# \#optimization **URL:** https://devcommunity.amd.com/tag/optimization/117.md [Latest](https://devcommunity.amd.com/latest.md) · [Categories](https://devcommunity.amd.com/categories.md) · [Tags](https://devcommunity.amd.com/tags.md) --- ## [ROCm 10.1: Breaking the Data-Movement Bottleneck - Your training runs aren't compute-bound anymore, they're traffic-bound](https://devcommunity.amd.com/t/rocm-10-1-breaking-the-data-movement-bottleneck-your-training-runs-arent-compute-bound-anymore-theyre-traffic-bound/1191)
**Author:** [@Uncy\_AMD](https://devcommunity.amd.com/u/Uncy_AMD)\ **Replies:** 0\ **Last updated:** [October 14, 2026, 2:41pm UTC](https://devcommunity.amd.com/t/rocm-10-1-breaking-the-data-movement-bottleneck-your-training-runs-arent-compute-bound-anymore-theyre-traffic-bound/1191 "2026-10-14T14:41:52Z")
What’s in it: ROCm 10.1 goes after the thing quietly capping your throughput: data movement, not raw FLOPs. This point release ships a broad slate of compiler, library, and tooling improvements aimed squarely at the m… --- ## [SyntheDrive: Bare-Metal Memory Pipeline Optimization Engine for AMD SEV-SNP](https://devcommunity.amd.com/t/synthedrive-bare-metal-memory-pipeline-optimization-engine-for-amd-sev-snp/1187)
**Author:** [@Kalu\_Abbott\_Uduma](https://devcommunity.amd.com/u/Kalu_Abbott_Uduma)\ **Replies:** 2\ **Last updated:** [October 13, 2026, 3:23pm UTC](https://devcommunity.amd.com/t/synthedrive-bare-metal-memory-pipeline-optimization-engine-for-amd-sev-snp/1187 "2026-10-13T15:23:03Z")
Project Architecture Status: Closed-Source Enterprise Framework / Open-Source Benchmark Harness Cryptographic memory isolation inside TEEs like AMD SEV-SNP introduces an intense performance tax. Hardware encryption cont… --- ## [Video Rater: rating rendering artifacts the way players actually see them](https://devcommunity.amd.com/t/video-rater-rating-rendering-artifacts-the-way-players-actually-see-them/1176)
**Author:** [@Uncy\_AMD](https://devcommunity.amd.com/u/Uncy_AMD)\ **Replies:** 0\ **Last updated:** [October 9, 2026, 4:57pm UTC](https://devcommunity.amd.com/t/video-rater-rating-rendering-artifacts-the-way-players-actually-see-them/1176 "2026-10-09T16:57:20Z")
Most video quality metrics (PSNR, VMAF) were built for compression in filmed video, so they miss a lot of what goes wrong with temporal upscaling and anti-aliasing in games: blur, ghosting, flickering, aliasing, moir… --- ## [Enabling DeepSeek-V4-Flash Training on AMD Instinct MI355X GPUs with Primus](https://devcommunity.amd.com/t/enabling-deepseek-v4-flash-training-on-amd-instinct-mi355x-gpus-with-primus/1041)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [September 17, 2026, 12:02pm UTC](https://devcommunity.amd.com/t/enabling-deepseek-v4-flash-training-on-amd-instinct-mi355x-gpus-with-primus/1041 "2026-09-17T12:02:49Z")
DeepSeek-AI released the DeepSeek-V4 series on April 24, 2026: a preview pair of MIT-licensed Mixture-of-Experts models, with DeepSeek-V4-Flash at 284B total parameters (13B activated) and a one-million-token context win… --- ## [Optimizing ATOM and vLLM-ATOM for High-Interactivity Inference](https://devcommunity.amd.com/t/optimizing-atom-and-vllm-atom-for-high-interactivity-inference/944)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [September 2, 2026, 9:49am UTC](https://devcommunity.amd.com/t/optimizing-atom-and-vllm-atom-for-high-interactivity-inference/944 "2026-09-02T09:49:34Z")
When you type a question into a chat assistant, ask a coding agent to make an edit, or watch tokens stream back from a model, what you feel is not throughput. You feel responsiveness. The pause before the first word appe… --- ## [ROCm 7.14: TheRock Goes Production and Expands AMD’s AI Software Platform](https://devcommunity.amd.com/t/rocm-7-14-therock-goes-production-and-expands-amd-s-ai-software-platform/729)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [July 20, 2026, 12:35pm UTC](https://devcommunity.amd.com/t/rocm-7-14-therock-goes-production-and-expands-amd-s-ai-software-platform/729 "2026-07-20T12:35:19Z")
In this blog, you will explore the key updates and improvements that make the ROCm 7.14 release a major milestone in the evolution of AMD’s open source AI software ecosystem. At the heart of this release is the produc… --- ## [Serving NVFP4 Models on AMD Instinct™ MI355 Accelerators](https://devcommunity.amd.com/t/serving-nvfp4-models-on-amd-instinct-mi355-accelerators/726)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [July 20, 2026, 12:30pm UTC](https://devcommunity.amd.com/t/serving-nvfp4-models-on-amd-instinct-mi355-accelerators/726 "2026-07-20T12:30:39Z")
NVFP4 is an increasingly common deployment format: NVIDIA, AMD, and the open-source community have published NVFP4 quantized checkpoints of frontier models such as moonshotai/Kimi-K2.6, and many users want to deploy t… --- ## [SGLang-ATOM: Bring ROCm-Native Acceleration to SGLang Serving](https://devcommunity.amd.com/t/sglang-atom-bring-rocm-native-acceleration-to-sglang-serving/711)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [July 14, 2026, 11:53am UTC](https://devcommunity.amd.com/t/sglang-atom-bring-rocm-native-acceleration-to-sglang-serving/711 "2026-07-14T11:53:15Z")
Large language model serving teams often face two competing goals: keeping the flexibility and developer velocity of an ecosystem serving framework, while also reaching strong throughput, latency, and cost efficiency … --- ## [Primus Tuning Agent: Closing the Configuration-Search Loop](https://devcommunity.amd.com/t/primus-tuning-agent-closing-the-configuration-search-loop/710)
**Author:** [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\ **Replies:** 0\ **Last updated:** [July 14, 2026, 11:50am UTC](https://devcommunity.amd.com/t/primus-tuning-agent-closing-the-configuration-search-loop/710 "2026-07-14T11:50:52Z")
Our earlier companion blog, Primus Projection: Estimate Memory and Performance Before You Train, showed how to answer two questions about a single training configuration before launching a run: “Will it fit?” and “How…