AMD's flagship data center GPU for large-model inference.
Weekly median price per GPU per hour · Get the data
By provider shows one card per company, placed where its best offer ranked. By configuration lists every offer.
What you can rent comes first: in stock, then waitlist, then not reported, then out of stock. Priced offers rank ahead of quote-only, and on-demand ahead of other billing types.
Within a group, five factors set the order:
We match the provider, its country, GPU model, form factor, billing type, availability and instance name. Matching is partial and case-insensitive. Everyday words work too, so "interruptible" finds spot and "sold out" finds out of stock.
No providers match .
Heads up: A provider's own page may quote a different figure, for example on tax, region or a promotion. A monthly-only plan shows a derived hourly rate. Verify before provisioning. More on how we price.
One MI300X has 192 GB of VRAM. In practice, that's enough memory for roughly 303B parameters at 4-bit or 81B at 16-bit, assuming a 32K context. Below are some open-weight LLMs, with the estimated memory and GPUs each one needs.
| Model | Memory (INT4 / FP4) | MI300Xs needed | Cost /hr | Cost /mo |
|---|---|---|---|---|
Qwen3.8-27B
27B
|
18 GB
|
1
|
$4.41
|
$3,175
|
Gemma 4 31B
30.7B
|
22 GB
|
1
|
$4.41
|
$3,175
|
GPT-OSS-120B
117B · 5.1B active
|
67 GB
|
1
|
$4.41
|
$3,175
|
GLM-5.3-Flash
320B · 18B active
|
178 GB
|
2
|
$8.82
|
$6,350
|
MiniMax-M3
428B · 23B active
|
239 GB
|
2
|
$8.82
|
$6,350
|
DeepSeek V4.1 Flash
552B · 16B active
|
306 GB
|
2
|
$8.82
|
$6,350
|
Kimi K3
2.8T · 104B active
|
1,544 GB
|
9
|
–
|
–
|
Estimates based on the median on-demand rate. Memory is weights plus FP8 KV cache at 32K context per request (FP16/BF16 in the 16-bit column). GPU counts assume 90% of advertised VRAM is usable. No guarantee of runtime support, usable performance, or that a matching quantized build exists. Pricing methodology.
AMD MI300X · Per GPU
| Compute · dense | |
|---|---|
| FP8 | 2,614.9 TFLOPS 5,229.8 with sparsity |
| FP16 / BF16 | 1,307.4 TFLOPS 2,614.9 with sparsity |
| INT8 | 2,614.9 TOPS 5,229.8 with sparsity |
| FP32 | 163.4 TFLOPS |
| FP64 | 81.7 TFLOPS |
| Precision support | FP8FP16BF16TF32FP32FP64INT8 |
| Memory | |
|---|---|
| Capacity | 192 GB HBM3 |
| Bandwidth | 5,300 GB/s |
| Bus width | 8,192-bit |
| ECC | Yes |
| Silicon | |
|---|---|
| Architecture | CDNA 3 |
| Process | TSMC 5nm / 6nm FinFET |
| Transistors | 153 billion |
| Shader cores | 19,456 Stream processors |
| Matrix cores | 1,216 |
| Compute units | 304 |
| Fabric and host | |
|---|---|
| GPU interconnect | Infinity Fabric 896 GB/s |
| Host interface | PCIe 5.0 x16 |
| Power | |
|---|---|
| Board power | 750 W |
| Cooling | Passive |
| Platform | |
|---|---|
| Partitioning | Up to 8 partitions |
Source: official AMD MI300X datasheet.
As of September 28, 2026, the median on-demand price is $4.41 per GPU per hour across 4 providers with a priced on-demand config.
| Billing type | Configs | Median /GPU/hr | Cheapest |
|---|---|---|---|
On-demand | 14 | $4.41 | $2.59 |
Reserved | 23 | $2.80 | $1.61 (12 mo) |
Spot | 3 | $1.60 |
No median for spot: we only show this when at least 3 providers list the GPU on that billing type. 2 providers also quote the MI300X on a custom contract, priced per deal.
The lowest listed on-demand price for the MI300X is $2.59 per GPU per hour from DigitalOcean, though we haven't verified current stock.
At 720 hours per month, one MI300X costs an estimated $3,175 at the median on-demand price.
GetDeploying currently tracks MI300X configs from 11 providers. See the full price comparison above for every provider and config.
As of September 28, 2026, the median on-demand price has been flat over the past 90 days, at about $4.41 per GPU per hour.
One MI300X runs models up to roughly 303B parameters at 4-bit quantization or 81B at 16-bit, assuming a 32K context. Larger models run across multiple GPUs: the model table above shows the estimated memory and GPU count for popular open-weight LLMs.
Comes with 192GB HBM3 - the largest memory of any single GPU. 5.3 TB/s memory bandwidth. Strong ROCm ecosystem for PyTorch workloads. Competitive pricing against H100 for inference.
ROCm software ecosystem, while rapidly improving, is narrower than CUDA. Some frameworks and libraries may have limited AMD support. Multi-node training tooling is less mature than Nvidia's.
Nvidia H100
80 GB HBM3 · 37% less bandwidth than the MI300X
Nvidia A100
80 GB HBM2e · 63% less bandwidth than the MI300X
AMD MI325X
256 GB HBM3e · released Q4 2024
Last updated
