31 accelerators across NVIDIA, AMD, Intel, Google TPU, Groq LPU, Cerebras wafer-scale, and Apple Silicon. Each page shows what the card can actually serve, how fast, and what it costs per million tokens.
Runs 28 models on one card · 1888 tok/s on Llama 3.3 70B · $0.7358/1M tok
Runs 28 models on one card · 1127 tok/s on Llama 3.3 70B · $0.9860/1M tok
Runs 35 models on one card · 795.5 tok/s on Llama 3.3 70B · $1.40/1M tok
Runs 38 models on one card · 1326 tok/s on Llama 3.3 70B · $1.26/1M tok
Runs 40 models on one card · 1326 tok/s on Llama 3.3 70B · $1.68/1M tok
Runs 28 models on one card · 1127 tok/s on Llama 3.3 70B · $0.5916/1M tok
Runs 20 models on one card · 3493 tok/s on Llama 3.3 70B · $0.2863/1M tok
Runs 23 models on one card · 486.8 tok/s on Llama 3.3 70B · $0.7988/1M tok
Runs 9 models on one card · 1352 tok/s on Llama 3.3 70B · $0.2958/1M tok
Runs 3 models on one card · 1836 tok/s on Llama 3.3 70B · $0.2421/1M tok
Runs 38 models on one card · 878.3 tok/s on Llama 3.3 70B · $0.7906/1M tok
Runs 40 models on one card · 994.3 tok/s on Llama 3.3 70B · $0.9778/1M tok
Runs 35 models on one card · 546.9 tok/s on Llama 3.3 70B · $1.02/1M tok
Runs 9 models on one card · 676.1 tok/s on Llama 3.3 70B · $0.6573/1M tok
Runs 23 models on one card · 540.9 tok/s on Llama 3.3 70B · $0.8216/1M tok
Runs 35 models on one card · 297.0 tok/s on Llama 3.3 70B · $1.40/1M tok
Runs 3 models on one card · 4039 tok/s on Llama 3.3 70B · $0.1651/1M tok
Runs 12 models on one card · 4039 tok/s on Llama 3.3 70B · $0.1376/1M tok
Runs 9 models on one card · 2272 tok/s on Llama 3.3 70B · $0.1956/1M tok
Runs 9 models on one card · 2017 tok/s on Llama 3.3 70B · $0.1377/1M tok
Runs 3 models on one card · 6464 tok/s on Llama 3.3 70B · $0.1031/1M tok
Runs 3 models on one card
Runs 38 models on one card · 95.61 tok/s on Llama 3.3 70B
Runs 38 models on one card · 95.61 tok/s on Llama 3.3 70B
Runs 26 models on one card · 198.3 tok/s on Llama 3.3 70B
Runs 35 models on one card · 60.20 tok/s on Llama 3.3 70B
Looking for the model side of the catalog? See all models and their VRAM requirements.