Local AI hardware
21 configurations across 10 products. Sorted by memory bandwidth, because single-stream decode speed is almost entirely a function of it. Prices are Dutch retail including 21% VAT, and every price states whether it is a published list price or our derivation.
All configurations (21)
| Configuration | Architecture | Memory | Bandwidth | Achieved | Load power | Price | GB/s per €1k |
|---|---|---|---|---|---|---|---|
| Quad RTX 5090 workstation (4x 32 GB) NVIDIA | 4× GPU | 128 GB | 7,168 GB/s | 79% | 2250 W | €13.999 * | 405 |
| Dual RTX 5090 workstation (2x 32 GB) NVIDIA | 2× GPU | 64 GB | 3,584 GB/s | 79% | 1180 W | €6.499 * | 437 |
| Dual used RTX 4090 workstation (48 GB) NVIDIA | 2× GPU | 48 GB | 2,016 GB/s | 78% | 980 W | €4.099 * | 384 |
| RTX PRO 6000 Blackwell workstation (96 GB) NVIDIA | Discrete GPU | 96 GB | 1,792 GB/s | 74% | 660 W | €11.499 * | 116 |
| RTX PRO 6000 Max-Q workstation (96 GB, 300 W) NVIDIA | Discrete GPU | 96 GB | 1,792 GB/s | 74% | 390 W | €11.999 * | 111 |
| RTX 5090 workstation (1x 32 GB) NVIDIA | Discrete GPU | 32 GB | 1,792 GB/s | 79% | 640 W | €3.699 * | 384 |
| Mac Studio M5 Ultra 512 GB Apple · announced | Unified memory | 512 GB | 1,200 GB/s | 88% | 225 W | €12.599 * | 83 |
| Mac Studio M5 Ultra 256 GB Apple | Unified memory | 256 GB | 1,200 GB/s | 88% | 215 W | €9.599 * | 110 |
| Mac Studio M5 Ultra 96 GB Apple | Unified memory | 96 GB | 1,200 GB/s | 88% | 190 W | €6.599 | 159 |
| Used RTX 4090 workstation (24 GB) NVIDIA | Discrete GPU | 24 GB | 1,008 GB/s | 80% | 520 W | €2.299 * | 351 |
| Mac Studio M3 Ultra 512 GB Apple · superseded | Unified memory | 512 GB | 819 GB/s | 71% | 205 W | €9.199 * | 63 |
| Mac Studio M3 Ultra 256 GB Apple · superseded | Unified memory | 256 GB | 819 GB/s | 71% | 195 W | €6.299 * | 92 |
| Radeon AI PRO R9700 workstation (32 GB) AMD | Discrete GPU | 32 GB | 644 GB/s | 95% | 420 W | €2.799 * | 219 |
| Mac Studio M5 Max 128 GB Apple | Unified memory | 128 GB | 614 GB/s | 88% | 122 W | €4.799 * | 112 |
| Mac Studio M5 Max 64 GB Apple | Unified memory | 64 GB | 614 GB/s | 88% | 118 W | €3.899 * | 138 |
| Mac Studio M5 Max 36 GB Apple | Unified memory | 36 GB | 614 GB/s | 88% | 105 W | €2.999 | 179 |
| NVIDIA DGX Spark 128 GB NVIDIA | Unified LPDDR | 128 GB | 273 GB/s | 67% | 170 W | €4.299 * | 43 |
| Mac mini M4 Pro 64 GB Apple | Unified memory | 64 GB | 273 GB/s | 70% | 62 W | €2.499 * | 76 |
| Framework Desktop (Ryzen AI Max+ 395, 128 GB) AMD | Unified LPDDR | 128 GB | 256 GB/s | 48% | 130 W | €2.399 * | 51 |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) AMD | Unified LPDDR | 128 GB | 256 GB/s | 42% | 135 W | €1.899 * | 56 |
| CPU-only workstation (Ryzen 9950X, 192 GB DDR5) Generic | Discrete GPU | 192 GB | 90 GB/s | 50% | 190 W | €1.699 * | 26 |
* Derived price rather than a published list price — the configuration page explains how each was arrived at. “Achieved” is the fraction of theoretical bandwidth real inference engines reach on that platform; where we have measurements it is calibrated from them, and otherwise inherited from a sibling configuration with the same silicon.
Multi-GPU builds
Discrete GPU
Unified memory
- Mac Studio M5 Ultra 512 GB 512 GB
- Mac Studio M5 Ultra 256 GB 256 GB
- Mac Studio M5 Ultra 96 GB 96 GB
- Mac Studio M3 Ultra 512 GB 512 GB
- Mac Studio M3 Ultra 256 GB 256 GB
- Mac Studio M5 Max 128 GB 128 GB
- Mac Studio M5 Max 64 GB 64 GB
- Mac Studio M5 Max 36 GB 36 GB
- Mac mini M4 Pro 64 GB 64 GB
Unified LPDDR