Independent LLM benchmarks

💰

LLM Economics

Compare hardware, models and inference setups by operating cost, token output and total cost of ownership.

💸 Operating cost🖥️ TCO⚡ Tokens / euro🔋 Tokens / kWh

🎮 GPUs 🧠 Models ⚖️ Compare 📚 Methodology

Settings are stored in your browser and shared via the URL. Power draw estimated from GPU TDP + CPU (idle + 15 %) + board. Methodology →

GPU economics

26 GPU configurations

Aggregated per configuration (incl. GPU count) – economics computed per run, then AVG/MIN/MAX. Number after the GPU = runs · models.

GPU configurations chart

Bars per configuration – switch between cost per 1M tokens, tokens per euro and throughput, plus best values (top model) and AVG. Below each GPU is the model that reaches the value. Cheaper/faster = longer bar.

List of graphics cards

#	GPU configuration	Gen tok/s i	Cost/h	EUR / 1M tok i	Tok/€	Tok/kWh	TCO i
1	NVIDIA GB10 (DGX Spark)i128 GB VRAM · EUR 4,000 · 140 W GPU-TDP	903.3 tok/s gemma-4-E2B-it	EUR 0.042	EUR 0.013	77.43M	23.23M	EUR 736
2	NVIDIA Tesla V100 PCIe 32GBi32 GB VRAM · EUR 600 · 250 W GPU-TDP	361.8 tok/s gemma-4-E2B-it	EUR 0.075	EUR 0.058	17.37M	5.21M	EUR 1,314
3	AMD Radeon RX 7900 XTXi24 GB VRAM · EUR 1,049 · 355 W GPU-TDP	495.5 tok/s Qwen3-0.6B	EUR 0.11	EUR 0.061	16.29M	4.89M	EUR 1,918
4	AMD Radeon PRO W7800 48GBi48 GB VRAM · EUR 2,808 · 260 W GPU-TDP	702.9 tok/s gemma-4-E2B-it	EUR 0.081	EUR 0.032	31.24M	9.37M	EUR 1,419
5	AMD Radeon 8060S GraphicsiEUR 3,000 · 120 W GPU-TDP	326.9 tok/s Nemotron-Cascade-2-30B-A3B	EUR 0.036	EUR 0.031	32.69M	9.81M	EUR 631
6	NVIDIA RTX A6000i48 GB VRAM · EUR 3,000 · 300 W GPU-TDP	1,122.5 tok/s gemma-4-E2B-it	EUR 0.11	EUR 0.028	36.31M	10.89M	EUR 1,950
7	AMD Radeon AI PRO R9700i32 GB VRAM · EUR 1,400 · 300 W GPU-TDP	561.4 tok/s Qwen3-0.6B	EUR 0.090	EUR 0.045	22.46M	6.74M	EUR 1,577
8	NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editioni96 GB VRAM · EUR 13,000 · 300 W GPU-TDP	1,100.7 tok/s Ornith-1.0-9B	EUR 0.091	EUR 0.023	44.03M	13.21M	EUR 1,577
9	3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editioni288 GB VRAM · EUR 39,000 · 900 W GPU-TDP	2,143.6 tok/s gemma-4-E2B-it	EUR 0.29	EUR 0.038	26.18M	7.85M	EUR 5,033
10	NVIDIA GeForce RTX 4090i24 GB VRAM · EUR 2,000 · 450 W GPU-TDP	400.4 tok/s Qwen3-Coder-30B-A3B-Instruct	EUR 0.14	EUR 0.094	10.68M	3.20M	EUR 2,365
11	2× NVIDIA RTX A6000i96 GB VRAM · EUR 6,000 · 600 W GPU-TDP	1,568.0 tok/s gemma-4-E2B-it	EUR 0.20	EUR 0.036	28.04M	8.41M	EUR 3,154
12	AMD Radeon PRO W7900 Dual Sloti48 GB VRAM · EUR 3,911 · 295 W GPU-TDP	343.9 tok/s gpt-oss-20b	EUR 0.091	EUR 0.074	13.53M	4.06M	EUR 1,603
13	2× AMD Radeon PRO W7800 48GBi96 GB VRAM · EUR 5,616 · 520 W GPU-TDP	323.1 tok/s Meta-Llama-3.1-8B-Instruct	EUR 0.16	EUR 0.14	7.32M	2.19M	EUR 2,786
14	NVIDIA Tesla P100 PCIe 16GBi16 GB VRAM · EUR 70 · 250 W GPU-TDP	111.1 tok/s gpt-oss-20b	EUR 0.075	EUR 0.19	5.33M	1.60M	EUR 1,314
15	2× AMD Radeon PRO W7900 Dual Sloti96 GB VRAM · EUR 7,822 · 590 W GPU-TDP	344.1 tok/s gpt-oss-20b	EUR 0.18	EUR 0.15	6.88M	2.06M	EUR 3,154
16	NVIDIA GeForce RTX 5070 Tii16 GB VRAM · EUR 994 · 300 W GPU-TDP	1,744.0 tok/s gemma-4-E2B-it	EUR 0.093	EUR 0.015	67.51M	20.25M	EUR 1,577
17	2× NVIDIA GeForce RTX 2060i12 GB VRAM · EUR 220 · 320 W GPU-TDP	259.9 tok/s Qwen2.5-3B-Instruct	EUR 0.10	EUR 0.12	8.69M	2.61M	EUR 1,682
18	4× AMD Radeon AI PRO R9700i128 GB VRAM · EUR 5,600 · 1,200 W GPU-TDP	290.3 tok/s Nemotron-3.5-Lightning-30B-A3B	EUR 0.38	EUR 0.36	2.74M	822.17K	EUR 6,680
19	2× AMD Radeon RX 7900 XTXi48 GB VRAM · EUR 2,098 · 710 W GPU-TDP	717.6 tok/s Llama-3.2-1B-Instruct	EUR 0.22	EUR 0.084	11.96M	3.59M	EUR 3,784
20	Intel Arc Pro B70i32 GB VRAM · EUR 1,258 · 230 W GPU-TDP	826.2 tok/s gemma-4-E2B-it	EUR 0.069	EUR 0.023	43.11M	12.93M	EUR 1,209
21	NVIDIA GeForce RTX 3090 Tii24 GB VRAM · EUR 999 · 450 W GPU-TDP	1,492.8 tok/s gemma-4-E2B-it	EUR 0.14	EUR 0.027	37.53M	11.26M	EUR 2,365
22	3× AMD Radeon AI PRO R9700i96 GB VRAM · EUR 4,200 · 900 W GPU-TDP	667.0 tok/s Nemotron-3-Nano-4B	EUR 0.29	EUR 0.12	8.24M	2.47M	EUR 5,104
23	NVIDIA GeForce RTX 5090i32 GB VRAM · EUR 3,300 · 575 W GPU-TDP	2,491.2 tok/s gemma-4-E2B-it	EUR 0.19	EUR 0.021	48.10M	14.43M	EUR 3,267
24	NVIDIA RTX PRO 6000 Blackwell Workstation Editioni96 GB VRAM · EUR 13,000 · 600 W GPU-TDP	2,414.4 tok/s gemma-4-E2B-it	EUR 0.19	EUR 0.022	45.01M	13.50M	EUR 3,384
25	AMD Liverpool (PlayStation 4)ino price · –	43.2 tok/s Llama-3.2-1B-Instruct	–	–	–	–	–
26	AMD Oberon (PlayStation 5)ino price · –	302.6 tok/s Qwen3-0.6B	–	–	–	–	–

MIN = lowest cost (highest speed / lowest draw), MAX = highest cost. Every cost metric is bound to a real run (computed per run, then aggregated). Prices = GPU acquisition (unit price × count).

GPUby GPU

NVIDIA GeForce RTX 5090 2.491,2 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 2.414,4 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 2.143,6 tok/sNVIDIA GeForce RTX 5070 Ti 1.744,0 tok/sNVIDIA RTX A6000 1.568,0 tok/sNVIDIA GeForce RTX 3090 Ti 1.492,8 tok/sNVIDIA GB10 (DGX Spark) 903,3 tok/sIntel Arc Pro B70 826,2 tok/sAMD Radeon RX 7900 XTX 717,6 tok/sAMD Radeon PRO W7800 48GB 702,9 tok/s

Speed vs. tokens per euro

Speed vs. tokens per euro – bubble size = VRAM. Top-right = fast and economical.

Over 2 years projected

Projection per GPU configuration for the selected period: electricity cost and generated output tokens (per-run AVG × effective operating hours). Adjust period/utilization above.

Cost vs. generated tokens (2 years)

Cost over 2 years

Generated tokens over 2 years

Cost & tokens grow linearly with runtime. Power = GPU TDP + CPU (idle + 15 %) + board (estimated). Charts show the top 12 most economical configurations.