Compute-балл — для general-purpose GPU compute: CUDA/OpenCL workload, научные расчёты, симуляции, видеомонтаж и энкодинг через NVENC. Формула: FP32 + (CUDA cores ÷ 1000) × 0.4 + memory bandwidth ÷ 100.
Фильтры: VRAM 8 ГБ + NVIDIA + Pro и Server
| # | Модель | Вычисления |
|---|---|---|
| #1 | NVIDIA RTX A5000-8Q Ampere | 19.9 |
| #2 | NVIDIA CMP 170HX 8 GB Ampere | 15.1 |
| #3 | NVIDIA RTX 2000 Embedded Ada Generation Ada Lovelace | 9.3 |
| #4 | NVIDIA RTX A1000 Ampere | 5.8 |
| #5 | NVIDIA Quadro RTX 4000 Turing | 5.2 |
| #6 | NVIDIA Tesla P4 Pascal | 2.9 |
| #7 | NVIDIA Quadro P4000 Pascal | 2.8 |
| #8 | NVIDIA T1000 8 GB Turing | 2.3 |
| #9 | NVIDIA Tesla M60 Maxwell 2.0 | 1.8 |
| #10 | NVIDIA Quadro M5000 Maxwell 2.0 | 1.7 |
| #11 | NVIDIA Quadro K5200 Kepler | 1.6 |
| #12 | NVIDIA Quadro M5000M Maxwell 2.0 | 1.3 |
| #13 | NVIDIA Quadro M4000 Maxwell 2.0 | 1.3 |
| #14 | NVIDIA GRID M60-8Q Maxwell 2.0 | 1.2 |
| #15 | NVIDIA Tesla K8 Kepler | 1.1 |
| #16 | NVIDIA Tesla M10 Maxwell | 0.9 |
| #17 | NVIDIA GRID M6-8Q Maxwell 2.0 | 0.8 |
| #18 | NVIDIA Quadro K5100M Kepler | 0.7 |
| #19 | NVIDIA GRID M10-8Q Maxwell | 0.6 |
| #20 | NVIDIA GRID M40 Maxwell | 0.4 |
Подробности расчёта — страница методики.