Strix HaloAvailable
GMKtec EVO-X2
Ryzen AI Max+ 395, 64GB / 1TB
$2,200at GMKtec store
Cheapest that fits. Lowest price we've seen on any tracked rig that holds Gemma 4 31B at 4-bit (21 of 48 GB usable).
Fits~11 tok/sQ4_K_M
Check priceGoogle DeepMind · 31B total / 31B active · dense
Best hardware to run Gemma 4 31B locally
Gemma 4 31B needs about 21 GB at Q4_K_M. 16 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.
Our picks
Mac StudioAvailable
Mac Studio M5 Max
18-core CPU / 40-core GPU, 48GB
$3,099at Apple Store
Best value. Lowest price per token/sec among complete systems where it fits comfortably: about $109 per tok/s at 4-bit.
Fits~28 tok/sQ4_K_M
Check price
RTX 5090In stock
NVIDIA GeForce RTX 5090 Founders Edition
32GB card
$8,500at Newegg
Fastest. Most memory bandwidth of the rigs that fit it (1,792 GB/s), so the quickest decode: ~73 tok/s at 4-bit.
Fits~73 tok/sQ4_K_M
GPU price is card only: you still need a host PC.
Check priceWhich quant to use
Use Q4_K_M on most rigs: it needs ~21 GB. Q8_0 (~36 GB) is a bit more accurate and fits comfortably on 2 of these rigs, e.g. the Mac Studio M5 Ultra.
Every rig that fits it
| Machine | Price | 4-bit fit & speed | Suggested quant | $ per tok/s |
|---|---|---|---|---|
| GMKtec EVO-X2Strix Halo | $2,200 | Fits~11 tok/s | Q4_K_M | $202 |
| Mac Studio M5 MaxMac Studio | $2,499 | Fits~21 tok/s | Q4_K_M | $116 |
| Mac Studio M5 MaxMac Studio | $3,099 | Fits~28 tok/s | Q4_K_M | $109 |
| Framework Desktop DIYStrix Halo | $3,449 | Fits~11 tok/s | Q4_K_M | $317 |
| GMKtec EVO-X2Strix Halo | $3,500 | Fits~11 tok/s | Q4_K_M | $322 |
| Minisforum MS-S1 MaxStrix Halo | $3,919 | Fits~11 tok/s | Q4_K_M | $360 |
| Beelink GTR9 ProStrix Halo | $4,349 | Fits~11 tok/s | Q4_K_M | $400 |
| Gigabyte GeForce RTX 4090 Windforce V2RTX 4090 (card only, needs a host PC) | $4,899 | Tight fit~43 tok/s | Q4_K_M | $115 |
| Mac Studio M5 UltraMac Studio | $5,499 | Fits~49 tok/s | Q8_0 | $112 |
| Lenovo ThinkStation PGXDGX Spark / GB10 | $5,859 | Fits~11 tok/s | Q4_K_M | $524 |
| ASUS Ascent GX10DGX Spark / GB10 | $5,999 | Fits~11 tok/s | Q4_K_M | $536 |
| NVIDIA DGX SparkDGX Spark / GB10 | $6,950 | Fits~11 tok/s | Q4_K_M | $621 |
| HP ZGX Nano G1nDGX Spark / GB10 | $7,699 | Fits~11 tok/s | Q4_K_M | $688 |
| Gigabyte AI TOP ATOMDGX Spark / GB10 | $8,499 | Fits~11 tok/s | Q4_K_M | $760 |
| NVIDIA GeForce RTX 5090 Founders EditionRTX 5090 (card only, needs a host PC) | $8,500 | Fits~73 tok/s | Q4_K_M | $116 |
| PNY NVIDIA RTX PRO 6000 BlackwellRTX PRO 6000 (card only, needs a host PC) | $17,899 | Fits~73 tok/s | Q8_0 | $244 |
Sorted by lowest price. Speeds are estimated single-user decode at ~8K context. Full fit table, including rigs that don't fit: Gemma 4 31B memory requirements. How we estimate
FAQ
How much memory does Gemma 4 31B need?
About 21 GB at Q4_K_M and 36 GB at Q8_0, including runtime overhead and ~8K tokens of context for one user. Longer context needs more.
What's the cheapest hardware to run Gemma 4 31B locally?
Of the rigs we track, the GMKtec EVO-X2 at $2,200, estimated ~11 tok/s at 4-bit. Prices checked 2026-10-23.
Which quant should I use for Gemma 4 31B?
Q4_K_M is the default: about 21 GB. Q8_0 needs about 36 GB; use it only if you have the room and still get usable speed.
How fast does Gemma 4 31B run?
It has 31B active parameters per token, so speed is set by memory bandwidth: from ~11 tok/s on the slowest rig that fits to ~73 tok/s on the NVIDIA GeForce RTX 5090 Founders Edition. Estimates, not benchmarks.