Inference · hardware

What each kind of machine does.

Every kind of machine that has signed a result here — by its card, or its CPU when it has none — with the fastest model it runs, the cheapest million tokens, and its parts as the machines benchmarked them. Same signed results as the provider standings, grouped by machine instead of by person.

14kinds of machine
17machines
11with a card
MachineMachines · ownersSigned runsFastest modelCheapest per MCardfp16 TFLOP/s, medianMemoryGB/s, median
nvidiaTesla T4·a5be46·e78fd22 · 292Qwen2.5 0.5B Instruct239 tok/s$0.56Qwen2.5 0.5B Instruct——
nvidiaNVIDIA GeForce RTX 3080 Ti·84bf541 · 140Llama 3.2 3B225 tok/s$1.14Llama 3.2 3B56.117.7
nvidia2× Tesla T4·e61d3b1 · 178Hunyuan MT 1.5 1.8B132 tok/s$1.03Hunyuan MT 1.5 1.8B11.53.4
appleApple M4·afbb8a1 · 142Sarvam-162 tok/s$3.16Sarvam-1——
intelIntel(R) UHD Graphics 730·9e3c791 · 163Gemma 3 270M52 tok/s$1.38Gemma 3 270M—3.6
nvidiaNVIDIA GeForce RTX 3050 Laptop GPU·53b2291 · 14Phi 4 Mini-Reasoning42 tok/s$5.19Phi 4 Mini-Reasoning——
Linux, CPU only · 4 CPUs·acbe75·e3654f2 · 285FunctionGemma 270M40 tok/s$0.76FunctionGemma 270M—2.5
amdAMD Radeon(TM) Vega 8 Mobile Graphics·d061c91 · 117Gemma 3 270M27 tok/s$6.41Gemma 3 270M—2.7
nvidiaNVIDIA GeForce GTX 1650·12954d1 · 13Sarvam-123 tok/s$7.51Sarvam-1——
appleApple M1·0548a8·0c52652 · 217Llama 3.2 1B20 tok/s$8.60Llama 3.2 1B——
Windows, CPU only · 12 CPUs·9a1db01 · 17Gemma 4 E2B15 tok/s$14Gemma 4 E2B——
intelIntel(R) Iris(R) Xe Graphics·2e5c021 · 17Qwen2.5 0.5B Instruct11 tok/s$6.75Qwen2.5 0.5B Instruct——
appleapple·d686721 · 121Llama 3.2 1B5 tok/s$6.56Llama 3.2 1B—0.1
Linux, CPU only · ? CPUs·15d9e91 · 13————

Ranked by the fastest signed result. A dash is a part no machine of that kind has benchmarked and signed yet. Prices are the network's metered rates on each machine.