Apertus-8B-Instruct · 4 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for Apertus-8B-Instruct?
Apertus-8B-Instruct on 4 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 257.7 tok/s, RTX 4060 Ti 16GB trails at 53.9 tok/s, and it peaked at 6GB of VRAM.
Benchmarked weights: unsloth/Apertus-8B-Instruct-2509-GGUF
NVIDIA GeForce RTX 5090
257.7 tok/s on Apertus-8B-Instruct, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 257.7 tok/s on Apertus-8B-Instruct
- 32GB, clears the Apertus-8B-Instruct floor
- 575W board rating
NVIDIA GeForce RTX 4090
163.1 tok/s on Apertus-8B-Instruct, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 163.1 tok/s on Apertus-8B-Instruct
- 24GB, clears the Apertus-8B-Instruct floor
- 450W board rating
NVIDIA GeForce RTX 3060
63.45 tok/s on Apertus-8B-Instruct, lowest launch price that still fits. Measured on our bench. 12GB of VRAM, $329 at launch.
- 63.45 tok/s on Apertus-8B-Instruct
- 12GB, clears the Apertus-8B-Instruct floor
- 170W board rating
NVIDIA GeForce RTX 4060 Ti 16GB
53.92 tok/s on Apertus-8B-Instruct, most speed per dollar. Measured on our bench. 16GB of VRAM, $499 at launch. That is 108.1 tok/s per $1,000 of launch price.
- 53.92 tok/s on Apertus-8B-Instruct
- 16GB, clears the Apertus-8B-Instruct floor
- 165W board rating
What GPU Do You Need for Apertus-8B-Instruct?, tok/s by GPU
Efficiency: tok/s per 100W drawn
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
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Apertus-8B-Instruct. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 257.7 | 14404.3 | 0.81 | 318.8 W |
| NVIDIA GeForce RTX 4090 | 163.1 | 11080.5 | 0.85 | 190.8 W |
| NVIDIA GeForce RTX 3060 | 63.45 | 2164.8 | 0.46 | 136.8 W |
| NVIDIA GeForce RTX 4060 Ti 16GB | 53.92 | 3486.9 | 0.49 | 109.5 W |
What the numbers show. Across 4 GPUs measured on our own bench, RTX 5090 is fastest at 258 tok/s. The slowest, RTX 4060 Ti 16GB, manages 53.9, so the spread is 4.8x from top to bottom. RTX 4090 is the most efficient, 163 tok/s at 191W. Per dollar of launch price, RTX 3060 gives the most (192.9 tok/s per $1,000). The fastest card with 16GB or less is RTX 3060 at 63.5 tok/s.
About Apertus-8B-Instruct. Apertus-8B-Instruct: from swiss-ai, 8.1B parameters, on Hugging Face since August 2025, Apache 2.0 licence. 452,881 downloads in the last 30 days.
How it compares. RTX 5090: Apertus-8B-Instruct 257.7 tok/s, Llama-3.1-8B 268.1 (8B), Llama 3 8B 259.2 (8B), Qwen3 8B 243.9 (8B), Qwen2.5-7B 284.6 (8B). 3 of 4 beat Apertus-8B-Instruct here.
Cost on a rented GPU. 1M generated tokens of Apertus-8B-Instruct: $0.16 on a RTX 3060 ($0.036/hr, 4.4 hours), $0.42 on a RTX 5090 ($0.39/hr, 65 min, 2.7x the cost).
Apertus-8B-Instruct: 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 | 63.45 | $0.16 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 257.7 | $0.42 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 163.1 | $0.57 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Apertus-8B-Instruct. 30+ tok/s: 4 (RTX 5090, RTX 4090, RTX 3060). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Apertus-8B-Instruct writes anything it reads the input: 14404.3 tok/s on the RTX 5090 (0.3s for a 4,000-token prompt), 11080.5 on the RTX 4090 (0.4s), 2164.8 on the RTX 3060 (1.8s). Long documents and big code files feel this number more than the generation speed.
VRAM for Apertus-8B-Instruct. Measured peak 4.9GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3060 (12GB). With long context: Q4_K_M 7GB (tested), Q2_K 5GB, Q3_K_M 6GB, Q5_K_M 9GB, Q6_K 10GB.
Power on Apertus-8B-Instruct. Most efficient: RTX 4090, 191W, 0.32 kWh per 1M generated tokens. Hungriest: RTX 5090, 319W, 0.34 kWh. At $0.15/kWh: $0.049 per 1M generated tokens.
Our verdict
Fastest on Apertus-8B-Instruct: NVIDIA GeForce RTX 5090, 257.7 tok/s. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3060 ($329, 63.45 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3060, $0.16 per 1M generated tokens.
FAQ
- What GPU do I need to run Apertus-8B-Instruct?
- About 5GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3060 (12GB, 63.45 tok/s).
- How fast is Apertus-8B-Instruct on the NVIDIA GeForce RTX 5090?
- 257.7 tok/s at 319W, faster than every datacenter card we ran it on.
- How much does it cost to run Apertus-8B-Instruct in the cloud?
- $0.16 per 1M generated tokens on a NVIDIA GeForce RTX 3060 at $0.036/hr, cheapest of 3 rentable cards we measured.
- Can I run Apertus-8B-Instruct on a 12GB, 16GB or 24GB card?
- It used 4.9GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the RTX 5090 or the RTX 4090 faster for Apertus-8B-Instruct?
- The RTX 5090: 257.7 vs 163.1 tok/s, 58% faster on our bench.
- Should I buy or rent a RTX 5090 for Apertus-8B-Instruct?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 4,767x 1M generated tokens of Apertus-8B-Instruct. 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.