NVIDIA · 120B total / 12B active · MoE
Nemotron 3 Super 120B-A12B needs about 78 GB at Q4_K_M. 10 of the 17 priced rigs we track can hold it. Here's the cheapest, the best value and the fastest. Prices & stock checked 2026-10-23. Prices change daily and some retailers block our checker, so always confirm on the store page.
Ryzen AI Max+ 395, 128GB
Cheapest that fits. Lowest price we've seen on any tracked rig that holds Nemotron 3 Super 120B-A12B at 4-bit (78 of 96 GB usable).
Ryzen AI Max+ 395, 128GB
Best value. Lowest price per token/sec among complete systems where it fits comfortably: about $140 per tok/s at 4-bit. Same machine as the cheapest pick.
96GB card
Fastest. Most memory bandwidth of the rigs that fit it (1,792 GB/s), so the quickest decode: ~135 tok/s at 4-bit.
GPU price is card only: you still need a host PC.
Check priceUse Q4_K_M on most rigs: it needs ~78 GB. Q8_0 would need ~135 GB, which doesn't fit comfortably at a usable speed on anything we track.
| Machine | Price | 4-bit fit & speed | Suggested quant | $ per tok/s |
|---|---|---|---|---|
| Framework Desktop DIYStrix Halo | $3,449 | Fits~25 tok/s | Q4_K_M | $140 |
| GMKtec EVO-X2Strix Halo | $3,500 | Fits~25 tok/s | Q4_K_M | $142 |
| Minisforum MS-S1 MaxStrix Halo | $3,919 | Fits~25 tok/s | Q4_K_M | $159 |
| Beelink GTR9 ProStrix Halo | $4,349 | Fits~25 tok/s | Q4_K_M | $177 |
| Lenovo ThinkStation PGXDGX Spark / GB10 | $5,859 | Fits~26 tok/s | Q4_K_M | $223 |
| ASUS Ascent GX10DGX Spark / GB10 | $5,999 | Fits~26 tok/s | Q4_K_M | $228 |
| NVIDIA DGX SparkDGX Spark / GB10 | $6,950 | Fits~26 tok/s | Q4_K_M | $264 |
| HP ZGX Nano G1nDGX Spark / GB10 | $7,699 | Fits~26 tok/s | Q4_K_M | $293 |
| Gigabyte AI TOP ATOMDGX Spark / GB10 | $8,499 | Fits~26 tok/s | Q4_K_M | $323 |
| PNY NVIDIA RTX PRO 6000 BlackwellRTX PRO 6000 (card only, needs a host PC) | $17,899 | Fits~135 tok/s | Q4_K_M | $132 |
Sorted by lowest price. Speeds are estimated single-user decode at ~8K context. Full fit table, including rigs that don't fit: Nemotron 3 Super 120B-A12B memory requirements. How we estimate
About 78 GB at Q4_K_M and 135 GB at Q8_0, including runtime overhead and ~8K tokens of context for one user. Longer context needs more.
Of the rigs we track, the Framework Desktop DIY at $3,449, estimated ~25 tok/s at 4-bit. Prices checked 2026-10-23.
Q4_K_M is the default: about 78 GB. Q8_0 needs about 135 GB; use it only if you have the room and still get usable speed.
It has 12B active parameters per token, so speed is set by memory bandwidth: from ~25 tok/s on the slowest rig that fits to ~135 tok/s on the PNY NVIDIA RTX PRO 6000 Blackwell. Estimates, not benchmarks.
One short email a week: what got cheaper, what's back in stock, and which new open models your machine can run. Unsubscribe anytime.
