Wave 0 Hardware Hubs
Hardware Bandwidth, Compute & UMA Ceiling Directory (28 Profiles)
Decode speed scales with memory bandwidth; prefill TTFT scales with FP16 TFLOPS. Click any node below to inspect its compatible 32-model matrix and launch commands.
Apple Silicon (Unified Memory)
| Hardware | Raw RAM/VRAM | Usable (Tuned) | Bandwidth | FP16 TFLOPS | TDP | Price |
|---|---|---|---|---|---|---|
| Apple Mac Mini M4 (16 GB) | 16 GB | 10 (11.5) GiB | 120 GB/s | 8.6 | 45W | $599 |
| Apple Mac Mini M4 (24 GB) | 24 GB | 18 (19.5) GiB | 120 GB/s | 8.6 | 45W | $799 |
| Apple Mac Mini M4 Pro (48 GB) | 48 GB | 37.44 (42.24) GiB | 273 GB/s | 18.4 | 95W | $1,799 |
| Apple Mac Mini M4 Pro (64 GB) | 64 GB | 49.92 (56.32) GiB | 273 GB/s | 18.4 | 95W | $1,999 |
| Apple MacBook Pro M3 Max (64 GB) | 64 GB | 49.92 (56.32) GiB | 400 GB/s | 28.4 | 115W | $3,499 |
| Apple MacBook Pro M3 Max (128 GB) | 128 GB | 99.84 (112.64) GiB | 400 GB/s | 28.4 | 115W | $4,299 |
| Apple Mac Studio M4 Max (64 GB) | 64 GB | 49.92 (56.32) GiB | 546 GB/s | 36.8 | 140W | $2,499 |
| Apple Mac Studio M4 Max (128 GB) | 128 GB | 99.84 (112.64) GiB | 546 GB/s | 36.8 | 140W | $3,499 |
| Apple Mac Studio M2 Ultra (192 GB) | 192 GB | 149.76 (168.96) GiB | 800 GB/s | 54.4 | 215W | $4,999 |
| Apple Mac Studio M3 Ultra (256 GB) | 256 GB | 199.68 (225.28) GiB | 819 GB/s | 58 | 230W | $5,599 |
| Apple Mac Studio M3 Ultra (512 GB) | 512 GB | 399.36 (450.56) GiB | 819 GB/s | 58 | 240W | $9,499 |
AMD Strix Halo & AI Mini-PCs
| Hardware | Raw RAM/VRAM | Usable (Tuned) | Bandwidth | FP16 TFLOPS | TDP | Price |
|---|---|---|---|---|---|---|
| Ryzen AI Max+ 395 Strix Halo (128 GB) | 128 GB | 96 (110) GiB | 256 GB/s | 59.4 | 140W | $1,899 |
| Ryzen AI Max+ 395 Strix Halo (64 GB) | 64 GB | 48 (54) GiB | 256 GB/s | 59.4 | 130W | $1,499 |
| Ryzen AI 9 HX 370 Mini-PC (64 GB) | 64 GB | 48 (54) GiB | 120 GB/s | 23 | 65W | $899 |
Desktop & Workstation GPUs
| Hardware | Raw RAM/VRAM | Usable (Tuned) | Bandwidth | FP16 TFLOPS | TDP | Price |
|---|---|---|---|---|---|---|
| NVIDIA RTX 5090 (32 GB) | 32 GB | 31 GiB | 1792 GB/s | 209.5 | 575W | $2,499 |
| NVIDIA RTX 5080 (16 GB) | 16 GB | 15 GiB | 960 GB/s | 112.6 | 360W | $1,199 |
| NVIDIA RTX 5070 Ti (16 GB) | 16 GB | 15 GiB | 896 GB/s | 87.9 | 300W | $849 |
| NVIDIA RTX 5060 Ti (16 GB) | 16 GB | 15 GiB | 448 GB/s | 47.4 | 180W | $449 |
| NVIDIA RTX 4090 (24 GB) | 24 GB | 23 GiB | 1008 GB/s | 165.2 | 450W | $1,999 |
| NVIDIA RTX 3090 (24 GB Used) | 24 GB | 23 GiB | 936 GB/s | 71 | 350W | $799 |
| 2x NVIDIA RTX 3090 Rig (48 GB) | 48 GB | 46.65 GiB | 936 GB/s | 71 | 680W | $1,950 |
| NVIDIA RTX 3060 (12 GB) | 12 GB | 11 GiB | 360 GB/s | 25.5 | 170W | $289 |
| AMD Radeon RX 7900 XTX (24 GB) | 24 GB | 23 GiB | 960 GB/s | 122.8 | 355W | $949 |
| NVIDIA RTX PRO 6000 Blackwell (96 GB) | 96 GB | 95 GiB | 1792 GB/s | 250 | 600W | $8,499 |
Cloud SXM / Inference GPUs
| Hardware | Raw RAM/VRAM | Usable (Tuned) | Bandwidth | FP16 TFLOPS | TDP | Price |
|---|---|---|---|---|---|---|
| Cloud NVIDIA H200 SXM (141 GB) | 141 GB | 140.4 GiB | 4800 GB/s | 989 | 700W | $3.49/hr |
| Cloud NVIDIA H100 SXM (80 GB) | 80 GB | 79.4 GiB | 3350 GB/s | 989 | 700W | $2.39/hr |
| Cloud NVIDIA A100 SXM (80 GB) | 80 GB | 79.4 GiB | 2039 GB/s | 312 | 400W | $1.39/hr |
| Cloud NVIDIA L40S (48 GB) | 48 GB | 47.4 GiB | 864 GB/s | 362 | 350W | $0.89/hr |