✅ PASST IN DEN VRAM (FP8 (W8A8 / E4M3))
Qwen2.5-Coder-32B-Instruct on 2x NVIDIA RTX 3090 (48GB) (48GB VRAM)
Unquantized FP16 (63.96 GB) exceeds 48 GB, but 4-bit Q4_K_M requires 21.23 GB (44% VRAM utilization) at ~32.5 tokens/sec.
🎛️ Interaktiver VRAM-, Kontext- & Offload-Konfigurator
📊 Echtzeit-Speicher- & Performance-Telemetrie
21.23 GB / 48 GB
🟦 Modellgewichte: 18.73 GB
🟪 KV-Cache: 2 GB
🟨 Engine-Puffer: 0.49 GB
Benötigter GPU-VRAM21.23 GB
Benötigter System-RAM (CPU)0.00 GB
Verfügbarer Hardware-VRAM48 GB (1x 2x NVIDIA RTX 3090 (48GB))
VRAM-Reserve / Defizit26.77 GB
Geschätzte Decode-Geschwindigkeit~32.5 tokens/sec
Geschätzte Prefill-Geschwindigkeit~1127 tokens/sec
Time-to-First-Token (4K Prompt)~3634 ms
Transformer-TopologieDense-GQA (64L, 8/40 KV/Q heads)
🚀 Startbefehl (vLLM / llama.cpp / Ollama)
ollama run qwen2.5-coder:32b # Or llama.cpp server (8K context, Flash Attention): llama-server -hf Qwen/Qwen2.5-Coder-32B-Instruct-GGUF -c 8192 -ngl 99 -fa
CLOUD-GPU PREISVERGLEICH & MIETEN-VS-KAUFEN RECHNER
Nicht genug lokaler VRAM? Cloud-GPU in 60 Sekunden starten (Qwen2.5-Coder-32B-Instruct: 21.23 GB in 4-bit)
Recommended Cloud Tier: RTX 4090 (24GB) · Full FP16 Needs: 63.96 GB
Amortisationszeit (Kauf vs. Miete): ~32 months (vs $1,500 hardware)
| Cloud Provider | RTX 4090 (24GB) | A100 (80GB) | H100 (80GB) | Est. Monthly Cost (4h/day) | Free Credit / Trial | Deploy |
|---|---|---|---|---|---|---|
| RunPod | $0.44 / hr | $1.39 / hr | $2.39 / hr | $52.80 / mo | Instant PyTorch / vLLM Templates + Serverless GPU | Auf RunPod starten → |
| Vast.ai | $0.29 / hr | $0.95 / hr | $1.79 / hr | $34.80 / mo | Save 50–70% on RTX 4090 / 5090 / A100 rentals | Spot-GPU mieten → |
| Vultr Cloud GPU | — | $1.79 / hr | $2.69 / hr | $214.80 / mo (A100) | $300 Free Credit | $300 Guthaben sichern → |
| DigitalOcean GPU | — | — | $2.99 / hr | $358.80 / mo (H100) | $200 Free Credit (60d) | $200 Startguthaben → |
Alle Quantisierungsstufen im Detail: Qwen2.5-Coder-32B-Instruct × 2x NVIDIA RTX 3090 (48GB)
| Quantization Format | Core / Emb BPW | Weights (GiB) | KV Cache (8K) | Total VRAM | 1x 2x NVIDIA RTX 3090 (48GB) (48GB) | Decode / Prefill | Min GPUs |
|---|---|---|---|---|---|---|---|
| FP16 / BF16 (Unquantized) | 16 / 16 | 61.02 GB | 2 GB | 63.51 GB | Nein (+15.5GB) | ~2.9 / 1127 tok/s | 2x |
| FP8 (W8A8 / E4M3) | 8.25 / 8.5 | 31.51 GB | 2 GB | 34 GB | Ja (71%) | ~19 / 1127 tok/s | 1x |
| GGUF Q8_0 (8.50 bpw) | 8.5 / 8.5 | 32.42 GB | 2 GB | 34.91 GB | Ja (73%) | ~18.5 / 1127 tok/s | 1x |
| GGUF Q6_K (6.56 bpw) | 6.56 / 6.56 | 25.02 GB | 2 GB | 27.51 GB | Ja (57%) | ~23.9 / 1127 tok/s | 1x |
| GGUF Q5_K_M (5.69 bpw) | 5.69 / 6.56 | 21.86 GB | 2 GB | 24.35 GB | Ja (51%) | ~27.6 / 1127 tok/s | 1x |
| GGUF Q4_K_M / AWQ (4.83 bpw) | 4.83 / 6.56 | 18.73 GB | 2 GB | 21.23 GB | Ja (44%) | ~32.5 / 1127 tok/s | 1x |
| GGUF IQ4_XS (4.25 bpw) | 4.25 / 5.5 | 16.44 GB | 2 GB | 18.93 GB | Ja (39%) | ~37 / 1127 tok/s | 1x |
| GGUF Q3_K_M (3.91 bpw) | 3.91 / 5 | 15.11 GB | 2 GB | 17.6 GB | Ja (37%) | ~40.2 / 1127 tok/s | 1x |
Dieses Modell auf anderen GPUs testen
Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 5090 (32GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 4090 (24GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 5080 (16GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 4080 Super (16GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 4070 Ti Super (16GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 3090 (24GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 4060 Ti 16GB (16GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 3060 12GB (12GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX 6000 Ada (48GB)Qwen2.5-Coder-32B-Instruct on NVIDIA RTX A6000 (48GB)
Weitere Modelle auf dieser GPU
DeepSeek-R1 671B (MoE) on 2x NVIDIA RTX 3090 (48GB)DeepSeek-V3-0324 671B (MoE) on 2x NVIDIA RTX 3090 (48GB)DeepSeek-R1-Distill-Llama-70B on 2x NVIDIA RTX 3090 (48GB)DeepSeek-R1-Distill-Qwen-32B on 2x NVIDIA RTX 3090 (48GB)DeepSeek-R1-Distill-Qwen-14B on 2x NVIDIA RTX 3090 (48GB)DeepSeek-R1-Distill-Llama-8B on 2x NVIDIA RTX 3090 (48GB)DeepSeek-R1-Distill-Qwen-7B on 2x NVIDIA RTX 3090 (48GB)Qwen3-235B-A22B (MoE) on 2x NVIDIA RTX 3090 (48GB)Qwen3-30B-A3B (MoE) on 2x NVIDIA RTX 3090 (48GB)Qwen3-32B on 2x NVIDIA RTX 3090 (48GB)