Stretch: run Slipstream in 2 GB of VRAM (iGPUs / very low spec) -- define a very-low tier and the features it cuts #408

New issue

Open

opened 2026-10-05 03:23:41 +00:00 by sky_line · 0 comments

sky_line commented

2026-10-05 03:23:41 +00:00

Owner

Part of #405. Stretch goal.

Make Slipstream run usably within 2 GB of VRAM: integrated GPUs (Intel Iris Xe / Arc iGPU, AMD Radeon 680M/780M/890M, Steam Deck-class APUs) and older 2 GB discrete cards. Integrated GPUs also share system RAM, so the RAM side (#405 part B) matters just as much.

This needs a defined "very low memory" tier with features cut or reduced. Under the #405 rule it's an explicit tier or preset, either chosen by the user or offered at first run on hardware detection, never a silent change.

Budget sketch (1080p or 900p, to be measured)

Fixed: swapchain + core targets ~150-300 MB, depending on resolution and HDR.
Geometry: ~300-500 MB in a busy region; needs tighter mesh LOD and distance.
Textures: ~600-900 MB at a coarse mip level.
Everything else: must fit in what's left. That's the work.

Candidate cuts and reductions (decide which, and measure each)

Ray tracing off (no BLAS/TLAS, no RT mask). This alone frees 0.7-2 GB.
Reflection probes: a single low-resolution probe or the sky only; no realtime probes; no hero/mirror planar reflections.
Shadows: sun only at 1K, or off; no projector shadows.
Render scale: FSR performance mode, rendering at about 50-67% internal resolution.
Post-process: no SSAO / DoF / bloom chains (or half-res); FXAA instead of TAA history buffers.
Textures: a max texture resolution cap (512-1024), with BC compression always on.
Geometry and avatars: shorter draw distance; impostors or jelly-dolls beyond a short range; lower mesh LOD factor; cap fully-rendered avatars.
Cull and instance buffers: start small and grow on demand.
Media: cap media texture count and resolution.
GPU-driven buffers: check that the persistent SSBOs, cull lists and staging rings scale down, with no fixed large minimums.

Also needed

Hardware detection that offers the tier on first run (VRAM below ~3 GB or an integrated GPU), with the user choosing.
RAM: integrated GPUs share RAM, so serve textures from the GPU-ready disk cache (#274, #405 part B) and keep RSS low.
A test target such as a Steam Deck / 780M laptop or a 2 GB card, using the low-memory benchmark preset from #163.

Done when

A defined very-low tier with a documented feature list.
A busy-region benchmark stays under 2 GB of process VRAM without thrashing at 30 FPS or better on a reference iGPU.
Offered at first run on qualifying hardware; never applied silently.

Part of #405. **Stretch goal.** Make Slipstream run usably within **2 GB of VRAM**: integrated GPUs (Intel Iris Xe / Arc iGPU, AMD Radeon 680M/780M/890M, Steam Deck-class APUs) and older 2 GB discrete cards. Integrated GPUs also share system RAM, so the RAM side (#405 part B) matters just as much. This needs a defined **"very low memory" tier** with features cut or reduced. Under the #405 rule it's an explicit tier or preset, either chosen by the user or offered at first run on hardware detection, never a silent change. ## Budget sketch (1080p or 900p, to be measured) - **Fixed:** swapchain + core targets ~150-300 MB, depending on resolution and HDR. - **Geometry:** ~300-500 MB in a busy region; needs tighter mesh LOD and distance. - **Textures:** ~600-900 MB at a coarse mip level. - **Everything else:** must fit in what's left. That's the work. ## Candidate cuts and reductions (decide which, and measure each) - **Ray tracing off** (no BLAS/TLAS, no RT mask). This alone frees 0.7-2 GB. - **Reflection probes:** a single low-resolution probe or the sky only; no realtime probes; no hero/mirror planar reflections. - **Shadows:** sun only at 1K, or off; no projector shadows. - **Render scale:** FSR performance mode, rendering at about 50-67% internal resolution. - **Post-process:** no SSAO / DoF / bloom chains (or half-res); FXAA instead of TAA history buffers. - **Textures:** a max texture resolution cap (512-1024), with BC compression always on. - **Geometry and avatars:** shorter draw distance; impostors or jelly-dolls beyond a short range; lower mesh LOD factor; cap fully-rendered avatars. - **Cull and instance buffers:** start small and grow on demand. - **Media:** cap media texture count and resolution. - **GPU-driven buffers:** check that the persistent SSBOs, cull lists and staging rings scale down, with no fixed large minimums. ## Also needed - Hardware detection that offers the tier on first run (VRAM below ~3 GB or an integrated GPU), with the user choosing. - RAM: integrated GPUs share RAM, so serve textures from the GPU-ready disk cache (#274, #405 part B) and keep RSS low. - A test target such as a Steam Deck / 780M laptop or a 2 GB card, using the low-memory benchmark preset from #163. ## Done when - [ ] A defined very-low tier with a documented feature list. - [ ] A busy-region benchmark stays under 2 GB of process VRAM without thrashing at 30 FPS or better on a reference iGPU. - [ ] Offered at first run on qualifying hardware; never applied silently.

sky_line added the

perf

platform

rendering

labels

2026-10-05 03:23:41 +00:00

sky_line added a new dependency

2026-10-05 03:23:41 +00:00

#405 Epic: RAM/VRAM management -- one budget, tiered texture storage, run on a stated minimum

sky_line referenced this issue

2026-10-05 03:23:41 +00:00

Epic: RAM/VRAM management -- one budget, tiered texture storage, run on a stated minimum #405