DeepSeek-Coder-V2-Lite · 11 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for DeepSeek-Coder-V2-Lite?
DeepSeek-Coder-V2-Lite on 11 GPUs, measured first-party: NVIDIA RTX PRO 6000 Blackwell Workstation Edition leads at 352.5 tok/s, T4 trails at 83.5 tok/s, and it peaked at 11GB of VRAM.
Benchmarked weights: bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
352.5 tok/s on DeepSeek-Coder-V2-Lite, the ceiling. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 352.5 tok/s on DeepSeek-Coder-V2-Lite
- 96GB, clears the DeepSeek-Coder-V2-Lite floor
- Rentable by the hour rather than bought
- 600W board rating
- Datacenter or workstation hardware, not a retail purchase
What GPU Do You Need for DeepSeek-Coder-V2-Lite?, tok/s by GPU
Efficiency: tok/s per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
DeepSeek-Coder-V2-Lite. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 352.5 | 13620.6 | 2.28 | 154.5 W |
| NVIDIA B300 | 340.5 | 7122.7 | 1.22 | 278.5 W |
| NVIDIA B200 | 325.9 | 10809.7 | 1.1 | 295.2 W |
| NVIDIA H200 | 312.1 | 10345.8 | 1.89 | 165.5 W |
| NVIDIA H100 80GB HBM3 | 309.1 | 10627.6 | 1.79 | 173.1 W |
| NVIDIA L40S | 257.6 | 11547.5 | 1.93 | 133.7 W |
| NVIDIA A100 80GB SXM4 | 213.9 | 5358.9 | 1.43 | 149.9 W |
| NVIDIA A100 40GB SXM4 | 205.4 | 5289.1 | 1.71 | 120.3 W |
| NVIDIA A10G | 160.1 | 4004.5 | 1.54 | 104.0 W |
| NVIDIA L4 | 110.1 | 3580.1 | 2.15 | 51.1 W |
| NVIDIA T4 | 83.49 | 1475.5 | 1.52 | 54.9 W |
What the numbers show. Across 11 GPUs measured on our own bench, RTX PRO 6000 Blackwell Workstation Edition is fastest at 353 tok/s. The slowest, T4, manages 83.5, so the spread is 4.2x from top to bottom. Per dollar of launch price, A10G gives the most (57.2 tok/s per $1,000). The fastest card with 16GB or less is T4 at 83.5 tok/s.
About DeepSeek-Coder-V2-Lite. DeepSeek-Coder-V2-Lite: from deepseek-ai, 16B parameters, on Hugging Face since June 2024. 1,230,139 downloads in the last 30 days and 1 community quantizations.
How it compares. H100 80GB HBM3: DeepSeek-Coder-V2-Lite 309.1 tok/s, Qwen2.5-Coder-14B 144.8 (15B), Qwen3 14B 151.2 (15B), Qwen2.5-14B 144.7 (15B), gpt-oss-20b 346.8 (21B). 1 of 4 beat DeepSeek-Coder-V2-Lite here.
Cost on a rented GPU. 1M generated tokens of DeepSeek-Coder-V2-Lite: $0.45 on a T4 ($0.14/hr, 3.3 hours), $0.85 on a RTX PRO 6000 Blackwell Workstation Edition ($1.08/hr, 47 min, 1.9x the cost).
DeepSeek-Coder-V2-Lite: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA T4 | $0.14/hr | 83.49 | $0.45 |
| NVIDIA A100 40GB SXM4 | $0.47/hr | 205.4 | $0.64 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 352.5 | $0.85 |
| NVIDIA L40S | $0.79/hr | 257.6 | $0.85 |
| NVIDIA L4 | $0.44/hr | 110.1 | $1.11 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 213.9 | $1.23 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 309.1 | $1.92 |
| NVIDIA H200 | $3.59/hr | 312.1 | $3.20 |
| NVIDIA B200 | $5.98/hr | 325.9 | $5.10 |
| NVIDIA B300 | $6.94/hr | 340.5 | $5.66 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for DeepSeek-Coder-V2-Lite. 30+ tok/s: 11 (RTX PRO 6000 Blackwell Workstation Edition, B300, B200). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before DeepSeek-Coder-V2-Lite writes anything it reads the input: 13620.6 tok/s on the RTX PRO 6000 Blackwell Workstation Edition (0.3s for a 4,000-token prompt), 1475.5 on the T4 (2.7s). Long documents and big code files feel this number more than the generation speed.
VRAM for DeepSeek-Coder-V2-Lite. Measured peak 10.5GB, so 12GB is the smallest common card size; smallest card it ran on: T4 (16GB). With long context: Q4_K_M 14GB (tested), Q2_K 9GB, Q3_K_M 11GB, Q5_K_M 17GB, Q6_K 20GB.
Power on DeepSeek-Coder-V2-Lite. Most efficient: RTX PRO 6000 Blackwell Workstation Edition, 154W, 0.12 kWh per 1M generated tokens. Hungriest: B200, 295W, 0.25 kWh. At $0.15/kWh: $0.018 per 1M generated tokens.
Our verdict
Fastest on DeepSeek-Coder-V2-Lite: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 352.5 tok/s. Cheapest to rent per job: NVIDIA T4, $0.45 per 1M generated tokens.
FAQ
- What GPU do I need to run DeepSeek-Coder-V2-Lite?
- About 11GB. Smallest card that ran it: NVIDIA T4 (16GB).
- How much does it cost to run DeepSeek-Coder-V2-Lite in the cloud?
- $0.45 per 1M generated tokens on a NVIDIA T4 at $0.14/hr, cheapest of 10 rentable cards we measured.
- Can I run DeepSeek-Coder-V2-Lite on a 12GB, 16GB or 24GB card?
- It used 10.5GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for DeepSeek-Coder-V2-Lite?
- The H100 80GB HBM3: 309.1 vs 213.9 tok/s, 45% faster on our bench.
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.