Hardware requirements
What hardware do I need to run gpt-oss 20B?
20.9B MoE · 3.6B active
128K context
Apache 2.0
At MXFP4 the weights are 10.8 GB; with an 8K KV cache and runtime overhead that is 11.6 GB, so it wants a 16 GB card to run with headroom — in practice 16 GB VRAM: GeForce RTX 5080 and 17 others. Every figure below is computed from the model’s published shape against each device’s usable memory; tokens per second are estimates. The method.
Minimum practical
16 GB VRAM
GeForce RTX 5080 — MXFP4, 11.6 GB of 14.4 GB, ~120 tok/s, up to 126K context.
the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running
Recommended
16 GB VRAM
GeForce RTX 5080 — MXFP4, 11.6 GB of 14.4 GB, ~120 tok/s, up to 126K context.
runs the recommended quantisation with headroom at 8K context
High quality
Nothing in the catalogue
No single device we list clears this bar — this is a multi-GPU or server-class model.
Long context
20 GB VRAM
Radeon RX 7900 XT — MXFP4, 14.4 GB of 18.4 GB, ~100 tok/s at 128K.
128K tokens (or the model’s maximum) at the recommended quantisation
Apple Silicon
24 GB unified
M2 · 24 GB — MXFP4, 11.6 GB of 18.0 GB, ~14 tok/s, up to 128K context.
smallest Mac that runs the recommended quantisation with headroom
CPU only
32 GB RAM
CPU only · DDR4 dual-channel — MXFP4, 11.6 GB of 25.6 GB, ~5.1 tok/s, up to 128K context.
no GPU — weights stream from system RAM
How much VRAM at each quantisation
| Quant | Weights | +KV 8K | +KV 32K | +KV 128K | Total @ 8K | Needs | Quality |
|---|---|---|---|---|---|---|---|
| MXFP4 | 10.8 GB | 0.19 GB | 0.75 GB | 3.0 GB | 11.6 GB | 16 GB card | Reference |
Totals add 0.6 GB runtime overhead and an f16 KV cache; "needs" is the smallest discrete card class whose usable memory (after display and driver reserve) leaves 15% headroom. A q8_0 cache roughly halves the KV columns.
NVIDIA · GeForce
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| GeForce RTX 5090 | 32 GB | MXFP4 | MXFP4 | 128K | 224 | yes | Check |
| GeForce RTX 5080 | 16 GB | MXFP4 | MXFP4 | 126K | 120 | no | Check |
| GeForce RTX 5070 Ti | 16 GB | MXFP4 | MXFP4 | 126K | 112 | no | Check |
| GeForce RTX 5070 | 12 GB | — | offload | — | ~43 | no | Check |
| GeForce RTX 5060 Ti 16 GB | 16 GB | MXFP4 | MXFP4 | 126K | 56 | no | Check |
| GeForce RTX 5060 | 8 GB | — | offload | — | ~15 | no | Check |
| GeForce RTX 4090 | 24 GB | MXFP4 | MXFP4 | 128K | 126 | yes | Check |
| GeForce RTX 4080 Super | 16 GB | MXFP4 | MXFP4 | 126K | 92 | no | Check |
| GeForce RTX 4080 | 16 GB | MXFP4 | MXFP4 | 126K | 90 | no | Check |
| GeForce RTX 4070 Ti Super | 16 GB | MXFP4 | MXFP4 | 126K | 84 | no | Check |
| GeForce RTX 4070 Ti | 12 GB | — | offload | — | ~37 | no | Check |
| GeForce RTX 4070 Super | 12 GB | — | offload | — | ~37 | no | Check |
| GeForce RTX 4070 | 12 GB | — | offload | — | ~37 | no | Check |
| GeForce RTX 4060 Ti 16 GB | 16 GB | MXFP4 | MXFP4 | 126K | 36 | no | Check |
| GeForce RTX 4060 Ti | 8 GB | — | offload | — | ~14 | no | Check |
| GeForce RTX 4060 | 8 GB | — | offload | — | ~14 | no | Check |
| GeForce RTX 3090 Ti | 24 GB | MXFP4 | MXFP4 | 128K | 126 | yes | Check |
| GeForce RTX 3090 | 24 GB | MXFP4 | MXFP4 | 128K | 117 | yes | Check |
| GeForce RTX 3080 Ti | 12 GB | — | offload | — | ~49 | no | Check |
| GeForce RTX 3080 12 GB | 12 GB | — | offload | — | ~49 | no | Check |
| GeForce RTX 3080 | 10 GB | — | offload | — | ~24 | no | Check |
| GeForce RTX 3070 Ti | 8 GB | — | offload | — | ~16 | no | Check |
| GeForce RTX 3070 | 8 GB | — | offload | — | ~15 | no | Check |
| GeForce RTX 3060 Ti | 8 GB | — | offload | — | ~15 | no | Check |
| GeForce RTX 3060 12 GB | 12 GB | — | offload | — | ~31 | no | Check |
| GeForce RTX 2080 Ti | 11 GB | — | offload | — | ~29 | no | Check |
| GeForce RTX 2060 12 GB | 12 GB | — | offload | — | ~29 | no | Check |
| GeForce GTX 1080 Ti | 11 GB | — | offload | — | ~27 | no | Check |
NVIDIA · GeForce laptop
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| GeForce RTX 5090 Laptop | 24 GB | MXFP4 | MXFP4 | 128K | 112 | yes | Check |
| GeForce RTX 4090 Laptop | 16 GB | MXFP4 | MXFP4 | 126K | 72 | no | Check |
| GeForce RTX 4080 Laptop | 12 GB | — | offload | — | ~34 | no | Check |
| GeForce RTX 4070 Laptop | 8 GB | — | offload | — | ~13 | no | Check |
NVIDIA · Workstation
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| RTX 6000 Ada | 48 GB | MXFP4 | MXFP4 | 128K | 120 | yes | Check |
| RTX 5000 Ada | 32 GB | MXFP4 | MXFP4 | 128K | 72 | yes | Check |
| RTX 4000 Ada | 20 GB | MXFP4 | MXFP4 | 128K | 45 | yes | Check |
| RTX 2000 Ada | 16 GB | MXFP4 | MXFP4 | 126K | 28 | no | Check |
| RTX A6000 | 48 GB | MXFP4 | MXFP4 | 128K | 96 | yes | Check |
| RTX A5000 | 24 GB | MXFP4 | MXFP4 | 128K | 96 | yes | Check |
| RTX A4000 | 16 GB | MXFP4 | MXFP4 | 126K | 56 | no | Check |
NVIDIA · Data centre
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| L40S | 48 GB | MXFP4 | MXFP4 | 128K | 108 | yes | Check |
| L4 | 24 GB | MXFP4 | MXFP4 | 128K | 37 | yes | Check |
| A10 | 24 GB | MXFP4 | MXFP4 | 128K | 75 | yes | Check |
| A100 40 GB | 40 GB | MXFP4 | MXFP4 | 128K | 194 | yes | Check |
| A100 80 GB | 80 GB | MXFP4 | MXFP4 | 128K | 255 | yes | Check |
| H100 PCIe | 80 GB | MXFP4 | MXFP4 | 128K | 250 | yes | Check |
| H100 SXM | 80 GB | MXFP4 | MXFP4 | 128K | 418 | yes | Check |
| H200 SXM | 141 GB | MXFP4 | MXFP4 | 128K | 599 | yes | Check |
| Tesla P40 | 24 GB | MXFP4 | MXFP4 | 128K | 43 | yes | Check |
AMD · Radeon
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Radeon RX 9070 XT | 16 GB | MXFP4 | MXFP4 | 126K | 81 | no | Check |
| Radeon RX 9070 | 16 GB | MXFP4 | MXFP4 | 126K | 81 | no | Check |
| Radeon RX 7900 XTX | 24 GB | MXFP4 | MXFP4 | 128K | 120 | yes | Check |
| Radeon RX 7900 XT | 20 GB | MXFP4 | MXFP4 | 128K | 100 | yes | Check |
| Radeon RX 7900 GRE | 16 GB | MXFP4 | MXFP4 | 126K | 72 | no | Check |
| Radeon RX 7800 XT | 16 GB | MXFP4 | MXFP4 | 126K | 78 | no | Check |
| Radeon RX 7700 XT | 12 GB | — | offload | — | ~34 | no | Check |
| Radeon RX 7600 XT | 16 GB | MXFP4 | MXFP4 | 126K | 36 | no | Check |
| Radeon RX 6900 XT | 16 GB | MXFP4 | MXFP4 | 126K | 64 | no | Check |
| Radeon RX 6800 XT | 16 GB | MXFP4 | MXFP4 | 126K | 64 | no | Check |
| Radeon RX 6700 XT | 12 GB | — | offload | — | ~32 | no | Check |
AMD · Radeon Pro
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Radeon Pro W7900 | 48 GB | MXFP4 | MXFP4 | 128K | 108 | yes | Check |
| Radeon Pro W7800 | 32 GB | MXFP4 | MXFP4 | 128K | 72 | yes | Check |
AMD · Instinct
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Instinct MI210 | 64 GB | MXFP4 | MXFP4 | 128K | 204 | yes | Check |
| Instinct MI300X | 192 GB | MXFP4 | MXFP4 | 128K | 662 | yes | Check |
AMD · Ryzen AI Max
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Ryzen AI Max+ 395 · 128 GB | 128 GB | MXFP4 | MXFP4 | 128K | 36 | yes | Check |
| Ryzen AI Max+ 395 · 64 GB | 64 GB | MXFP4 | MXFP4 | 128K | 36 | yes | Check |
Apple · Apple Silicon
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| M1 · 16 GB | 16 GB | — | no | — | — | no | Check |
| M1 Pro · 32 GB | 32 GB | MXFP4 | MXFP4 | 128K | 28 | yes | Check |
| M1 Max · 64 GB | 64 GB | MXFP4 | MXFP4 | 128K | 56 | yes | Check |
| M1 Ultra · 128 GB | 128 GB | MXFP4 | MXFP4 | 128K | 112 | yes | Check |
| M2 · 24 GB | 24 GB | MXFP4 | MXFP4 | 128K | 14 | yes | Check |
| M2 Pro · 32 GB | 32 GB | MXFP4 | MXFP4 | 128K | 28 | yes | Check |
| M2 Max · 96 GB | 96 GB | MXFP4 | MXFP4 | 128K | 56 | yes | Check |
| M2 Ultra · 192 GB | 192 GB | MXFP4 | MXFP4 | 128K | 112 | yes | Check |
| M3 · 24 GB | 24 GB | MXFP4 | MXFP4 | 128K | 14 | yes | Check |
| M3 Pro · 36 GB | 36 GB | MXFP4 | MXFP4 | 128K | 21 | yes | Check |
| M3 Max · 64 GB | 64 GB | MXFP4 | MXFP4 | 128K | 56 | yes | Check |
| M3 Max · 128 GB | 128 GB | MXFP4 | MXFP4 | 128K | 56 | yes | Check |
| M3 Ultra · 256 GB | 256 GB | MXFP4 | MXFP4 | 128K | 112 | yes | Check |
| M3 Ultra · 512 GB | 512 GB | MXFP4 | MXFP4 | 128K | 112 | yes | Check |
| M4 · 32 GB | 32 GB | MXFP4 | MXFP4 | 128K | 17 | yes | Check |
| M4 Pro · 48 GB | 48 GB | MXFP4 | MXFP4 | 128K | 38 | yes | Check |
| M4 Pro · 64 GB | 64 GB | MXFP4 | MXFP4 | 128K | 38 | yes | Check |
| M4 Max · 64 GB | 64 GB | MXFP4 | MXFP4 | 128K | 57 | yes | Check |
| M4 Max · 128 GB | 128 GB | MXFP4 | MXFP4 | 128K | 76 | yes | Check |
Intel · Arc
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| Arc B580 | 12 GB | — | offload | — | ~35 | no | Check |
| Arc B570 | 10 GB | — | offload | — | ~20 | no | Check |
| Arc A770 16 GB | 16 GB | MXFP4 | MXFP4 | 126K | 70 | no | Check |
| Arc A750 | 8 GB | — | offload | — | ~15 | no | Check |
CPU · CPU + system RAM
| Device | Memory | Best quant with headroom | Fits at all | Max context @ MXFP4 | Tok/s est. | 128K | |
|---|---|---|---|---|---|---|---|
| CPU only · DDR4 dual-channel | 32 GB | MXFP4 | MXFP4 | 128K | 5.1 | yes | Check |
| CPU only · DDR5 dual-channel | 64 GB | MXFP4 | MXFP4 | 128K | 9.0 | yes | Check |
| CPU only · DDR5 8-channel server | 256 GB | MXFP4 | MXFP4 | 128K | 31 | yes | Check |