Performance, cost, power and VRAM — weighted by what you actually value
Four axes for Qwen 3 8B, each normalised across the hardware shown, so you can see at a glance where a machine is strong and where it gives something up.
There is no score here until you make one. A single blended number with weights we picked would look objective while hiding an opinion you never saw — so the default is the four axes, and the ranking appears once you say what matters to you. Your weights go in the URL, so a link you share shows what you valued.
What do you actually want from this machine?
Pick a starting point — each is an opinion, named for the person it suits — or set the sliders yourself. Nothing is ranked until you do.
| Hardware | Speed | Running cost | Efficiency | Memory | Evidence |
|---|---|---|---|---|---|
0.03 | 0.40 | 0.07 | 0.05 | ||
0.09 | 0.33 | 0.05 | 0.24 | ||
0.00 | 0.58 | 0.13 | 0.02 | ||
0.12 | 0.97 | 0.78 | 0.11 | ||
0.05 | 0.83 | 0.34 | 0.05 | ||
0.25 | 0.95 | 0.70 | 0.24 | ||
0.01 | 0.75 | 0.24 | 0.03 | ||
0.12 | 0.92 | 0.54 | 0.17 | ||
0.05 | 0.83 | 0.34 | 0.05 | ||
0.25 | 0.95 | 0.70 | 0.37 | ||
11Apple M3 | 0.01 | 0.71 | 0.21 | 0.03 | |
0.12 | 0.95 | 0.63 | 0.24 | ||
0.09 | 0.89 | 0.46 | 0.17 | ||
0.03 | 0.75 | 0.24 | 0.06 | ||
0.25 | 0.96 | 0.70 | 1.00 | ||
16Apple M4 | 0.02 | 0.78 | 0.27 | 0.05 | |
0.17 | 0.99 | 0.85 | 0.24 | ||
0.13 | 0.95 | 0.63 | 0.06 | ||
0.08 | 0.91 | 0.50 | 0.03 | ||
20Apple M5 | 0.03 | 0.84 | 0.35 | 0.05 | |
0.19 | 1.00 | 1.00 | 0.24 | ||
0.14 | 0.97 | 0.70 | 0.06 | ||
0.09 | 0.93 | 0.57 | 0.11 | ||
0.36 | 0.88 | 0.43 | 1.00 | ||
25Apple M6 | 0.04 | 0.84 | 0.36 | 0.05 | |
0.12 | 0.36 | 0.06 | 0.00 | ||
0.15 | 0.30 | 0.04 | 0.01 | ||
0.20 | 0.44 | 0.07 | 0.05 | ||
0.20 | 0.35 | 0.05 | 0.05 | ||
0.07 | 0.56 | 0.12 | 0.02 | ||
0.15 | 0.26 | 0.04 | 0.03 | ||
0.22 | 0.66 | 0.17 | 0.03 | ||
0.60 | 0.76 | 0.25 | 0.14 | ||
0.14 | 0.42 | 0.07 | 0.24 | ||
0.14 | 0.47 | 0.08 | 0.24 | ||
0.75 | 0.78 | 0.27 | 0.14 | ||
1.00 | 0.60 | 0.13 | 0.14 | ||
0.17 | 0.84 | 0.35 | 0.03 | ||
0.44 | 0.63 | 0.15 | 0.08 | ||
0.44 | 0.55 | 0.11 | 0.08 | ||
0.24 | 0.27 | 0.04 | 0.05 | ||
0.29 | 0.42 | 0.07 | 0.08 | ||
0.13 | 0.30 | 0.04 | 0.01 | ||
0.17 | 0.33 | 0.05 | 0.00 | ||
0.17 | 0.25 | 0.04 | 0.00 | ||
0.23 | 0.25 | 0.03 | 0.00 | ||
0.28 | 0.32 | 0.04 | 0.00 | ||
0.33 | 0.37 | 0.06 | 0.01 | ||
0.33 | 0.37 | 0.06 | 0.01 | ||
0.19 | 0.00 | 0.00 | 0.02 | ||
0.34 | 0.39 | 0.06 | 0.03 | ||
0.36 | 0.23 | 0.03 | 0.03 | ||
0.15 | 0.62 | 0.15 | 0.00 | ||
0.14 | 0.60 | 0.13 | 0.00 | ||
0.16 | 0.44 | 0.07 | 0.02 | ||
0.16 | 0.47 | 0.08 | 0.00 | ||
0.28 | 0.62 | 0.14 | 0.01 | ||
0.28 | 0.57 | 0.12 | 0.01 | ||
0.28 | 0.40 | 0.07 | 0.01 | ||
0.36 | 0.56 | 0.11 | 0.02 | ||
0.38 | 0.52 | 0.10 | 0.02 | ||
0.24 | 0.00 | 0.00 | 0.01 | ||
0.39 | 0.53 | 0.10 | 0.02 | ||
0.58 | 0.55 | 0.11 | 0.03 | ||
0.31 | 0.78 | 0.28 | 0.02 | ||
0.31 | 0.56 | 0.12 | 0.05 | ||
0.16 | 0.60 | 0.13 | 0.00 | ||
0.23 | 0.69 | 0.19 | 0.00 | ||
0.20 | 0.00 | 0.00 | 0.00 | ||
0.23 | 0.58 | 0.13 | 0.02 | ||
0.23 | 0.58 | 0.13 | 0.00 | ||
0.33 | 0.59 | 0.13 | 0.01 | ||
0.20 | 0.78 | 0.27 | 0.00 | ||
0.42 | 0.61 | 0.14 | 0.02 | ||
0.33 | 0.88 | 0.43 | 0.01 | ||
0.45 | 0.54 | 0.11 | 0.02 | ||
0.42 | 0.87 | 0.40 | 0.02 | ||
0.70 | 0.52 | 0.10 | 0.05 | ||
0.42 | 0.87 | 0.40 | 0.03 | ||
0.48 | 0.67 | 0.17 | 0.08 | ||
0.28 | 0.55 | 0.11 | 0.03 | ||
0.28 | 0.38 | 0.06 | 0.08 | ||
0.14 | 0.81 | 0.31 | 0.02 | ||
0.33 | 0.00 | 0.00 | 0.03 | ||
0.42 | 0.78 | 0.27 | 0.05 | ||
0.58 | 0.74 | 0.23 | 0.08 | ||
0.58 | 0.74 | 0.23 | 0.13 | ||
0.42 | 0.00 | 0.00 | 0.03 | ||
0.70 | 0.49 | 0.09 | 0.17 | ||
0.70 | 0.00 | 0.00 | 0.17 | ||
0.10 | 0.15 | 0.01 | 0.00 | ||
0.10 | 0.00 | 0.00 | 0.02 | ||
0.16 | 0.11 | 0.01 | 0.01 | ||
0.22 | 0.32 | 0.04 | 0.02 | ||
0.21 | 0.27 | 0.04 | 0.02 | ||
0.28 | 0.33 | 0.04 | 0.02 | ||
0.32 | 0.34 | 0.05 | 0.03 | ||
0.12 | 0.32 | 0.04 | 0.02 | ||
0.12 | 0.37 | 0.06 | 0.00 | ||
0.24 | 0.50 | 0.10 | 0.02 | ||
0.17 | 0.27 | 0.04 | 0.01 | ||
0.24 | 0.26 | 0.04 | 0.02 |
SoC marks the 27 unified-memory systems in this ranking. Their power figure is whole-package power — CPU, GPU and memory controller on one die — where a discrete card reports board power for the card alone. Neither is the wall draw of the complete machine. They are not the same measurement, so read the ranking as a real difference in efficiency that is partly a difference in what is being counted.
How this is calculated
Decode speed comes from the same engine as the rest of the site — a three-term latency model with efficiency fitted per GPU architecture. Full methodology
each axis is min-max normalised across the hardware shown: axis = (value - min) / (max - min) 0 = weakest, 1 = strongest cost is inverted first, so higher is better on all four axes. an axis every row agrees on scores 0 -- it carries no information, and scoring it 1 would let a flat axis dominate your composite. composite = SUM(axis x your_weight) / SUM(your_weights)
Watts is the heat, and the heat is the noise — but the mapping between them is the cooler's job, and we hold no cooler data. A 450 W card in a well-designed case can be quieter than a 250 W card with a bad fan curve, so nothing here should be read as a loudness ranking. If you own a sound meter, we would like your reading.
lub porównaj w Vast.ai od $0.35/godz. (typical low · varies)
Jako partner Amazon zarabiamy na kwalifikujących się zakupach. Linki do chmurowych GPU są linkami polecającymi — możemy otrzymać prowizję bez dodatkowych kosztów dla Ciebie.