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
/
JIRU-W
JIRU-W/upcycle_dense15000_e32top8_noaux_rewarm200
Hugging Face
3.7B params · MoE 32 experts, top-8
MaskMoeForCausalLM
text-generation
apache-2.0
4K trained context
6.9 GiB repo
RTX 3060 12GB
◐
Tight fit
BF16
68.8 tok/s
RTX 4060 Ti 16GB
✓
Fits
BF16
56.0 tok/s
RTX 4090 24GB
✓
Fits
BF16
173 tok/s
RTX 5090 32GB
✓
Fits
BF16
272 tok/s
2x RTX 3090 (48GB)
✓
Fits
BF16
162 tok/s
Radeon RX 7900 XTX 24GB
✓
Fits
BF16
84.2 tok/s
Apple M4 32GB
✓
Fits
BF16
21.1 tok/s
Apple M4 Pro 48GB
✓
Fits
BF16
42.7 tok/s
Apple M4 Max 128GB
✓
Fits
BF16
71.6 tok/s
Apple M3 Ultra 512GB
✓
Fits
BF16
92.5 tok/s
NVIDIA DGX Spark (GB10) 128GB
✓
Fits
BF16
53.3 tok/s
AMD Ryzen AI Max+ 395 128GB
✓
Fits
BF16
29.9 tok/s
H100 80GB SXM
✓
Fits
BF16
415 tok/s
NVIDIA DGX/HGX H100 8x80GB (640GB)
✓
Fits
BF16
415 tok/s
Apple iPhone 17 Pro / Pro Max (A19 Pro) 12GB
✕
Won't fit
BF16
Samsung Galaxy S25 Ultra (Snapdragon 8 Elite) 12GB
✕
Won't fit
BF16
Raspberry Pi 5 16GB
✓
Fits
BF16
2.10 tok/s
NVIDIA Jetson AGX Thor 128GB
✓
Fits
BF16
25.4 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