DeepSeek Coder 7B Instruct v1.5 · 33 GPUs measured first-party · llama.cpp Q4_K_M · Updated October 2026

What GPU Do You Need for DeepSeek Coder 7B Instruct v1.5?

We ran DeepSeek Coder 7B Instruct v1.5 (Q4_K_M, llama.cpp) on 33 GPUs, from budget cards to datacenter parts, and measured generation speed, prompt processing, power draw and peak VRAM on every one. The fastest was NVIDIA H100 80GB HBM3 at 293.9 tok/s.

Benchmarked weights: mradermacher/deepseek-coder-7b-instruct-v1.5-GGUF

Fastest we measured

NVIDIA H100 80GB HBM3

293.9 tok/s on DeepSeek Coder 7B Instruct v1.5, the ceiling. Measured on our bench. 80GB of VRAM, $30,000 at launch.

Pros

• 293.9 tok/s on DeepSeek Coder 7B Instruct v1.5
• 80GB, clears the DeepSeek Coder 7B Instruct v1.5 floor
• Rentable by the hour rather than bought

Cons

• 700W board rating
• Datacenter or workstation hardware, not a retail purchase

Best consumer card

NVIDIA GeForce RTX 5090

291.4 tok/s on DeepSeek Coder 7B Instruct v1.5, fastest card you can buy at retail. Measured on our bench. 32GB of VRAM, $1,999 at launch.

Pros

• 291.4 tok/s on DeepSeek Coder 7B Instruct v1.5
• 32GB, clears the DeepSeek Coder 7B Instruct v1.5 floor

Cons

• 575W board rating

Cheapest card that runs it

GeForce RTX 4060

59.11 tok/s on DeepSeek Coder 7B Instruct v1.5, lowest launch price that still fits. Measured on our bench. 8GB of VRAM, $299 at launch.

Pros

• 59.11 tok/s on DeepSeek Coder 7B Instruct v1.5
• 8GB, clears the DeepSeek Coder 7B Instruct v1.5 floor

Cons

• 115W board rating

Best value

NVIDIA GeForce RTX 3060 Ti

96.35 tok/s on DeepSeek Coder 7B Instruct v1.5, most speed per dollar. Measured on our bench. 8GB of VRAM, $399 at launch. That is 241.5 tok/s per $1,000 of launch price.

Pros

• 96.35 tok/s on DeepSeek Coder 7B Instruct v1.5
• 8GB, clears the DeepSeek Coder 7B Instruct v1.5 floor

Cons

• 200W board rating

What the numbers show. Across 33 GPUs measured on our own bench, H100 80GB HBM3 is fastest at 294 tok/s. Per dollar of launch price, RTX 5060 Ti gives the most (265.1 tok/s per $1,000). The fastest card with 16GB or less is RTX 5080 at 181 tok/s. The measured peak was ~5GB, so it runs on 6GB cards and up. That floor is a property of the model, so it applies to every GPU, measured or not.

About DeepSeek Coder 7B Instruct v1.5. DeepSeek Coder 7B Instruct v1.5: from deepseek-ai, 6.9B parameters, on Hugging Face since January 2024. 648,036 downloads in the last 30 days.

How it compares. H100 80GB HBM3: DeepSeek Coder 7B Instruct v1.5 293.9 tok/s, Llama-2-7B 284.6 (7B), Mistral-7B-Instruct-v0.2 276.7 (7B), Qwen2.5-7B 264.2 (8B), Qwen2.5-Coder 7B 263.9 (8B). DeepSeek Coder 7B Instruct v1.5 beats all 4 here.

Cost on a rented GPU. 1M generated tokens of DeepSeek Coder 7B Instruct v1.5: $0.16 on a RTX 3080 ($0.082/hr, 118 min), $2.02 on a H100 80GB HBM3 ($2.14/hr, 57 min, 12.6x the cost).

Speed tiers for DeepSeek Coder 7B Instruct v1.5. 30+ tok/s: 33 (RTX 5090, RTX 4090, RTX 3090 Ti). 30 tok/s is roughly where replies outpace reading.

Reading your prompt. Before DeepSeek Coder 7B Instruct v1.5 writes anything it reads the input: 16072.3 tok/s on the RTX 5090 (0.2s for a 4,000-token prompt), 12396.8 on the RTX 4090 (0.3s), 1348.6 on the T4 (3.0s). Long documents and big code files feel this number more than the generation speed.

VRAM for DeepSeek Coder 7B Instruct v1.5. Measured peak 4.1GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Ti (8GB). With long context: Q4_K_M 6GB (tested), Q2_K 4GB, Q3_K_M 5GB, Q5_K_M 7GB, Q6_K 9GB.

Power on DeepSeek Coder 7B Instruct v1.5. Most efficient: H100 80GB HBM3, 224W, 0.21 kWh per 1M generated tokens. Hungriest: RTX 3090 Ti, 293W, 0.44 kWh. At $0.15/kWh: $0.032 per 1M generated tokens.

Our verdict

Fastest on DeepSeek Coder 7B Instruct v1.5: NVIDIA H100 80GB HBM3, 293.9 tok/s. Best desktop card: NVIDIA GeForce RTX 5090, 291.4 tok/s. Cheapest consumer card that ran it: GeForce RTX 4060 ($299, 59.11 tok/s). Cheapest to rent per job: NVIDIA GeForce RTX 3080, $0.16 per 1M generated tokens.

FAQ

What GPU do I need to run DeepSeek Coder 7B Instruct v1.5?

About 4GB. Cheapest consumer card that ran it: GeForce RTX 4060 (8GB, 59.11 tok/s).

How fast is DeepSeek Coder 7B Instruct v1.5 on the NVIDIA GeForce RTX 5090?

291.4 tok/s at 266W, 99% of the NVIDIA H100 80GB HBM3.

How much does it cost to run DeepSeek Coder 7B Instruct v1.5 in the cloud?

$0.16 per 1M generated tokens on a NVIDIA GeForce RTX 3080 at $0.082/hr, cheapest of 28 rentable cards we measured.

Can I run DeepSeek Coder 7B Instruct v1.5 on a 12GB, 16GB or 24GB card?

It used 4.1GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.

Is the RTX 4090 or the RTX 3090 faster for DeepSeek Coder 7B Instruct v1.5?

The RTX 4090: 188.6 vs 164.7 tok/s, 14% faster on our bench.

Should I buy or rent a RTX 5090 for DeepSeek Coder 7B Instruct v1.5?

Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 5,390x 1M generated tokens of DeepSeek Coder 7B Instruct v1.5. Buy only if you'll run more than that.

How we test

llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.