Skip to content
modelread
Ask
Browse
Devices
Connect
Docs
Methodology
Search
⌘K
My hardware
All
Loading devices…
System RAM / unified memory override
GB
Remembered on this device. Multi-GPU: search “2x”, “8x”, “node”.
Ask
Browse
Devices
Connect
Docs
Methodology
On your hardware
Search models, devices and pages
Type a model or device name.
Pages
Ask
what should I run?
Browse
best models per device
Devices
all hardware
Connect your agent
MCP · API
How the numbers work
calibration
Models
/
openai
openai/gpt-oss-20b
Hugging Face
21B params · MoE 32 experts, top-4
GptOssForCausalLM
text-generation
apache-2.0
128K trained context
38.5 GiB repo
RTX 3060 12GB
⇅
Offload
F16
53.7 tok/s
RTX 4060 Ti 16GB
✓
Fits
F16
80.7 tok/s
RTX 4090 24GB
✓
Fits
F16
219 tok/s
RTX 5090 32GB
✓
Fits
MXFP4
214 tok/s
2x RTX 3090 (48GB)
✓
Fits
MXFP4
130 tok/s
Radeon RX 7900 XTX 24GB
✓
Fits
F16
109 tok/s
Apple M4 32GB
✓
Fits
MLX-8BIT
20.9 tok/s
Apple M4 Pro 48GB
✓
Fits
MLX-8BIT
42.2 tok/s
Apple M4 Max 128GB
✓
Fits
MXFP4
62.3 tok/s
Apple M3 Ultra 512GB
✓
Fits
MXFP4
82.1 tok/s
NVIDIA DGX Spark (GB10) 128GB
✓
Fits
MXFP4
44.3 tok/s
AMD Ryzen AI Max+ 395 128GB
✓
Fits
MXFP4
25.6 tok/s
H100 80GB SXM
✓
Fits
MXFP4
316 tok/s
NVIDIA DGX/HGX H100 8x80GB (640GB)
✓
Fits
MXFP4
316 tok/s
Apple iPhone 17 Pro / Pro Max (A19 Pro) 12GB
✕
Won't fit
MXFP4
Samsung Galaxy S25 Ultra (Snapdragon 8 Elite) 12GB
✕
Won't fit
MXFP4
Raspberry Pi 5 16GB
✓
Fits
F16
3.40 tok/s
NVIDIA Jetson AGX Thor 128GB
✓
Fits
MXFP4
21.3 tok/s
My hardware
All
Loading devices…
System RAM / unified memory override
GB
Remembered on this device. Multi-GPU: search “2x”, “8x”, “node”.
Context
4K
8K
16K
32K
64K
128K
256K
1M
KV cache
f16
q8_0
q4_0