# Cloud & Edge AI

**URL:** https://devcommunity.amd.com/c/ai-full/cloud-edge-ai/141.md

[Latest](https://devcommunity.amd.com/latest.md) · [Categories](https://devcommunity.amd.com/categories.md) · [Tags](https://devcommunity.amd.com/tags.md)

---

## [Fireworks credits](https://devcommunity.amd.com/t/fireworks-credits/704)

<div class="topic-metadata">

**Author:** [@Rashid\_Hussain](https://devcommunity.amd.com/u/Rashid_Hussain)\
**Replies:** 3\
**Last updated:** [November 22, 2026, 1:27am UTC](https://devcommunity.amd.com/t/fireworks-credits/704 "2026-11-22T01:27:12Z")

</div>

the three days to go but i didn’t recieve the promo code for to collect the fireworks credits

---

## [\[Video Tutorial\] Using the AMD Developer Cloud in Minutes](https://devcommunity.amd.com/t/video-tutorial-using-the-amd-developer-cloud-in-minutes/274)

<div class="topic-metadata">

**Author:** [@Mithun\_AMD](https://devcommunity.amd.com/u/Mithun_AMD)\
**Replies:** 5\
**Last updated:** [November 13, 2026, 5:39pm UTC](https://devcommunity.amd.com/t/video-tutorial-using-the-amd-developer-cloud-in-minutes/274 "2026-11-13T17:39:13Z")

</div>

Get instant access to powerful AMD Instinct MI300X GPUs through the AMD Developer Cloud. Perfect for AI, ML, and HPC developers, run workloads in the browser with pre-installed Docker containers and Jupyter Notebooks.…

---

## [Getting started with vLLM for a small tool-using AI agent](https://devcommunity.amd.com/t/getting-started-with-vllm-for-a-small-tool-using-ai-agent/1260)

<div class="topic-metadata">

**Author:** [@Aviv\_Morad](https://devcommunity.amd.com/u/Aviv_Morad)\
**Replies:** 1\
**Last updated:** [October 27, 2026, 3:11pm UTC](https://devcommunity.amd.com/t/getting-started-with-vllm-for-a-small-tool-using-ai-agent/1260 "2026-10-27T15:11:54Z")

</div>

I’m a computer science student at Afeka working on AI and agent projects, and I’m exploring AMD Developer Cloud for a first GPU experiment. I’d like to serve an open model with vLLM and connect it to a simple tool-using…

---

## [VLLM\_ROCM\_USE\_AITER=1 makes Gemma 4 12B fp8 slower on MI300X: untuned a8w8 shapes, Triton attention](https://devcommunity.amd.com/t/vllm-rocm-use-aiter-1-makes-gemma-4-12b-fp8-slower-on-mi300x-untuned-a8w8-shapes-triton-attention/1181)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 5\
**Last updated:** [October 11, 2026, 3:20pm UTC](https://devcommunity.amd.com/t/vllm-rocm-use-aiter-1-makes-gemma-4-12b-fp8-slower-on-mi300x-untuned-a8w8-shapes-triton-attention/1181 "2026-10-11T15:20:15Z")

</div>

A follow-up to the Gemma 4 size sweep on one MI300X. I ran Gemma 4 12B in fp8 twice on the same Developer Cloud droplet, the same day and the same image (vllm/vllm-openai-rocm nightly, vLLM 0.31.1rc1.dev23): once stock a…

---

## [Gemma 4 from E2B to 31B on one MI300X: fp8 overtakes bf16 from 12B up](https://devcommunity.amd.com/t/gemma-4-from-e2b-to-31b-on-one-mi300x-fp8-overtakes-bf16-from-12b-up/1174)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 0\
**Last updated:** [October 9, 2026, 2:16pm UTC](https://devcommunity.amd.com/t/gemma-4-from-e2b-to-31b-on-one-mi300x-fp8-overtakes-bf16-from-12b-up/1174 "2026-10-09T14:16:40Z")

</div>

A follow-up to the E2B weight-format sweep. E2B is small enough that each token reads few weight bytes, so I ran the rest of the Gemma 4 family on the same MI300X Developer Cloud droplet and the same vLLM image (vllm/vll…

---

## [Ten weight formats of Gemma 4 E2B on one MI300X: fp8 is the only one at bf16's pace](https://devcommunity.amd.com/t/ten-weight-formats-of-gemma-4-e2b-on-one-mi300x-fp8-is-the-only-one-at-bf16s-pace/1171)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 0\
**Last updated:** [October 8, 2026, 5:46pm UTC](https://devcommunity.amd.com/t/ten-weight-formats-of-gemma-4-e2b-on-one-mi300x-fp8-is-the-only-one-at-bf16s-pace/1171 "2026-10-08T17:46:57Z")

</div>

A follow-up to my earlier post on the numeric formats an MI300X executes. That post timed raw matrix multiplies. This one serves a real model: ten weight formats of Gemma 4 E2B, one after another on one MI300X Developer …

---

## [GPU Droplet Availability](https://devcommunity.amd.com/t/gpu-droplet-availability/1162)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 0\
**Last updated:** [October 7, 2026, 4:34pm UTC](https://devcommunity.amd.com/t/gpu-droplet-availability/1162 "2026-10-07T16:34:14Z")

</div>

I went to spin up a GPU droplet on the MI300X on the AMD developer cloud in Atlanta and can’t get a basic GPU setup. this was the message: We’re out of GPU’s right now. We are currently experiencing high demand and our…

---

## [Optimizing vLLM Inference on AMD GPUs with ROCm and AITER](https://devcommunity.amd.com/t/optimizing-vllm-inference-on-amd-gpus-with-rocm-and-aiter/1070)

<div class="topic-metadata">

**Author:** [@Kashif\_Syed](https://devcommunity.amd.com/u/Kashif_Syed)\
**Replies:** 0\
**Last updated:** [September 22, 2026, 7:59am UTC](https://devcommunity.amd.com/t/optimizing-vllm-inference-on-amd-gpus-with-rocm-and-aiter/1070 "2026-09-22T07:59:44Z")

</div>

Optimizing vLLM Inference on AMD GPUs with ROCm and AITER Introduction Large Language Model (LLM) inference requires efficient GPU utilization, memory management, and high-performance communication between worker process…

---

## [Which numeric formats an MI300X actually executes — measured on a Developer Cloud droplet](https://devcommunity.amd.com/t/which-numeric-formats-an-mi300x-actually-executes-measured-on-a-developer-cloud-droplet/1043)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 5\
**Last updated:** [September 18, 2026, 6:34pm UTC](https://devcommunity.amd.com/t/which-numeric-formats-an-mi300x-actually-executes-measured-on-a-developer-cloud-droplet/1043 "2026-09-18T18:34:39Z")

</div>

I spent a few days driving a single MI300X droplet on AMD Developer Cloud entirely through an MCP server, and the part worth sharing here isn’t the tooling — it’s what the card said when I stopped reading spec tables a…

---

## [Serving Gemma 4 E2B on one MI300X with vLLM](https://devcommunity.amd.com/t/serving-gemma-4-e2b-on-one-mi300x-with-vllm/1048)

<div class="topic-metadata">

**Author:** [@William\_Mclean](https://devcommunity.amd.com/u/William_Mclean)\
**Replies:** 0\
**Last updated:** [September 18, 2026, 3:05pm UTC](https://devcommunity.amd.com/t/serving-gemma-4-e2b-on-one-mi300x-with-vllm/1048 "2026-09-18T15:05:46Z")

</div>

A follow-up to my numeric-formats post, from the same Developer Cloud droplet. vLLM was pulled to the Debian Trixie image running on the Droplet to serve Gemma4 2B. Why? I used Trixie to keep my base system as consis…

---

## [AMD GPUs not available in certain regions](https://devcommunity.amd.com/t/amd-gpus-not-available-in-certain-regions/1013)

<div class="topic-metadata">

**Author:** [@muhammad\_kabeer\_khan](https://devcommunity.amd.com/u/muhammad_kabeer_khan)\
**Replies:** 1\
**Last updated:** [September 14, 2026, 11:44am UTC](https://devcommunity.amd.com/t/amd-gpus-not-available-in-certain-regions/1013 "2026-09-14T11:44:58Z")

</div>

After 10 days of continuous learning , today was about to take the first step in building my first ai agent solo only to be stopped at this screen where i found out that AMD GPUs are not available in my region . Still…

---

## [⚠️ AMD Developer Cloud is in high demand!](https://devcommunity.amd.com/t/amd-developer-cloud-is-in-high-demand/714)

<div class="topic-metadata">

**Author:** [@Ayllee\_AMD](https://devcommunity.amd.com/u/Ayllee_AMD)\
**Replies:** 5\
**Last updated:** [September 9, 2026, 8:00am UTC](https://devcommunity.amd.com/t/amd-developer-cloud-is-in-high-demand/714 "2026-09-09T08:00:26Z")

</div>

Our developer community is keeping AMD Developer Cloud busier than ever. While we support developers at AMD Advancing AI next week, approvals for AMD Cloud Credits will be delayed due to high demand. If you have not rec…

---

## [Energy per token on Radeon: 8.7× efficiency from batching, and 4 ways to measure it wrong](https://devcommunity.amd.com/t/energy-per-token-on-radeon-8-7x-efficiency-from-batching-and-4-ways-to-measure-it-wrong/783)

<div class="topic-metadata">

**Author:** [@Saad\_Kaicar](https://devcommunity.amd.com/u/Saad_Kaicar)\
**Replies:** 1\
**Last updated:** [August 28, 2026, 11:36pm UTC](https://devcommunity.amd.com/t/energy-per-token-on-radeon-8-7x-efficiency-from-batching-and-4-ways-to-measure-it-wrong/783 "2026-08-28T23:36:16Z")

</div>

I spent last week measuring the actual energy cost of local LLM inference on a Radeon PRO W7900 — joules per token, not just tokens per second. Four things bit me, and each one produced numbers that looked plausible enou…

---

## [Credit request order for Radeon cloud](https://devcommunity.amd.com/t/credit-request-order-for-radeon-cloud/702)

<div class="topic-metadata">

**Author:** [@Danila\_Permogorskii](https://devcommunity.amd.com/u/Danila_Permogorskii)\
**Replies:** 0\
**Last updated:** [July 11, 2026, 9:15am UTC](https://devcommunity.amd.com/t/credit-request-order-for-radeon-cloud/702 "2026-07-11T09:15:57Z")

</div>

Hi everyone! I’ve noticed that a new radeon cloud is becoming in ai developer portal. Now everyone has 2 gpu credits by default but it’s not possible to request them for now. You can check the message from the profile: …

---

## [Amdgpu VF init fails on managed SR-IOV GPU instance — "Doesn't get ack from pf", REQ\_INIT\_DATA unsupported, probe fails with -62](https://devcommunity.amd.com/t/amdgpu-vf-init-fails-on-managed-sr-iov-gpu-instance-doesnt-get-ack-from-pf-req-init-data-unsupported-probe-fails-with-62/553)

<div class="topic-metadata">

**Author:** [@Danila\_Permogorskii](https://devcommunity.amd.com/u/Danila_Permogorskii)\
**Replies:** 4\
**Last updated:** [May 29, 2026, 7:39pm UTC](https://devcommunity.amd.com/t/amdgpu-vf-init-fails-on-managed-sr-iov-gpu-instance-doesnt-get-ack-from-pf-req-init-data-unsupported-probe-fails-with-62/553 "2026-05-29T19:39:17Z")

</div>

Posting in case others on managed SR-IOV GPU instances hit this, and to ask about guest-VF ↔ host-PF version pairing. Setup: a cloud provider’s GPU instance exposing an AMD Instinct accelerator as an SR-IOV Virtual Func…

---

## [AMD edge devices (similar to Raspberry Pi 5 or Jetson Orin Nano) suitable for robotics applications?](https://devcommunity.amd.com/t/amd-edge-devices-similar-to-raspberry-pi-5-or-jetson-orin-nano-suitable-for-robotics-applications/415)

<div class="topic-metadata">

**Author:** [@Azmyin\_Md\_Kamal](https://devcommunity.amd.com/u/Azmyin_Md_Kamal)\
**Replies:** 4\
**Last updated:** [March 19, 2026, 4:37pm UTC](https://devcommunity.amd.com/t/amd-edge-devices-similar-to-raspberry-pi-5-or-jetson-orin-nano-suitable-for-robotics-applications/415 "2026-03-19T16:37:34Z")

</div>

Hello AMD Developer Community. Good morning. This is my first time posting in this community. I have been working on a local trajectory controller that has high SIMD like data structures. So far I have been testing it w…

---

## [How to Get Started on the AMD Developer Cloud](https://devcommunity.amd.com/t/how-to-get-started-on-the-amd-developer-cloud/273)

<div class="topic-metadata">

**Author:** [@Mithun\_AMD](https://devcommunity.amd.com/u/Mithun_AMD)\
**Replies:** 0\
**Last updated:** [January 12, 2026, 5:36pm UTC](https://devcommunity.amd.com/t/how-to-get-started-on-the-amd-developer-cloud/273 "2026-01-12T17:36:58Z")

</div>

The newly launched AMD Developer Cloud offers developers easy access to high-performance AMD Instinct™ MI300X GPUs in a cloud environment. With seamless GitHub-based login, preconfigured ROCm™ software, and Docker setups…
