TUNEDb443rel/engines-tools4 min read

Valve plans FEX code pre-caching for Arm game launches

Valve has enabled FEX caching for Proton Experimental on Arm and plans Steam Pre-Caching integration. A 1GB default cap, missing first-run data and no release date remain open issues.

Valve wants Steam to squash Arm gaming stutter before you even hit play
Valve wants Steam to squash Arm gaming stutter before you even hit playaxonite / pixabay

Patch notes

  • Valve enabled FEX’s persistent on-disk JIT cache in Proton Experimental for Arm after the September release.

  • Griffais says the cache may improve 1% low frame rates, particularly on repeat runs.

  • A 60fps frame leaves about 16.7 milliseconds for all work.

  • The October FEX update set a configurable cache limit that defaults to 1GB.

  • Steam Pre-Caching integration has no announced date or published performance data.

By Elena Vasquez

Valve enabled FEX’s persistent on-disk JIT cache in Proton Experimental for Arm after its September debut, with potential gains in 1% low frame rates on repeat runs.

Valve plans to integrate that cache with Steam Pre-Caching, software engineer Pierre-Loup Griffais told PCWorld, so translated code can exist before an application’s first local run.

How FEX shifts translation work

Most Windows games target x86 processors from AMD and Intel. Arm chips use a different instruction set. FEX translates x86 code into Arm instructions as a game runs, allowing existing Windows builds to work across the two architectures.

That translation does not cover the whole system. “When provided with a proper ARM execution environment, such as when integrated with Proton ARM, FEX only translates the CPU code of the application,” Griffais explained.

Proton can switch to native Arm execution at the Win32 API boundary. The operating system and graphics drivers run natively too, leaving FEX responsible for the game’s CPU instructions.

Griffais says translation overhead is typically modest, especially in GPU-bound games. CPU work can still disrupt play. A 60fps frame has about 16.7 milliseconds to complete, and translating unfamiliar code mid-game can consume enough time to cause a hitch.

The cache stores translated code for reuse instead of compiling it again. Its on-disk form preserves that work between sessions, allowing later runs to avoid some processing that the first run required.

FEX developers introduced the persistent JIT cache in September. Valve then enabled it in Proton Experimental for Arm. Griffais pointed to potential gains in 1% low frame rates, particularly on repeat runs.

Those lows expose slow frames that an average frame-rate number can hide. They do not capture every source of stutter, but they offer a more revealing measure than a high average alone.

What does this change for studios?

“FEX code caching is vital for frame time consistency,” Griffais told PCWorld. He said the future Steam Pre-Caching integration would make cached code available even before an application’s first local run.

For studios, the change sits in Steam’s compatibility and distribution layer. Valve can prepare x86-to-Arm translation artifacts for Windows builds, rather than leaving the work for the opening moments of play.

The approach supports Valve’s wider goal of running thousands of Windows games on processors they were not designed for. The mechanism translates x86 Windows code rather than replacing every title with a native Arm executable. Hardware, driver and game compatibility still determine actual results.

However, Valve has announced neither a timetable nor performance data for Pre-Caching. Until the integration ships, the existing cache offers its clearest benefit on repeat runs. Under the current cache, the first local run still builds translation data.

Any useful studio evaluation must cover frame-time consistency before and after the first launch, battery impact and cache stability.

A broader push to move work before play

Valve is not alone in moving compilation work away from gameplay. Microsoft and Nvidia’s Advanced Shader Delivery gives supported games shaders compiled for their particular hardware configuration, rather than compiling everything locally.

Intel has also worked on its Graphics Shader Distribution Service to reduce graphics-driver compilation delays. These systems handle GPU shaders. FEX caches translated CPU code, so the mechanisms are related but not interchangeable.

None of these systems removes every possible hitch. Games can still stutter through asset streaming, CPU bottlenecks or poor optimization. Their shared principle is straightforward: finish expensive work before it interrupts a frame.

Steam gives Valve an established route for that work. The platform already delivers games to players, and Valve has spent years building shader pre-caching infrastructure for SteamOS. Adding translated CPU code could make Arm systems feel less like a compatibility experiment.

The target hardware includes Qualcomm’s Snapdragon Elite chips and Nvidia’s RTX Spark platform. Valve’s work may ease a specific CPU-translation problem on those machines, but it will not make every Arm GPU equivalent to a desktop graphics system.

Reliability remains unfinished

The first September cache release had rough edges. Caches could grow without restriction, and stale entries did not always disappear when game files changed.

The October FEX update addressed several problems. It added:

  • a configurable cache limit that defaults to 1GB;
  • cleanup of outdated caches after FEX updates; and
  • an optional memory buffer for frequently accessed code.

Griffais also warned that Arm GPUs can differ substantially from the desktop GPUs games target. New graphics features may carry performance costs. CPU translation can also affect battery life and power consumption, even when its frame-rate impact remains modest.

Independent testing remains limited. No comprehensive data yet shows how much FEX caching improves gameplay across different Arm chips. Valve has not publicly demonstrated its Steam Pre-Caching integration either.

Valve spent nearly a decade developing FEX until it could run modern PC games with relatively little translation overhead. The remaining engineering task is to make that overhead largely disappear before players notice it.

The next market signal will be Valve’s first public Pre-Caching demonstration and independent frame-time data, including first-run, repeat-run and battery results. Until those measurements arrive, the integration remains a credible plan without a delivery date.

via pcworld.com (Source)

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Staff writer covering regulation, policy and their impact on video game industry at Game Dev Wire.

248 articles