Architecture truth card — Qwen3.8-2.4T-A95B
Exact safetensors params2,446,182,725,504 (2446.2B)
Hidden layers92
KV heads (GQA)4
Head dim256
Data snapshot (UTC)2026-11-18T18:29:13+00:00
Snapshot SHA-256d89025de2d15…
← Open the interactive VRAM & cost calculator on the homepage
Exact VRAM matrix — 3 precisions × 3 context lengths
| Precision | 8192 tokens ctx | 32768 tokens ctx | 131072 tokens ctx |
|---|---|---|---|
| FP16/BF16 16-bit | 4556.37 / 2.88 / 229.32 Total 4788.56 GiB | 4556.37 / 11.50 / 229.32 Total 4797.19 GiB | 4556.37 / 46.00 / 229.32 Total 4831.69 GiB |
| FP8/INT8 8-bit | 2278.19 / 2.88 / 115.41 Total 2396.47 GiB | 2278.19 / 11.50 / 115.41 Total 2405.09 GiB | 2278.19 / 46.00 / 115.41 Total 2439.59 GiB |
| INT4/AWQ 4.5-bit | 1281.48 / 2.88 / 65.57 Total 1349.93 GiB | 1281.48 / 11.50 / 65.57 Total 1358.55 GiB | 1281.48 / 46.00 / 65.57 Total 1393.05 GiB |
Each cell: Weights GiB / GQA KV Cache GiB / Runtime Overhead GiB; bold line = Total VRAM GiB. Concurrency = 1, FP16 KV cache (2 bytes), overhead = 1.5 + 0.05 × weights GiB.
Terminal speed-math — no install needed
Re-run this exact sizing from your own terminal:
curl -fsSL https://gpuvram.pages.dev/vram-cli.py | python3 - qwen3-8-2-4t-a95b --ctx 32768
Live cheapest GPU configurations
| Scenario | Cheapest fitting rig (top 3) | Total VRAM | Lowest $/hr | 730 h/mo | Deploy |
|---|---|---|---|---|---|
| INT4 @ 32K ctx | 8x MI300X 192GB | 1536 GiB | $4.00 | $2920.00 | Deploy on RunPodCheck Vast.ai Verified |
| INT4 @ 32K ctx | 8x B200 180GB | 1440 GiB | $47.84 | $34923.20 | Deploy on RunPodCheck Vast.ai Verified |
| INT4 @ 32K ctx | 8x B300 288GB | 2304 GiB | $55.52 | $40529.60 | Deploy on RunPodCheck Vast.ai Verified |