lonewolf0622/HelixSR v3.2.1

HelixSR 3.2.1

on GitHub

Pre-release for testing. Latest stays 3.1.0.

Only tested on RDNA 1 (BC-250) under Linux. It may not work on other GPUs.

New

• Preset K for HLX DLSS (optional): sharper than Model E and a little slower. The setup can build it on your PC; pick it in HelixSR Scaler's DLSS Upscaler list. Tested on AMD RDNA GPUs only.
• HLX FSR 4 Full: the full FSR 4.1.2 network, as its own line in the Scaler's upscaler list. BC-250, RDNA 1 and RDNA 2 under Linux; only the BC-250 is tested. One DLL now.
• DeepReuse 1:2 and 1:4 for Preset K and HLX FSR 4 Full (not on RDNA 2): faster, with some loss of quality in motion. On a BC-250 at 1080p Quality: Preset K 2.54 / 1.96 / 1.66 ms, Full about 2.15 / 1.81 / 1.63 ms. Exact stays the default.
• HelixSR Scaler 2.2.0: based on the latest OptiScaler.

Fixes

• Newer Proton versions: HLX DLSS Model E times its two decoder forms at startup and keeps the faster. On Proton Experimental at 1080p Quality on a BC-250: 1.09 ms, was 1.51 ms. Proton 11.0 and Hotfix stay at 1.02 ms.
• HLX FSR 4 on a BC-250: no longer ends the game under Proton 11.0, and uses tier 0 first after one GPU hang with tier 7.
• A shader set that ends a game while it is built is left out on the next start.
• Setup and HelixSR Desktop reuse builds they already made.

Known issues

• A game can freeze at startup with Preset K when the Steam overlay is on. Turn the overlay off for that game if it does.
• Preset K with 1:2 or 1:4: one GPU hang on a BC-250 at 1440p with 1:2, cause not found. Exact is the safe choice.
• HLX FSR 4 Full with 1:2 or 1:4: some ghosting in motion, more at 1:4. Exact, the default, has none.
• Preset K can sparkle on foliage at any setting. HLX FSR 4 Full and Model E are steadier there.

Not tested in a game on this build

AMD's plain FSR entry, frame generation, and the final builds of Model E, HLX FSR 4 and HLX FSR 4 Full (their code was tested in replays).

Files

HelixSR-3.2.1.zip, HelixSR Desktop 1.4.0 (Windows, Linux), SHA256SUMS.txt.