Run it locally
Per-quant memory needs and a static "can you run it?" reference - no rig entry required
Q4_K_M
8.5GB
8.5GB min
10.0GB rec
Balanced - the usual local sweet spot
Q8_0
15.0GB
15.0GB min
17.0GB rec
Near-lossless
FP16
29.0GB
29.0GB min
32.0GB rec
Full quality, largest
The reference hardware
22 reference configs, drawn in-house. Scroll for more.
Can you run it? - reference rigs
| Rig |
Q4_K_M |
Q8_0 |
FP16 |
| 4x H100 80GB (320GB) |
fast 866.9t/s
|
fast 491.3t/s
|
fast 254.1t/s
|
| NVIDIA DGX Station 748GB |
fast 517.6t/s
|
fast 293.3t/s
|
fast 151.7t/s
|
| 8x RTX 3090 rack (192GB) |
fast 484.5t/s
|
fast 274.6t/s
|
fast 142.0t/s
|
| 4x RTX 5090 (128GB) |
fast 463.7t/s
|
fast 262.8t/s
|
fast 135.9t/s
|
| AMD Instinct MI300X (192GB) |
fast 344.5t/s
|
fast 195.2t/s
|
fast 101.0t/s
|
| 4x RTX 4090 (96GB) |
fast 260.9t/s
|
fast 147.8t/s
|
fast 76.5t/s
|
| 2x RTX 5090 (64GB) |
fast 231.9t/s
|
fast 131.4t/s
|
fast 68.0t/s
|
| 2x RTX 3090 (48GB) |
fast 121.1t/s
|
fast 68.7t/s
|
fast 35.5t/s
|
| Single RTX 5090 (32GB) |
fast 115.9t/s
|
fast 65.7t/s
|
fast 34.0t/s
|
| RTX PRO 6000 Blackwell (96GB) |
fast 115.9t/s
|
fast 65.7t/s
|
fast 34.0t/s
|
| Mac Studio M4 Ultra 192GB |
fast 77.1t/s
|
fast 43.7t/s
|
fast 22.6t/s
|
| Mac Studio M4 Ultra 512GB |
fast 77.1t/s
|
fast 43.7t/s
|
fast 22.6t/s
|
| Single RTX 4090 (24GB) |
fast 65.2t/s
|
fast 37.0t/s
|
offload
|
| MacBook Pro M5 Max 128GB |
fast 43.3t/s
|
fast 24.6t/s
|
ok 12.7t/s
|
| Single GTX 1080 Ti (11GB) |
fast 31.3t/s
|
offload
|
no -> cloud
|
| Dual EPYC 9004 + 768GB DDR5-4800 |
fast 29.8t/s
|
ok 16.9t/s
|
ok 8.7t/s
|
| DGX Spark 128GB unified |
ok 17.7t/s
|
ok 10.0t/s
|
slow 5.2t/s
|
| Ryzen AI Max+ 395 128GB |
ok 16.6t/s
|
ok 9.4t/s
|
slow 4.9t/s
|
| Jetson AGX Orin 64GB |
ok 13.3t/s
|
slow 7.5t/s
|
slow 3.9t/s
|
| Epyc + 512GB DDR4-3200 + 2x RTX 3090 |
ok 13.3t/s
|
slow 7.5t/s
|
slow 3.9t/s
|
| Epyc + 512GB DDR4-2400 + 2x RTX 3090 |
ok 9.9t/s
|
slow 5.6t/s
|
slow 2.9t/s
|
| NVIDIA Jetson Orin NX 16GB |
slow 6.6t/s
|
tight
|
no -> cloud
|
Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.
Or run it in the cloud
Live per-provider pricing, throughput and uptime - refreshed about 19 hours ago via OpenRouter. Click a column to sort.
|
Provider
|
Type
|
Input $/M
|
Output $/M
|
Cache $/M
|
Tok/s
|
Latency
|
Uptime
|
Value
|
|
DeepInfra
|
API
|
0.07 |
0.14 |
- |
- |
- |
100.00% |
best uptime
|
Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.
PRICE HISTORY
Inference cost over time
Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.
Loading price history...