Ornith 1.5 9B · 36 GPUs measured first-party · llama.cpp Q4_K_M · Updated October 2026
What GPU Do You Need for Ornith 1.5 9B?
We ran Ornith 1.5 9B (Q4_K_M, llama.cpp) on 36 GPUs, from budget cards to datacenter parts, and measured generation speed, prompt processing, power draw and peak VRAM on every one. The fastest was NVIDIA GeForce RTX 5090 at 225.8 tok/s.
Benchmarked weights: ornith-ai/Ornith-1.5-9B-GGUF
NVIDIA GeForce RTX 5090
225.8 tok/s on Ornith 1.5 9B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 225.8 tok/s on Ornith 1.5 9B
- 32GB, clears the Ornith 1.5 9B floor
- 575W board rating
NVIDIA GeForce RTX 4090
144.0 tok/s on Ornith 1.5 9B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 144.0 tok/s on Ornith 1.5 9B
- 24GB, clears the Ornith 1.5 9B floor
- 450W board rating
GeForce RTX 4060
45.62 tok/s on Ornith 1.5 9B, lowest launch price that still fits. Measured on our bench. 8GB of VRAM, $299 at launch.
- 45.62 tok/s on Ornith 1.5 9B
- 8GB, clears the Ornith 1.5 9B floor
- 115W board rating
NVIDIA GeForce RTX 3060 Ti
73.84 tok/s on Ornith 1.5 9B, most speed per dollar. Measured on our bench. 8GB of VRAM, $399 at launch. That is 185.1 tok/s per $1,000 of launch price.
- 73.84 tok/s on Ornith 1.5 9B
- 8GB, clears the Ornith 1.5 9B floor
- 200W board rating
What GPU Do You Need for Ornith 1.5 9B?, tok/s by GPU
Top 15 shown; 21 more cards in the full table below.
Efficiency: tok/s per 100W drawn
Top 15 shown; 20 more cards in the full table below.
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Top 15 shown; 21 more cards in the full table below.
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Ornith 1.5 9B. Measured generation speed by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 225.8 | 11863.4 | 0.74 | 304.7 W |
| NVIDIA H100 80GB HBM3 | 221.4 | 7293.2 | 0.9 | 246.2 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 201.5 | 10225.1 | 0.91 | 222.3 W |
| NVIDIA GeForce RTX 4090 | 144 | 9175.3 | 0.7 | 204.5 W |
| NVIDIA GeForce RTX 3090 Ti | 139.3 | 4777.5 | 0.46 | 302.0 W |
| GeForce RTX 5080 | 138.1 | 6571.2 | 0.77 | 179.8 W |
| NVIDIA A100 80GB SXM4 | 131.9 | 3834 | 0.69 | 192.3 W |
| NVIDIA GeForce RTX 3080 Ti | 125.2 | 4138.8 | 0.55 | 229.6 W |
| GeForce RTX 5070 Ti | 124.9 | 5356.5 | 0.76 | 165.2 W |
| NVIDIA GeForce RTX 3090 | 124.8 | 4021.1 | 0.47 | 263.9 W |
| NVIDIA L40S | 114.7 | 7543.9 | 0.56 | 205.1 W |
| GeForce RTX 4080 Super | 111.6 | 6234 | 0.68 | 163.5 W |
| NVIDIA GeForce RTX 4080 | 109.7 | 6159.9 | 0.65 | 169.3 W |
| NVIDIA GeForce RTX 3080 | 107.5 | 3646.2 | 0.52 | 207.1 W |
| NVIDIA RTX A6000 | 106.5 | 3846.2 | 0.54 | 197.9 W |
| NVIDIA GeForce RTX 4070 Ti Super | 104 | 5321.8 | 0.65 | 159.3 W |
| NVIDIA RTX A5000 | 102.9 | 3346.1 | 0.57 | 181.2 W |
| GeForce RTX 5070 | 99.41 | 4054.9 | 0.7 | 142.8 W |
| NVIDIA Titan RTX | 93.68 | 2424.6 | 0.43 | 215.7 W |
| NVIDIA RTX PRO 4000 Blackwell | 91.88 | 3742.3 | 0.88 | 104.6 W |
| NVIDIA GeForce RTX 3070 Ti | 91.43 | 2915.6 | 0.44 | 209.3 W |
| NVIDIA TITAN V | 85.44 | 2339 | 0.64 | 132.5 W |
| NVIDIA GeForce RTX 4070 | 80.73 | 3573 | 0.62 | 130.1 W |
| NVIDIA GeForce RTX 4070 Ti | 79.04 | 4461.7 | 0.57 | 139.7 W |
| NVIDIA A10G | 75.38 | 2501.3 | 0.64 | 118.1 W |
| NVIDIA GeForce RTX 3060 Ti | 73.84 | 2186.8 | 0.54 | 137.2 W |
| NVIDIA Quadro RTX 8000 | 72.54 | 2082.2 | 0.46 | 158.2 W |
| NVIDIA GeForce RTX 3070 Founders Edition | 70.34 | 2503.2 | 0.46 | 153.6 W |
| GeForce RTX 5060 Ti | 68.96 | 2652 | 0.91 | 75.4 W |
| NVIDIA RTX A4000 | 66.27 | 2202.3 | 0.54 | 123.6 W |
| NVIDIA RTX 4000 (Ada Generation) | 58.29 | 3095.8 | 0.58 | 100.4 W |
| NVIDIA GeForce RTX 3060 | 56.51 | 1693.6 | 0.42 | 133.0 W |
| NVIDIA GeForce RTX 4060 Ti 16GB | 48.42 | 2742.1 | 0.48 | 100.3 W |
| GeForce RTX 4060 | 45.62 | 1935.4 | — | — |
| NVIDIA L4 | 43.76 | 2271.9 | 0.72 | 61.0 W |
| NVIDIA T4 | 34.7 | 995.1 | 0.56 | 62.2 W |
What the numbers show. Across 36 GPUs measured on our own bench, RTX 5090 is fastest at 226 tok/s. RTX 5060 Ti is the most efficient, 69.0 tok/s at 75W (0.915 tok/W). Per dollar of launch price, RTX 5060 Ti gives the most (209.6 tok/s per $1,000). The fastest card with 16GB or less is RTX 5080 at 138 tok/s. The measured peak was ~6GB, so it runs on 6GB cards and up. That floor is a property of the model, so it applies to every GPU, measured or not.
About Ornith 1.5 9B. Ornith 1.5 9B: from ornith-ai, 9.7B parameters, on Hugging Face since August 2026, MIT licence. 6,592,357 downloads in the last 30 days and 3 community quantizations.
How it compares. H100 80GB HBM3: Ornith 1.5 9B 221.4 tok/s, gemma-2-9b 174.7 (9B), Nemotron Nano 9B v2 227.7 (9B), Qwen2.5-VL 7B Instruct 267.7 (8B), Qwen3 8B 244.2 (8B). 3 of 4 beat Ornith 1.5 9B here.
Cost on a rented GPU. 1M generated tokens of Ornith 1.5 9B: $0.18 on a RTX 3060 ($0.036/hr, 4.9 hours), $0.48 on a RTX 5090 ($0.39/hr, 74 min, 2.7x the cost).
Ornith 1.5 9B: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA GeForce RTX 3060 | $0.036/hr | 56.51 | $0.18 |
| NVIDIA GeForce RTX 3080 | $0.082/hr | 107.5 | $0.21 |
| NVIDIA GeForce RTX 3090 | $0.12/hr | 124.8 | $0.27 |
| NVIDIA GeForce RTX 3070 Founders Edition | $0.075/hr | 70.34 | $0.30 |
| NVIDIA RTX A4000 | $0.078/hr | 66.27 | $0.33 |
| GeForce RTX 5070 Ti | $0.15/hr | 124.9 | $0.33 |
| NVIDIA GeForce RTX 3080 Ti | $0.18/hr | 125.2 | $0.40 |
| GeForce RTX 5080 | $0.21/hr | 138.1 | $0.42 |
| NVIDIA GeForce RTX 4070 | $0.12/hr | 80.73 | $0.42 |
| NVIDIA RTX A5000 | $0.16/hr | 102.9 | $0.43 |
| NVIDIA GeForce RTX 4070 Ti Super | $0.16/hr | 104 | $0.43 |
| NVIDIA Titan RTX | $0.15/hr | 93.68 | $0.44 |
| NVIDIA TITAN V | $0.14/hr | 85.44 | $0.44 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 225.8 | $0.48 |
| NVIDIA GeForce RTX 4080 | $0.20/hr | 109.7 | $0.51 |
| GeForce RTX 5070 | $0.19/hr | 99.41 | $0.53 |
| NVIDIA GeForce RTX 3090 Ti | $0.27/hr | 139.3 | $0.54 |
| GeForce RTX 5060 Ti | $0.14/hr | 68.96 | $0.55 |
| GeForce RTX 4080 Super | $0.22/hr | 111.6 | $0.55 |
| NVIDIA RTX PRO 4000 Blackwell | $0.20/hr | 91.88 | $0.62 |
| NVIDIA GeForce RTX 4070 Ti | $0.18/hr | 79.04 | $0.62 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 144 | $0.65 |
| NVIDIA RTX A6000 | $0.33/hr | 106.5 | $0.86 |
| NVIDIA RTX 4000 (Ada Generation) | $0.20/hr | 58.29 | $0.95 |
| NVIDIA Quadro RTX 8000 | $0.26/hr | 72.54 | $0.98 |
| NVIDIA T4 | $0.14/hr | 34.7 | $1.09 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 201.5 | $1.48 |
| NVIDIA L40S | $0.79/hr | 114.7 | $1.91 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 131.9 | $1.99 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 221.4 | $2.68 |
| NVIDIA L4 | $0.44/hr | 43.76 | $2.79 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Ornith 1.5 9B. 30+ tok/s: 36 (RTX 5090, RTX 4090, RTX 3090 Ti). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Ornith 1.5 9B writes anything it reads the input: 11863.4 tok/s on the RTX 5090 (0.3s for a 4,000-token prompt), 9175.3 on the RTX 4090 (0.4s), 995.1 on the T4 (4.0s). Long documents and big code files feel this number more than the generation speed.
VRAM for Ornith 1.5 9B. Measured peak 5.1GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Ti (8GB). With long context: Q4_K_M 7GB (tested), Q5_K_M 9GB, Q6_K 10GB, Q8_0 12GB, BF16 23GB.
Power on Ornith 1.5 9B. Most efficient: RTX 5060 Ti, 75W, 0.30 kWh per 1M generated tokens. Hungriest: RTX 5090, 305W, 0.37 kWh. At $0.15/kWh: $0.046 per 1M generated tokens.
Our verdict
Fastest on Ornith 1.5 9B: NVIDIA GeForce RTX 5090, 225.8 tok/s. Cheapest consumer card that ran it: GeForce RTX 4060 ($299, 45.62 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3060, $0.18 per 1M generated tokens.
FAQ
- What GPU do I need to run Ornith 1.5 9B?
- About 5GB. Cheapest consumer card that ran it: GeForce RTX 4060 (8GB, 45.62 tok/s).
- How fast is Ornith 1.5 9B on the NVIDIA GeForce RTX 5090?
- 225.8 tok/s at 305W, faster than every datacenter card we ran it on.
- How much does it cost to run Ornith 1.5 9B in the cloud?
- $0.18 per 1M generated tokens on a NVIDIA GeForce RTX 3060 at $0.036/hr, cheapest of 31 rentable cards we measured.
- Can I run Ornith 1.5 9B on a 12GB, 16GB or 24GB card?
- It used 5.1GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the RTX 4090 or the RTX 3090 faster for Ornith 1.5 9B?
- The RTX 4090: 144.0 vs 124.8 tok/s, 15% faster on our bench.
- Should I buy or rent a RTX 5090 for Ornith 1.5 9B?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 4,177x 1M generated tokens of Ornith 1.5 9B. Buy only if you'll run more than that.
How we test
llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.