5 35B-A3B | 36.0B, 3B active | 23.0 GB | 20 KB | RTX 3090 (24 GB) |
| Qwen3.6 35B-A3B | 36.0B, 3B active | 23.0 GB | 20 KB | RTX 3090 (24 GB) |
| K2-Horizon MoVA 36B-A4B | 37.4B, 4B active | 25.4 GB | 192 KB | RTX 5090 (32 GB) |
Ornith 1.0 35B VRAM questions
How much VRAM do I need to run Ornith 1.0 35B locally?
About 22.4 GB at Q4_K_M with an 8K context and one request, by the bits-per-weight estimate, and 72.6 GB at BF16. The Q4_K_M GGUF deepreinforce-ai_Ornith-1.0-35B-Q4_K_M.gguf in bartowski/
Can I run Ornith 1.0 35B on a 24 GB GPU?
Yes. On the RTX 4090 (24 GB), the largest listed GGUF that fits with an 8K context and 0.5 GB free is Q4_K_M (19.9 GB), with up to 50K tokens of context.
Which quantization of Ornith 1.0 35B fits in 16 GB of VRAM?
On the RTX 5060 Ti 16GB, the largest listed GGUF that fits with an 8K context and 0.5 GB free is Q2_K (11.8 GB), with up to 96K tokens of context.
Can I run Ornith 1.0 35B on a Mac?
On the M4 Pro Mac (64 GB, 48 GB usable), the largest listed GGUF that fits with an 8K context and 0.5 GB free is Q8_0 (34.4 GB), with its full 256K-token context. On the M4 Max Mac (128 GB, 96 GB usable), the largest listed GGUF that fits with an 8K context and 0.5 GB free is BF16 (64.6 GB), with its full 256K-token context. By default macOS lets the GPU use about 75% of unified memory.
How much more VRAM does Ornith 1.0 35B need for 32K or 128K tokens of context?
Its FP16 KV cache is 160 MB at 8K, 640 MB at 32K, 2.50 GB at 128K for one request, so 128K adds 2.34 GB over 8K, plus 10% overhead. A q8_0 cache
Why does Ornith 1.0 35B use more VRAM than its GGUF file size?
The file holds only the weights. At Q4_K_M with an 8K context: 19.9 GB of weights (the file's 21,391,448,128 bytes) + 160 MB of FP16 KV cache + 2.51 GB of compute buffers and runtime (0.5 GB + 10%) = 22.6 GB.
Does Ornith 1.0 35B need less VRAM because only 3B parameters are active?
No. All 256 experts stay loaded, so the Q4_K_M file takes 19.9 GB. The active parameters decide speed: each generated token reads about 1.69 GB of them.
Badge for your model card
Paste it into a Hugging Face or GitHub README; it shows the Q4_K_M total above and links to this page. Q8_0 and FP16 badges.
Numbers read from the model files on Hugging Face on . Compare it with other models in the reproducible model and GPU report.