We’ve loved seeing what developers are building with our cloud browsers, especially across regulated industries. As these agents take on heavier, more resource-intensive websites, raw browser performance becomes the bottleneck, so we’re focused on making Kernel feel faster everywhere to unlock a lot more use cases. Also, it’s just fun.
Over the past few months, we’ve been tinkering with our open source project, Hypeman, pushing into territory that hasn’t really been explored in the open. We’re now ready to give a sneak peek at one of those explorations. Today, we’re launching gpu Acceleration for Kernel browsers as a limited research preview.
Why your browser wants a GPU
GPU acceleration lets you attach a virtual GPU to a cloud browser, giving your agent access to more of the internet instead of being limited to CPU-friendly sites. It’s basically hitting turbo on your browser. In our internal benchmarks, this makes browsing feel faster pretty much everywhere on the internet, but it shines on live view streaming, canvas-heavy applications, or more generally sites that rely on WebGL. Here’s a fun WebGL Aquarium benchmark we stumbled on the internet and used for testing our new 60 fps performance (6× faster than our regular browsers).
Many providers that don’t use gpu acceleration will render this same page as a single static frame. Yes, really, go try it yourself.
A quick tour of why GPUs meant moving past Firecracker
We’re happy about where we’ve landed with these gpu accelerated browsers. Getting there, though, meant revisiting some core infrastructure choices we’d been happily running for a while now. Firecracker was and still is a great fit for our lightweight cloud browsers, but it hits a hard ceiling for gpu acceleration. It has no device passthrough at all and explicitly does not support vGPU passthrough, so we had to build a brand new data plane around QEMU for these GPU accelerated browsers instead. We also spent time looking at Cloud Hypervisor, and Hypeman still supports it, but we found the ecosystem and our early experiments didn’t get us as far as QEMU did with vGPUs. So today gpu accelerated browsers run on QEMU.
So… how can you get your hands on this?
We’re actively iterating on the project and partnering with teams that want to run GPU‑powered browser agents in production. If you’d like to try the research preview, or just play doom 3 compiled to wasm on kernel, sign up for a Startup or Enterprise plan and we’ll get you early access. You can also learn more in our docs.
