The NVIDIA RTX A4000 is an enterprise-class GPU built on the Ampere architecture, featuring 16 GB of memory with 448 GB/s of memory bandwidth and a 140 W TDP. It delivers 19.2 TFLOPS of FP32 performance. Currently available from 5 providers starting at $0.12/GPU/hour, with a market median of $0.16/GPU/hour across 23 configurations.
Hardware specifications
Same across all providers
Median price across all providers
7 offerings from 5 providers
Sorted by price ascending. Compare configurations side-by-side.
| Provider | Count | vCPU | RAM | Region | Per GPU hour | Total/hr | Action |
|---|---|---|---|---|---|---|---|
| ×1 | — | — | -- | $0.250 | $0.25 | Launch | |
| ×2 | 8 | 31 GB |    +1 | From$0.120 0.0% 30d | From$0.24 | Launch | |
| ×1 | 4 | 21 GB |  | $0.150 0.0% 30d | From$0.15 | ||
Up to$500credit | ×1 | — | — | -- | $0.250 | $0.25 | Launch |
| ×1 | — | — | -- | $0.300 0.0% 30d | $0.30 | ||
| ×1 | 8 | 45 GB | | $0.840 +425.0% 30d | From$0.84 | Launch | |
 Sesterce CloudReferral link | ×1 | 8 | 45 GB | | From$0.880 0.0% 30d | From$0.88 | Launch |
| ×1 | 8 | 45 GB | | $1.040 +511.8% 30d | From$1.04 | Launch |
LLMs that fit on the NVIDIA RTX A4000
Estimated VRAM for popular open-weight LLMs at 16-bit (FP16/BF16), 8-bit (FP8/INT8) and 4-bit (INT4/MXFP4) precision, against 16 GB per GPU. Showing the 12 largest that fit.
| Model | Parameters | 16-bit | 8-bit | 4-bit |
|---|---|---|---|---|
| DeepSeek V4 Flash DeepSeek | 284B | — Not released | — Not released | 192 GB Tight on 11× · 10% spare |
| Solar Open2 250B Upstage | 250B 15B active | — Not released | — Not released | 192 GB Tight on 11× · 9% spare |
| Qwen3 235B-A22B Alibaba | 235B 22B active | — Not released | — Not released | 181 GB Tight on 11× · 16% spare |
| Laguna-S 2.1 Poolside | 118B 8B active | 282 GB Too large | 155 GB 11× · 21 GB free | 90 GB 6× · 5.7 GB free |
| gpt-oss-120b OpenAI | 117B 5.1B active | — Not released | — Not released | 78 GB 6× · 18 GB free |
| Qwen3.6 35B-A3B Alibaba | 35B 3B active | 86 GB 6× · 9.7 GB free | 47 GB Barely fits on 3× | 28 GB 2× · 4.4 GB free |
| Qwen3 32B Alibaba | 32.8B | 79 GB 6× · 17 GB free | 43 GB 3× · 4.8 GB free | 25 GB 2× · 6.8 GB free |
| Gemma 4 31B Google | 31.3B | 75 GB 6× · 21 GB free or 4× · 2% spare | 41 GB 3× · 6.7 GB free | 24 GB 2× · 8 GB free |
| GLM-4.7 Flash Z.ai | 31.2B | 75 GB 6× · 21 GB free or 4× · 2% spare | 41 GB 3× · 6.8 GB free | 24 GB 2× · 8 GB free |
| Qwen3.6 27B Alibaba | 27.8B | 67 GB 6× · 29 GB free or 4× · 15% spare | 37 GB 3× · 11 GB free or 2× · 4% spare | 21 GB 2× · 11 GB free |
| Mistral Small 3.2 24B Mistral AI | 24B | 58 GB 4× · 6.4 GB free | 32 GB Barely fits on 2× | 18 GB 2× · 14 GB free or 1× · 4% spare |
| gpt-oss-20b OpenAI | 21B 3.6B active | — Not released | — Not released | 17 GB 2× · 15 GB free or 1× · 16% spare |
Estimates: model weights plus 20% for KV cache and runtime overhead, at short context lengths. Long contexts and large batches need more. Rows marked “tight” hold the weights but not that full margin. A ×N figure is the total VRAM across N of these GPUs and makes no claim about interconnect throughput. How we estimate this · Start from a model instead
Want more from this page?
See incorrect data?