ai.canshu.com
中文
VRAM Calculator Advisor & TCO API Pricing Compare Models Leaderboard GPUs

Search models, GPUs, and tools

↑↓ Navigate · Enter Open · ★ Watchlist prioritized ai.canshu.com
Q4_K_M: Smooth Fit 6.09 GB / 23.5 GB Usable

DeepSeek-R1-Distill-Qwen-7B on NVIDIA GeForce RTX 4090 24GB: VRAM Requirement & Speed Estimate

Yes. At Q4_K_M 4-bit quantization with a 16K context window, DeepSeek-R1-Distill-Qwen-7B requires 6.09 GB of VRAM (4.61 GB weights + 0.94 GB KV cache + 0.54 GB overhead), fitting within RTX 4090's 23.5 GB usable memory at ~142.9 tokens/s.

Concurrency (Batch):
Dense 7.6B·28L (d=3584)·28Q/4KV·Ctx 128K
Runnable Models →
GPUs
24 GB VRAM·1008 GB/s·Safe Usable 23.5 GB
Smooth Fit6.09 / 23.5 GB~142.9 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
6.09/ 23.5 GB usable

Physical 24 GB · Load 25.9%

Theoretical Decode Speed
~142.9tokens/s

Actual speed varies ~±25% by runtime

Max Context at Batch=1
128K/ 128K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 4.61 GBKV Cache (1×16K): 0.94 GBOverhead: 0.54 GB
🚀 Blazing Fast (Instant CoT)~3.5s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run deepseek-r1:7b

RTX 4090 Fits DeepSeek-R1-Distill-Qwen-7B · Cost & TCO

Approx. $1,899 (~$79/GB VRAM) · 450W TDP

📊 All Quantization Levels for DeepSeek-R1-Distill-Qwen-7B on RTX 4090

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 16.89 GB Smooth Fit ~46.3 tok/s 128K
FP8 / Q8_0 9.63 GB Smooth Fit ~84.9 tok/s 128K
GGUF Q6_K 7.75 GB Smooth Fit ~108.3 tok/s 128K
GGUF Q5_K_M 6.9 GB Smooth Fit ~123.5 tok/s 128K
Q4_K_M / AWQ 6.09 GB Smooth Fit ~142.9 tok/s 128K
GGUF IQ4_XS 5.51 GB Smooth Fit ~161 tok/s 128K
GGUF Q3_K_M 5.18 GB Smooth Fit ~173.4 tok/s 128K
GGUF Q2_K 4.44 GB Smooth Fit ~209.6 tok/s 128K

💡 Recommended Alternative GPUs & Macs for DeepSeek-R1-Distill-Qwen-7B

Frequently Asked Questions

Can NVIDIA GeForce RTX 4090 24GB run DeepSeek-R1-Distill-Qwen-7B locally?

Yes. At Q4_K_M 4-bit quantization with a 16K context window, DeepSeek-R1-Distill-Qwen-7B requires 6.09 GB of VRAM (4.61 GB weights + 0.94 GB KV cache + 0.54 GB overhead), fitting within RTX 4090's 23.5 GB usable memory at ~142.9 tokens/s.

What is the best quantization for DeepSeek-R1-Distill-Qwen-7B on RTX 4090?

The highest-quality quantization that fits cleanly on RTX 4090 is FP16 / BF16 (Lossless Full Precision), using 16.89 GB VRAM at ~46.3 tokens/s with up to 128K context.

How much VRAM does DeepSeek-R1-Distill-Qwen-7B need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), DeepSeek-R1-Distill-Qwen-7B requires 16.89 GB VRAM (15.2 GB for weights alone).