Intel Quietly Sunset Its PlaidML Open-Source Deep Learning Software
Another hit to the open-source Intel software ecosystem this year was the company formally archiving/discontinuing work on the PlaidML deep learning software. PlaidML was the deep learning framework that Intel acquired back in 2018 as part of their acquisition of Vertex.AI. PlaidML had a goal of "deep learning for every platform" but unfortunately those ambitions didn't materialize.
Amid the recent Intel layoffs/restructuring at the company and seeing what other open-source/Linux software efforts are being impacted (e.g. Intel Announces It's Shutting Down Clear Linux, Another Longtime Intel Linux Engineer Leaves The Company, Latest Intel Engineering Layoffs Lead To An Intel Linux Driver Being Orphaned, etc), I learned this week that they pulled PlaidML off life support and officially archived the open-source project.
PlaidML was developed by Vertex.AI that Intel acquired back in 2018. As part of buying out that AI startup back, Intel planned to continue developing the PlaidML open-source deep learning engine. Development continued well initially and as part of "deep learning for every platform" was running well on AMD and NVIDIA GPUs too back in the 2018~2019 period. For CPUs and GPUs PlaidML had worked well for benchmarking for a period of time. But then the project was going through a big code restructuring and since then there had been little to report at large on the project... Fading away and now in 2025 is officially dead.
PlaidML during their big code overhaul was working on adopting LLVM's MLIR. Plus they had been working on other grand plans too. Last time I checked it was still in development albeit frustratingly slow from a user perspective. Fading away without any major releases to report while the AI world has been racing on.
Now this week finding out it's indeed no longer being developed by Intel. The PlaidML GitHub notes the project was archived back on 29 March of this year. The former ai.intel.com/plaidml product page also now just redirects to one of Intel's generic AI web pages. I hadn't been able to find any other public announcement or news of Intel ending PlaidML development since the March archival. A quiet and unfortunate passing and another wasted acquisition.
Amid the recent Intel layoffs/restructuring at the company and seeing what other open-source/Linux software efforts are being impacted (e.g. Intel Announces It's Shutting Down Clear Linux, Another Longtime Intel Linux Engineer Leaves The Company, Latest Intel Engineering Layoffs Lead To An Intel Linux Driver Being Orphaned, etc), I learned this week that they pulled PlaidML off life support and officially archived the open-source project.
PlaidML was developed by Vertex.AI that Intel acquired back in 2018. As part of buying out that AI startup back, Intel planned to continue developing the PlaidML open-source deep learning engine. Development continued well initially and as part of "deep learning for every platform" was running well on AMD and NVIDIA GPUs too back in the 2018~2019 period. For CPUs and GPUs PlaidML had worked well for benchmarking for a period of time. But then the project was going through a big code restructuring and since then there had been little to report at large on the project... Fading away and now in 2025 is officially dead.
PlaidML during their big code overhaul was working on adopting LLVM's MLIR. Plus they had been working on other grand plans too. Last time I checked it was still in development albeit frustratingly slow from a user perspective. Fading away without any major releases to report while the AI world has been racing on.
Now this week finding out it's indeed no longer being developed by Intel. The PlaidML GitHub notes the project was archived back on 29 March of this year. The former ai.intel.com/plaidml product page also now just redirects to one of Intel's generic AI web pages. I hadn't been able to find any other public announcement or news of Intel ending PlaidML development since the March archival. A quiet and unfortunate passing and another wasted acquisition.