AMD · CDNA 3 · 2023
AMD MI300X. Rental by the hour, on demand.
From $4.85/hr per GPU on an 8-GPU machine, pay as you go with no commitment. 192GB HBM3 in a single GPU, on the open ROCm stack.
MI300X on demand
from$4.85/hr per GPU
on an 8-GPU machine
- Availability
- Available now · 1 region
- Billing
- By the second on most GPUs, a few by the minute
- 8+ GPUs or a month
- Launch any size yourself, or prepay a machine for 1 to 12 months at a discount. Custom quotes on request.
The card
192 GB of HBM3. At 5.3 TB/s.
- Memory
- 192 GB
- HBM3
- Bandwidth
- 5.3 TB/s
- Memory to compute
- FP16 tensor
- 1307.4
- TFLOPS
- Per node
- 8 GPUs
- Infinity Fabric, 750 W each
Workloads
What teams run on the MI300X. Each one opens the launch form, sized.
Llama 405B InferenceFP8 weights (405 GB) fit across 4x MI300X (768 GB)Launch itDeepSeek-V3 ServingFP8 weights (671 GB) fit on one 8x MI300X nodeLaunch itTrain Mixtral 8x22BFull training state (about 2.3 TB) spans two 8x nodesLaunch itLong-context Inference192 GB per GPU leaves more room for long-context KV cacheLaunch it
Launch it
Ask your agent for an MI300X. Or read the call it makes.
Claude Code or Codex runs it for you: it starts the instance, copies your files over and runs the job. The SDKs make the same request.
Connect your agent first
Add Runcrate's MCP server to yourself: run curl -fsSL "https://runcrate.ai/mcp.md" and follow it.
Works inClaude CodeCodexOpenCodeClineRoo CodeKilo CodeCursor
Start an MI300X on Runcrate called my-gpu, copy train.py to it and run it.
create_instance { name: "my-gpu", gpu: "MI300X" }instance_status { id: "my-gpu" }file_upload { id: "my-gpu", content: "<train.py>", path: "/root/workspace/train.py" }ssh_execute { id: "my-gpu", command: "nohup python /root/workspace/train.py > train.log 2>&1 &" }Spec sheet
How the MI300X compares. Against its closest alternatives.
Alternatives
Alternatives to the MI300X. Same account, different price and performance.
| GPU | Memory · FP16 tensor | Stands out | Price gap | From |
|---|---|---|---|---|
| AMD MI300XThis GPU | 192 GB HBM3 · 1307.4 TFLOPS | $4.85/hr | ||
| NVIDIA GH200 | 96 GB HBM3 · 990 TFLOPS | -96 GB memory | Not in stock right now | |
| NVIDIA H200 | 141 GB HBM3e · 989 TFLOPS | -51 GB memory | +6% | $5.13/hr |
| NVIDIA H100 | 80 GB HBM3 · up to 989 TFLOPS | -112 GB memory | -19% | $3.91/hr |
| NVIDIA H200 NVL | 141 GB HBM3e · 835 TFLOPS | -51 GB memory | -4% | $4.67/hr |
Full specs
AMD MI300X specifications.
Memory
- Memory
- 192 GB
- Type
- HBM3
- Bandwidth
- 5.3 TB/s
Compute
- FP32
- 163 TFLOPS
- FP16 tensor
- 1307.4 TFLOPS
- INT8
- 2610 TOPS
- Stream processors
- 19,456
Power and form
- TDP
- 750 W
- Form factor
- OAM
- Architecture
- CDNA 3
- Released
- 2023
Node
- GPUs per node
- 8
- Interconnect
- Infinity Fabric
- Maker
- AMD
Questions
Renting the MI300X. What buyers ask first.
Other GPUs
Browse every GPU by architecture.
NVIDIA Blackwell
NVIDIA Hopper
NVIDIA Ampere
NVIDIA Ada Lovelace
NVIDIA Volta