AMD Developer Cloud - Free MI300X GPU Hours
Source: https://www.amd.com/en/developer/resources/cloud-access/amd-developer-cloud.html
Description
Create account to comment on specific lines or Sign in
+ 1 The AMD Developer Cloud provides free access to AMD Instinct MI300X GPUs with 192 GB HBM3 memory -- one of the most powerful accelerators available. New developers get 25 free hours of compute time, and you can earn up to 50 additional hours by passing the ROCm Star Application Developer Certificate. The platform runs on DigitalOcean infrastructure, supports single-GPU or 8-GPU (1,536 GB total VRAM) configurations, and comes pre-configured with ROCm, PyTorch, vLLM, and other AI frameworks.
No comments on this line yet.
+ 2
No comments on this line yet.
+
3
No comments on this line yet.
+ 4
No comments on this line yet.
+ 6
No comments on this line yet.
+ 7 1. Go to AMD Developer Cloud
No comments on this line yet.
+ 8 2. Click the link to access the cloud -- you'll be redirected to devcloud.amd.com (hosted on DigitalOcean infrastructure)
No comments on this line yet.
+ 9 3. Create a new account or log into an existing DigitalOcean account
No comments on this line yet.
+ 10 4. Complete the Request Credit form with your developer profile information
No comments on this line yet.
+ 11 5. AMD reviews applications and grants credits at their discretion to qualified developers
No comments on this line yet.
+ 12 6. Once approved, you'll receive 25 hours of free MI300X compute credits deposited to your account
No comments on this line yet.
+ 13 7. Credits appear in your dashboard and you can immediately launch GPU instances
No comments on this line yet.
+ 14
No comments on this line yet.
+ 15 Important:
No comments on this line yet.
+ 16 • Approval is not automatic -- AMD grants credits at their sole discretion
No comments on this line yet.
+ 17 • Credits expire 10 days from the date of deposit (unless stated otherwise in your confirmation email)
No comments on this line yet.
+ 18 • The 10-day expiration is aggressive -- plan your work before requesting credits
No comments on this line yet.
+ 19 • For questions or issues, contact: devcloudrequests@amd.com
No comments on this line yet.
+ 20
No comments on this line yet.
+
21
No comments on this line yet.
+ 22
No comments on this line yet.
+ 24
No comments on this line yet.
+ 25 ConfigurationGPU MemoryDetails
No comments on this line yet.
+ 26 1x MI300X192 GB HBM3Single GPU instance, $1.99/hour retail rate
No comments on this line yet.
+ 27 8x MI300X1,536 GB HBM3 (8 x 192 GB)Multi-GPU cluster, $15.92/hour retail rate
No comments on this line yet.
+ 28
No comments on this line yet.
+ 29 SpecValue
No comments on this line yet.
+ 30 ArchitectureAMD CDNA 3
No comments on this line yet.
+ 31 Memory TypeHBM3
No comments on this line yet.
+ 32 Memory per GPU192 GB
No comments on this line yet.
+ 33 Memory Bandwidth5.3 TB/s
No comments on this line yet.
+ 34 FP16 Performance1,307 TFLOPS
No comments on this line yet.
+ 35 Host CPUIntel Xeon Platinum 8568Y+ (Emerald Rapids)
No comments on this line yet.
+ 36 OSUbuntu 24.04 LTS
No comments on this line yet.
+ 37
No comments on this line yet.
+
38
No comments on this line yet.
+ 39
No comments on this line yet.
+ 41
No comments on this line yet.
+ 42 ComponentVersion/Details
No comments on this line yet.
+ 43 ROCm6.4.0 (pre-configured)
No comments on this line yet.
+ 44 Quick-start packagesJAX, Megatron, PyTorch, SGLang, vLLM
No comments on this line yet.
+ 45 JupyterLabAvailable for interactive development
No comments on this line yet.
+ 46 OSUbuntu 24.04 LTS
No comments on this line yet.
+ 47
No comments on this line yet.
+
48
No comments on this line yet.
+ 49
No comments on this line yet.
+ 51
No comments on this line yet.
+ 52 ResourceLimit
No comments on this line yet.
+ 53 Initial free hours25 hours MI300X
No comments on this line yet.
+ 54 Bonus hours (ROCm Star cert)Up to 50 additional hours
No comments on this line yet.
+ 55 Maximum free hours75 hours total
No comments on this line yet.
+ 56 Credit expiration10 days from deposit
No comments on this line yet.
+ 57 Application requiredYes, approval at AMD's discretion
No comments on this line yet.
+ 58 Concurrent instancesNot publicly documented
No comments on this line yet.
+ 59
No comments on this line yet.
+
60
No comments on this line yet.
+ 61
No comments on this line yet.
+ 63
No comments on this line yet.
+ 64 Earn up to 50 additional free hours by passing the ROCm Star certification:
No comments on this line yet.
+ 65
No comments on this line yet.
+ 66 • Format: 40-question online test
No comments on this line yet.
+ 67 • Topics covered: HIP programming, OpenMP offloading, ROCm software-accelerated libraries, porting CUDA applications to ROCm, application design across diverse hardware
No comments on this line yet.
+ 68 • Reward: Up to 50 additional free Developer Cloud hours upon passing
No comments on this line yet.
+ 69 • Purpose: Validates your ability to develop, refactor, and port applications for AMD GPU hardware
No comments on this line yet.
+ 70
No comments on this line yet.
+
71
No comments on this line yet.
+ 72
No comments on this line yet.
+ 74
No comments on this line yet.
+ 75 • Plan before requesting credits: The 10-day expiration is the biggest gotcha. Have your project, code, and data ready before you apply. Don't request credits "just to try it out" unless you can dedicate time immediately
No comments on this line yet.
+ 76 • 192 GB VRAM is massive: This is enough to run very large models (70B+ parameters) at full precision without quantization. Take advantage of this for workloads that won't fit on consumer GPUs
No comments on this line yet.
+
77
• ROCm compatibility: Most PyTorch code works out of the box on ROCm, but CUDA-specific kernels need porting via HIP. Use torch.cuda.is_available() -- ROCm maps to the CUDA API
No comments on this line yet.
+ 78 • vLLM works great on MI300X: The pre-installed vLLM package is optimized for MI300X and can serve large language models efficiently
No comments on this line yet.
+ 79 • Noisy neighbors warning: AMD engineers have cautioned that the shared infrastructure may have variable performance. Don't use it for benchmarking
No comments on this line yet.
+ 80 • ROCm version may lag: The pre-installed ROCm 6.4.0 may not be the latest stable release. Check if newer versions are available if you hit compatibility issues
No comments on this line yet.
+ 81 • Host CPU is Intel, not AMD: Despite being AMD's cloud, the host processors are Intel Xeon -- this doesn't affect GPU workloads but is worth noting
No comments on this line yet.
+ 82 • India-specific program: AMD announced 100,000 free hours specifically for Indian researchers and startups -- if you qualify, this is a much larger allocation
No comments on this line yet.
+ 83 • Combine with ROCm Star: Take the certification first to maximize your free hours to 75 total before the 10-day clock starts
No comments on this line yet.
+ 84
No comments on this line yet.
+
85
No comments on this line yet.
+ 86
No comments on this line yet.
+ 87 Sources:
No comments on this line yet.
+ 88 • AMD Developer Cloud
No comments on this line yet.
+ 89 • Introducing the AMD Developer Cloud - AMD Blog
No comments on this line yet.
+ 90 • ROCm Star Application Developer Certificate
No comments on this line yet.
+ 91 • AMD ROCm 7 and Developer Cloud Announcement
No comments on this line yet.
+ 92 • AMD Developer Cloud Review - Phoronix
No comments on this line yet.
+ 93 • AMD 100k Free Hours for Indian Developers
No comments on this line yet.
+ 94 • Complimentary AMD Developer Cloud Access - PyTorch Forum
No comments on this line yet.