AMD Instinct MI300X
192 GB HBM3 · CDNA 3 · 1,307 FP16 TFLOPS
Live AMD Instinct MI300X cloud prices from 3 providers, updated .
4 offerings from 3 providers
| Provider | Per GPU / hour | Action | |
|---|---|---|---|
1 GPU | $1.71 | Launch | |
| |||
Runpod*Up to$500credit 1 GPU | $2.39 | ||
| |||
Cyfuture AIUp to$1credit 8 GPUs · $22.38 total | from$2.798 | Launch | |
| |||
8 GPUs · $38.80 total | $4.85 | Launch | |
| |||
* We may earn a referral fee or commission from some providers when you sign up through our links. This does not affect ranking or ordering.
AMD Instinct MI300X price history
The median AMD Instinct MI300X price is up 13% since November 24, 2025.
LLMs that fit on the AMD Instinct MI300X
Popular open-weight LLMs against 192 GB per GPU: weights plus room for the KV cache, at the best precision that fits. Showing the 12 largest that fit.
Runs on one AMD Instinct MI300X
- DeepSeek V4 Flash4-bit · 192 GB
- Solar Open2 250B4-bit · tight, 19% spare
- Qwen3 235B-A22B4-bit · 181 GB
- Laguna-S 2.18-bit · 155 GB
- gpt-oss-120b4-bit · 78 GB
- Qwen3.6 35B-A3B16-bit · 86 GB
- Qwen3 32B16-bit · 79 GB
- Gemma 4 31B16-bit · 75 GB
Needs more than one AMD Instinct MI300X
- DeepSeek V4 Pro8 GPUs · 4-bit
- Kimi K2.54 GPUs · 4-bit
- GLM-5.24 GPUs · 4-bit
- DeepSeek R14 GPUs · 4-bit
Full table with every precision
Estimated VRAM for popular open-weight LLMs at 16-bit (FP16/BF16), 8-bit (FP8/INT8) and 4-bit (INT4/MXFP4) precision, against 192 GB per GPU.
| Model | Parameters | 16-bit | 8-bit | 4-bit |
|---|---|---|---|---|
| DeepSeek V4 Pro DeepSeek | 1.6T 49B active | — Not released | — Not released | 1,038 GB 8× · 498 GB free |
| Kimi K2.5 Moonshot AI | 1T 32B active | — Not released | — Not released | 714 GB 4× · 54 GB free |
| GLM-5.2 Z.ai | 753B | 1,808 GB Tight on 8× · 1% spare | 994 GB 8× · 542 GB free | 579 GB 4× · 189 GB free |
| DeepSeek R1 DeepSeek | 671B 37B active | — Not released | 826 GB 8× · 710 GB free or 4× · 11% spare | 481 GB 4× · 287 GB free |
| DeepSeek V4 Flash DeepSeek | 284B | — Not released | — Not released | 192 GB Barely fits on 1× |
| Solar Open2 250B Upstage | 250B 15B active | 601 GB 4× · 167 GB free | 330 GB 2× · 54 GB free | 192 GB 2× · 192 GB free or 1× · 19% spare |
| Qwen3 235B-A22B Alibaba | 235B 22B active | 564 GB 4× · 204 GB free | 310 GB 2× · 74 GB free | 181 GB 1× · 11 GB free |
| Laguna-S 2.1 Poolside | 118B 8B active | 282 GB 2× · 102 GB free | 155 GB 1× · 37 GB free | 90 GB 1× · 102 GB free |
| gpt-oss-120b OpenAI | 117B 5.1B active | — Not released | — Not released | 78 GB 1× · 114 GB free |
| Qwen3.6 35B-A3B Alibaba | 35B 3B active | 86 GB 1× · 106 GB free | 47 GB 1× · 145 GB free | 28 GB 1× · 164 GB free |
| Qwen3 32B Alibaba | 32.8B | 79 GB 1× · 113 GB free | 43 GB 1× · 149 GB free | 25 GB 1× · 167 GB free |
| Gemma 4 31B Google | 31.3B | 75 GB 1× · 117 GB free | 41 GB 1× · 151 GB free | 24 GB 1× · 168 GB free |
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
AMD Instinct MI300X specs
The AMD Instinct MI300X is an enterprise-class GPU built on the CDNA 3 architecture, featuring 192 GB of memory with 5,200 GB/s of memory bandwidth and a 750 W TDP. It delivers 163.4 TFLOPS of FP32 performance and 1,307 TFLOPS of FP16 tensor performance.
- Memory bandwidth
- 5,200 GB/s
- FP16 performance
- 1,307 TFLOPS
- peak
- FP32 performance
- 163.4 TFLOPS
- Architecture
- CDNA 3
- TDP
- 750 W
- Brand
- AMD
- Series
- Instinct MI
Similar GPUs
See more GPUs like this inHighest performance


