All models

DeepSeek / LANGUAGE MODEL

deepseek-coder-33b-instruct VRAM requirements

Explore memory tradeoffs with sourced model data and transparent estimates.

Q4_K_M estimate

24.3 GiB

at 8,192 tokens

FP16 estimate

72.6 GiB

at 8,192 tokens

Suggested GPU

1 × RTX 5090 32 GiB

Model architecture

Model computation diagram
  1. Token embedding · 7168
  2. 62 layers · GQA
  3. Output head → token scores
Parameters33,342,991,360 (33.34B)
Layers62
Attention heads56
KV heads8
Head dimension128
Hidden size7168
Maximum context16,384 tokens
NotesStandard transformer KV-cache estimate.

Minimum hardware and family comparison

Estimates use batch 1, an FP16 KV cache and 12% headroom. Multi-GPU capacity assumes even sharding and requires a compatible runtime.

DevicesRequired totalSmallest fitting device exampleUsable per device
124.29 GiBApple M3 Max (36 GiB)27.0 GiB
224.62 GiBRTX 4060 Ti 16 GiB16.0 GiB
425.30 GiBRTX 3060 Ti 8 GiB8.0 GiB
826.64 GiBGTX 1050 Ti 4 GiB4.0 GiB

Quantization comparison

Batch 1 · FP16 KV cache · 1 GPU · totals include safety margin. Contexts beyond the model maximum are hypothetical and marked *.

PrecisionWeightsTotal @ 4kTotal @ 8kTotal @ 32k *
FP16 / BF1616.0005 bpw · FP16 GGUF reference62.11 GiB71.49 GiB72.57 GiB79.08 GiB
FP88 bpw · Nominal bit width31.05 GiB36.70 GiB37.79 GiB44.30 GiB
INT88 bpw · Nominal bit width31.05 GiB36.70 GiB37.79 GiB44.30 GiB
NVFP66 bpw · Nominal bit width23.29 GiB28.01 GiB29.09 GiB35.60 GiB

GPU architecture and runtime requirements

NVIDIA architectureAttention backendPublished environment or release
Ampere / AdaFlashAttention-2CUDA ≥ 12.0 · PyTorch ≥ 2.2 · FP16 / BF16
Hopper (H100 / H800)FlashAttention-3CUDA ≥ 12.3
Blackwell (B200)FlashAttention-4 · CuTeDSLPyTorch 2.7 introduced Blackwell support in its CUDA 12.8 builds

Transformers version is checkpoint-specific. Model card · Kernel compatibility · PyTorch build notes

Recommended GPUs by memory fit

Q4_K_M @ 8k context, batch 1, FP16 KV cache. Smallest nominal memory first; green results only. Check backend and precision support separately.

Sources & verification

Try your configuration

Your workload

Workload

Precision of the weights held in memory.

Longer contexts make cache precision more important.

Runtime settingsRoofline estimate

Your hardware

Input settings

Sequences per batch. Serving concurrency and throughput require separate measurements.

Advanced runtime settings
Advanced · Model architecture
Layers
62
Attention heads
56
KV heads
8
Head dimension
128
Hidden size
7,168
Maximum context (tokens)
16,384
View model configuration

Performance and memory results

101.2%Memory use

Memory estimate

24.3 GiB needed

Tight0.3 GiB over budget24.0 GiB available

The estimate fits only without the safety margin. Reduce the workload to leave headroom.

Estimated GPU rental · $0.44/h · Per-GPU quote × devices

This configuration exceeds available memory. Adjust the model or hardware.

deepseek-coder-33b-instruct

RTX 4090 24 GiB

Q4_K_M · FP16 KV

Model specifications
Memory breakdownGiB

Vertical line = usable hardware memory

  • Weights (Q4_K_M)19.00 GiB
  • KV cache (FP16)1.94 GiB
  • Runtime overhead0.75 GiB
  • CUDA Graph capture0.00 GiB · 0.0%
  • Safety margin2.60 GiB
Framework overhead breakdown

Planning allocations. Actual driver, library and allocator use varies by runtime.

Device runtime and driver
0.50 GiB
Operator workspaces
0.20 GiB
CUDA Graph
0.00 GiB
Allocator base
0.05 GiB
Other runtime buffers and communication
0.00 GiB
Compare context lengths
Compare GPU capacity

Inference simulation

deepseek-coder-33b-instruct · Q4_K_M · RTX 4090 24 GiB

Prompt processing
—
Token generation
—
Estimated duration
—
Timing playback · sample text
Timing playback · sample text0 / 128 · 0 s
This configuration exceeds available memory. Adjust the model or hardware.
Measured latency percentiles and request throughput

Frequently asked questions

How much VRAM does deepseek-coder-33b-instruct need?

Our default Q4_K_M estimate at 8,192 context tokens, batch 1 and FP16 KV cache is 24.3 GiB including the safety margin. These are estimates, not measured guarantees.

Why does context length change memory use?

KV cache grows linearly with context length and batch size. We use KV heads for grouped-query attention and the recorded KV-cache layer count.

Does a green result guarantee compatibility?

No. Green means the estimated memory including safety headroom fits. Backend support, precision support, kernels and runtime allocation can differ. Confirm compatibility and measure your actual workflow.

Where do these estimates come from?

Architecture and file sizes are traced to model sources. The precision table identifies measured GGUF rates, block storage and nominal bit widths. The source data for deepseek-coder-33b-instruct was last verified on 2026-11-20. Memory estimates use GiB.