Measurements
The numbers behind the performance guide. Each table is dated and says what it was taken on. Compare numbers within one table, and across tables only when they say they used the same kit version and scene. The trace kit changes its version when a case changes, and its results compare within a version only.
What each part of the picture costs, trace kit V3
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 7900 XTX (a tester's card), ms | AMD Radeon RX 9060 XT (a tester's card), ms | RX 9070 XT against RTX 5080 | RX 7900 XTX against RTX 5080 | RX 9060 XT against RTX 5080 |
|---|---|---|---|---|---|---|---|
| Whole trace, default settings | 5.99 | 23.3 | 15.1 | 37.5 | 3.9 times | 2.5 times | 6.3 times |
| Whole trace, light reuse off | 4.05 | 10.5 | 8.13 | 18.8 | 2.6 times | 2.0 times | 4.6 times |
| Whole trace, the game case (worst-case fog) | 12.4 | 49.5 | 35.6 | 77.6 | 4.0 times | 2.9 times | 6.3 times |
| Whole trace, as advised for AMD (light reuse and fog off) | 6.10 | 13.7 | 10.6 | 24.2 | 2.2 times | 1.7 times | 4.0 times |
| Light reuse | 1.94 | 12.8 | 6.92 | 18.7 | 6.6 times | 3.6 times | 9.7 times |
| Foliage (leaves, grass, other cut-outs) | 2.53 | 5.77 | 5.01 | 9.65 | 2.3 times | 2.0 times | 3.8 times |
| Fog (a worst case) | 4.49 | 20.7 | 14.6 | 31.7 | 4.6 times | 3.3 times | 7.1 times |
| Classic clouds | 2.97 | 3.88 | 4.30 | 5.13 | 1.3 times | 1.4 times | 1.7 times |
| NRD's split signal | 0.26 | 0.31 | 1.05 | 0.05 |
Measured 2026-11-09. Trace kit V3 (rt_bench, the renderer without the game). The same tester's dusk scene as the other kit sets: grassland with trees, pools and a stone spire; 1280x720 trace (what 1440p traces at Performance upscaling). Settings: Three bounces at three quarters of the pixels, light reuse through RTXDI, radiance cache, 4 light candidates; the game case adds NRD's split signal, the thickest fog (density 0.05, traced exactly) and classic clouds, and the costs of those three are taken from it; the AMD row is the game case with light reuse and fog off. 240 frames, median of 3 rounds. No upscaler or denoiser in the timing. Build: RTX 5080: rt_bench from b6bdb183d, V3 as first built, run on 8 November. RX 9070 XT, RX 7900 XTX and RX 9060 XT: V3 rebuilt with the fix for the AMD shader compile crash (640fc43da), as sent to testers, the same build on all three. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1. AMD Radeon RX 7900 XTX (a tester's card): driver AMD Software 26.10.1. AMD Radeon RX 9060 XT (a tester's card): driver AMD Software 26.10.1. A cost is the whole trace's time with the part, less the time without it. NRD's split signal, 27 ms on the RX 9070 XT in V2, now costs every card little; its row has no comparison columns, since at a millisecond or less they would say little. Fog and clouds also cost the Radeons far less than in V2, with no change of their own (see the V2 against V3 table). Light reuse is now the largest cost that only the Radeons pay: in the main path tracing pass its code costs the RX 9070 XT and RX 7900 XTX 4 to 10 ms, against about 0.1 on the 5080, whose light reuse is RTXDI's own passes. The fog cost is a worst case: the kit traces the thickest fog exactly, at a moonlit dusk that this fog all but blocks, where the game uses a faster method for fog and usually lighter fog. The RX 9060 XT (RDNA4, 32 compute units to the RX 9070 XT's 64) takes about 1.6 times the 9070 XT's time.
What the leaner NRD split did, trace kit V2 against V3
| Part | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 7900 XTX (a tester's card), ms | NVIDIA GeForce RTX 5080, ms |
|---|---|---|---|
| NRD's split signal (the game case less the game case without it) | 26.96 → 0.26 | 20.60 → 1.08 | 0.55 → 0.26 |
| Fog, the kit's worst case (the game case less the game case without fog) | 35.20 → 20.51 | 27.64 → 15.10 | 5.47 → 5.04 |
| Classic clouds (the game case less the game case without them) | 16.47 → 3.09 | 16.09 → 3.81 | 2.57 → 2.70 |
| Light reuse (default settings less light reuse off) | 8.90 → 9.67 | 4.13 → 4.44 | 0.08 → 0.13 |
| The game case, main pass | 66.58 → 43.62 | 49.75 → 31.21 | 9.82 → 9.57 |
| The game case, whole trace | 71.41 → 49.48 | 53.95 → 35.56 | 12.66 → 12.40 |
Measured 2026-11-09. Trace kit V2 (8 November) against V3 (8 and 9 November), the same scene and settings on the same cards. Compare within this table only. The same tester's dusk scene as the other kit sets; 1280x720 trace. Settings: Each cell is V2's figure, then V3's. Every row but the last is the main path tracing pass alone; each part's cost is one case less another, as named. The kit's settings as in the other kit sets. Build: V2: rt_bench from cbf6190f8 on the RTX 5080, the V2 kit as sent to testers on the Radeons. V3: rt_bench from b6bdb183d on the RTX 5080, V3 rebuilt with the fix for the AMD shader compile crash (640fc43da) on the Radeons. AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1. AMD Radeon RX 7900 XTX (a tester's card): driver AMD Software 26.10.1. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). Between V2 and V3 NRD's split signal was made leaner: it carries less through each path. Fog and clouds changed nothing of their own; V2's split, carried through every ray of the path, made each of their extra rays dearer, as the continuations explanation predicts. Two builds, so differences of a millisecond or two are within what a rebuild moves; the falls on the Radeons are not. The fog rows are the kit's worst case. The main pass leaves out light reuse's own passes, so light reuse here is only what its code adds to the main pass.
Every trace kit V3 case, RTX 5080 (whole trace / main pass)
| Part | NVIDIA GeForce RTX 5080, ms |
|---|---|
| Default settings | 5.99 / 2.88 |
| No foliage or other cut-outs | 3.46 / 1.61 |
| Every surface opaque (no transparency tests) | 5.18 / 2.29 |
| One light bounce instead of three | 5.32 / 2.44 |
| Light reuse off | 4.05 / 2.75 |
| Light reuse through VSPT's own reservoirs, not RTXDI | 7.04 / 5.16 |
| Radiance cache off | 5.35 / 3.17 |
| As the game traces with NRD: split signal, the thickest fog, classic clouds | 12.40 / 9.57 |
| The game case without fog | 7.91 / 4.53 |
| The game case without classic clouds | 9.43 / 6.87 |
| The game case without NRD's split signal | 12.14 / 9.31 |
| As advised for AMD: the game case with light reuse and fog off | 6.10 / 4.60 |
| As advised for AMD, without NRD's split signal | 5.68 / 4.11 |
Measured 2026-11-08. Trace kit V3 (rt_bench, the renderer without the game). The same tester's dusk scene as the other kit sets; 1280x720 trace. Settings: As the default row of the kit sets, plus the change named in each row. Each cell is the whole trace, then the main path tracing pass. 240 frames, median of 3 rounds; one run of the radiance-cache-off case was left out as noisy. Build: rt_bench from b6bdb183d, V3 as first built, the kit unpacked as a tester gets it, the GPU held exclusively. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). NRD's split signal costs the 5080 0.26 ms of the game case's whole trace (0.48 in V2, another build: within what a rebuild moves) and 0.42 ms at the settings advised for AMD. The game cases' fog is a worst case that overstates fog in the game: the thickest fog, traced exactly, at a moonlit dusk that this fog all but blocks, where the game uses a faster method for fog and usually lighter fog.
Every trace kit V3 case, RX 9070 XT, RX 7900 XTX and RX 9060 XT (whole trace / main pass)
| Part | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 7900 XTX (a tester's card), ms | AMD Radeon RX 9060 XT (a tester's card), ms |
|---|---|---|---|
| Default settings | 23.26 / 17.96 | 15.05 / 10.40 | 37.53 / 28.86 |
| No foliage or other cut-outs | 17.49 / 14.36 | 10.04 / 7.71 | 27.88 / 23.08 |
| Every surface opaque (no transparency tests) | 23.27 / 18.07 | 14.15 / 9.90 | 38.35 / 29.11 |
| One light bounce instead of three | 19.93 / 15.00 | 13.05 / 8.86 | 31.81 / 23.52 |
| Light reuse off | 10.49 / 8.29 | 8.13 / 5.96 | 18.80 / 15.18 |
| Light reuse through VSPT's own reservoirs, not RTXDI | 22.01 / 18.59 | 14.85 / 11.78 | 35.37 / 30.24 |
| Radiance cache off | 22.10 / 18.51 | 12.74 / 9.94 | 35.95 / 30.29 |
| As the game traces with NRD: split signal, the thickest fog, classic clouds | 49.48 / 43.62 | 35.56 / 31.21 | 77.64 / 69.09 |
| The game case without fog | 28.82 / 23.11 | 20.96 / 16.11 | 45.93 / 36.60 |
| The game case without classic clouds | 45.60 / 40.53 | 31.26 / 27.40 | 72.51 / 64.82 |
| The game case without NRD's split signal | 49.17 / 43.36 | 34.51 / 30.13 | 77.59 / 69.32 |
| As advised for AMD: the game case with light reuse and fog off | 13.67 / 11.00 | 10.64 / 8.09 | 24.21 / 20.00 |
| As advised for AMD, without NRD's split signal | 13.22 / 10.55 | 10.45 / 7.90 | 23.08 / 19.33 |
Measured 2026-11-09. Trace kit V3 (rt_bench, the renderer without the game). The same tester's dusk scene as the other kit sets; 1280x720 trace. Settings: As the default row of the kit sets, plus the change named in each row. Each cell is the whole trace, then the main path tracing pass. 240 frames, median of 3 rounds; every run kept on all three cards. Build: V3 rebuilt with the fix for the AMD shader compile crash (640fc43da), as sent to testers, the same build on all three cards. V3 as first built did not run on AMD cards. AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1. AMD Radeon RX 7900 XTX (a tester's card): driver AMD Software 26.10.1. AMD Radeon RX 9060 XT (a tester's card): driver AMD Software 26.10.1. The RX 7900 XTX (RDNA3) is faster than the RX 9070 XT (RDNA4) in every case, as in V2; the RX 9060 XT (RDNA4, 32 compute units to the 9070 XT's 64) takes about 1.6 times the 9070 XT's time. NRD's split signal now costs little on all three (the game case against the game case without it). Still to explain: without the split, V3's game case is slower than V2's (main pass 43.36 against 39.62 ms on the 9070 XT, 30.13 against 29.15 on the 7900 XTX), and light reuse's code still costs the 9070 XT and 7900 XTX 4 to 10 ms of the main pass against about 0.1 on an RTX 5080. The game cases' fog is a worst case that overstates fog in the game: the thickest fog, traced exactly, at a moonlit dusk that this fog all but blocks.
What each part of the picture costs, trace kit V2
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 7900 XTX (a tester's card), ms | RX 9070 XT against RTX 5080 | RX 7900 XTX against RTX 5080 | Note |
|---|---|---|---|---|---|---|
| Whole trace, default settings | 5.87 | 22.1 | 15.0 | 3.8 times | 2.5 times | |
| Whole trace, light reuse off | 3.97 | 10.4 | 8.19 | 2.6 times | 2.1 times | For the 9070 XT, about the hardware's difference from the 5080 |
| Light reuse | 1.90 | 11.7 | 6.77 | 6.2 times | 3.6 times | |
| Foliage (leaves, grass, other cut-outs) | 2.43 | 5.61 | 4.84 | 2.3 times | 2.0 times | |
| Fog | 4.91 | 34.9 | 26.9 | 7.1 times | 5.5 times | A worst case, which overstates fog in the game. On the game rows |
| Classic clouds | 2.79 | 17.1 | 16.4 | 6.1 times | 5.9 times | On the game rows |
| NRD's split signal | 0.48 | 27.1 | 20.6 | 56.5 times | 42.9 times | On the game rows |
Measured 2026-11-08. Trace kit V2 (rt_bench, the renderer without the game). One scene captured in the game by a tester, at dusk: grassland with trees, pools and a stone spire; 1280x720 trace (what 1440p traces at Performance upscaling). Settings: Three bounces at three quarters of the pixels, light reuse through RTXDI, radiance cache, 4 light candidates; the game rows add NRD's split signal, the thickest fog (density 0.05, traced exactly) and classic clouds. 240 frames, median of 3 rounds. No upscaler or denoiser in the timing. Build: RTX 5080: rt_bench from cbf6190f8. RX 9070 XT and RX 7900 XTX: the kit as sent to testers, the same build on both. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). AMD Radeon RX 7900 XTX (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). A cost is the whole trace's time with the part, less the time without it. On both Radeons the excess grows with the features compiled into the main path tracing shader. A tester's analysis of the pipeline AMD's driver compiled found a likely reason: AMD's Windows driver compiles VSPT's ray tracing with continuations, so what a path keeps across each ray is saved to memory and loaded back on every ray, and each feature's extra state costs far more than on NVIDIA. That is not yet confirmed by a fix. The fog cost is a worst case: the kit traces the thickest fog exactly, at a moonlit dusk that this fog all but blocks, where the game uses a faster method for fog and usually lighter fog. For fog in the game, the in-game figures are the better guide: Fog off took a tester's RX 9070 XT from about 15 to about 30 frames a second. The 5080's light reuse is RTXDI's own passes (1.7 ms) and leaves the trace itself alone.
Every trace kit V2 case, RTX 5080
| Part | NVIDIA GeForce RTX 5080, ms |
|---|---|
| Default settings | 5.87 |
| No foliage or other cut-outs | 3.44 |
| Every surface opaque (no transparency tests) | 5.08 |
| One light bounce instead of three | 5.48 |
| Light reuse off | 3.97 |
| Light reuse through VSPT's own reservoirs, not RTXDI | 6.95 |
| Radiance cache off | 5.27 |
| As the game traces with NRD: split signal, the thickest fog, classic clouds | 12.7 |
| The game case without fog | 7.75 |
| The game case without classic clouds | 9.87 |
| The game case without NRD's split signal | 12.2 |
Measured 2026-11-08. Trace kit V2 (rt_bench, the renderer without the game). The same tester's dusk scene as the other kit sets; 1280x720 trace. Settings: As the default row of the first kit set, plus the change named in each row. 240 frames, median of 3 rounds; one run each of the radiance-cache-off and game cases was left out as noisy. Build: rt_bench from cbf6190f8, the kit unpacked as a tester gets it, the GPU held exclusively. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). Whole-trace times. On the 5080 light reuse is RTXDI's own passes and leaves the trace itself alone; VSPT's own reservoirs cost more, inside the trace. The game cases' fog is a worst case that overstates fog in the game: the thickest fog, traced exactly, at a moonlit dusk that this fog all but blocks, where the game uses a faster method for fog and usually lighter fog.
Every trace kit V2 case, RX 9070 XT and RX 7900 XTX
| Part | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 7900 XTX (a tester's card), ms |
|---|---|---|
| Default settings | 22.1 | 15.0 |
| No foliage or other cut-outs | 16.5 | 10.1 |
| Every surface opaque (no transparency tests) | 22.1 | 14.0 |
| One light bounce instead of three | 19.0 | 12.9 |
| Light reuse off | 10.4 | 8.19 |
| Light reuse through VSPT's own reservoirs, not RTXDI | 20.9 | 14.5 |
| Radiance cache off | 22.3 | 12.9 |
| As the game traces with NRD: split signal, the thickest fog, classic clouds | 71.4 | 54.0 |
| The game case without fog | 36.5 | 27.0 |
| The game case without classic clouds | 54.3 | 37.5 |
| The game case without NRD's split signal | 44.3 | 33.4 |
Measured 2026-11-08. Trace kit V2 (rt_bench, the renderer without the game). The same tester's dusk scene as the other kit sets; 1280x720 trace. Settings: As the default row of the first kit set, plus the change named in each row. 240 frames, median of 3 rounds; every run kept on both cards, and each case's runs on the RX 7900 XTX within 0.4 ms of each other. Build: The kit as sent to testers, the same build on both cards. AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). AMD Radeon RX 7900 XTX (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). Whole-trace times. The RX 7900 XTX (RDNA3) shows the RX 9070 XT's (RDNA4) pattern and is faster in every case, its main path tracing pass 1.3 to 1.9 times as fast; it has 96 compute units to the 9070 XT's 64. On both, each feature compiled into the main path tracing shader costs far more than the work it adds. The game cases' fog is a worst case that overstates fog in the game: the thickest fog, traced exactly, at a moonlit dusk that this fog all but blocks.
Trace kit V1, the first AMD run
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | RX 9070 XT against RTX 5080 |
|---|---|---|---|
| Default settings | 5.78 | 22.1 | 3.8 times |
| No foliage or other cut-outs | 3.37 | 16.4 | 4.9 times |
| Every surface opaque | 5.03 | 22.1 | 4.4 times |
| One light bounce instead of three | 5.20 | 18.9 | 3.6 times |
| Light reuse off | 3.95 | 10.3 | 2.6 times |
| Radiance cache off | 5.36 | 22.3 | 4.2 times |
Measured 2026-11-08. Trace kit V1 (the first version, 6 cases). Compare within a version only. The same tester's dusk scene as the V2 sets; 1280x720 trace. Settings: The default settings of the kit, with the change named in each row. 240 frames, median over the rounds. Build: rt_bench builds differ between the two cards' runs. NVIDIA GeForce RTX 5080: driver 32.0.16.2101 (Windows driver version string). AMD Radeon RX 9070 XT (a tester's card): driver 26.10.1. Without light reuse the RX 9070 XT takes 2.6 times the RTX 5080's time, about the difference between the cards; with it, 3.8 times. Bounces past the first are also out of line.
Light reuse and fog in the game, on an RX 9070 XT, in the first test build
| Setting | AMD Radeon RX 9070 XT (a tester's card), fps | Note |
|---|---|---|
| Light reuse on, fog on | 21 | |
| Fog off | 30 | From about 15 to about 30 |
| Light reuse off and fog off | 52 | Light reuse off alone took it from about 31 to 52 |
Measured 2026-11-08. A tester's play, read from the game's frame log. A tester's own world; 2560x1440, upscaled from 1280x720 (Performance), NRD REBLUR with FSR 4.1. Settings: View distance 256. Each setting was changed in turn; the starting point of each step differs a little with where the player stood and the sky. Build: VSPT 0.4.0-dev.1 with VSPT: Runtime 0.2.0-dev.1 (test build). AMD Radeon RX 9070 XT (a tester's card): driver not recorded. Taken in the first test build, 0.4.0-dev.1, before the leaner NRD split that 0.4.0-dev.3 has. The two steps were taken one after the other, so each starts from a different frame rate. Each change took five minutes to show in this build, because the renderer kept the shaders for the old settings; that is fixed in VSPT 4.0.
View distance and video memory on AMD cards
| Setting | AMD Radeon RX 9070 XT (a tester's card), fps | AMD Radeon RX 9060 XT (a tester's card), fps | Note |
|---|---|---|---|
| View distance 1536 (the game's own) | 1 to 2 | Scene grew to 65 M vertices, video memory full, trace 763 ms a frame. 1 to 2 fps with path tracing on or off | |
| View distance 128 | 17 | 3 M vertices, 11 GB of 16 free, trace about 41 ms, light reuse and fog on. At 128 to 256, 9 to 11 GB were free | |
| A bare world | 41 | 3.2 M vertices, no plants or lamps, trace 18.5 ms | |
| A world with plants and lamps | 10 to 16 | 0.6 to 1 M vertices, trace 50 to 77 ms, 10 to 16 fps |
Measured 2026-11-08. Testers' play, read from the game's logs. The testers' own worlds; 2560x1440, FSR 4.1 Performance. Settings: NRD REBLUR. The view distance varied as listed. Build: VSPT 0.4.0-dev.1 with VSPT: Runtime 0.2.0-dev.1 (test build). AMD Radeon RX 9070 XT (a tester's card): driver not recorded. AMD Radeon RX 9060 XT (a tester's card): driver not recorded. AMD's acceleration structures and their build scratch are much larger than NVIDIA's, so a 16 GB AMD card fills sooner. The renderer's memory budget does not yet account for this on AMD.
Clouds, by quality
| Part | NVIDIA GeForce RTX 5080, ms |
|---|---|
| Clouds off | 1.83 |
| Classic clouds | 2.25 |
| Volumetric, Low | 2.63 |
| Volumetric, Medium | 2.85 |
| Volumetric, High | 3.75 |
Measured 2026-10-27. rt_bench, trace stage. terrain3, over the sea (rise 40, look down -3), made-up fair sky; 3840x2160 from 1280x720, DLSS Ray Reconstruction Ultra Performance. Settings: Four rounds of 240 frames, spread within 0.1 ms. Build: 0.2.8 development build. NVIDIA GeForce RTX 5080: driver not recorded. These are the trace stage's totals, including the sea and sky. The cost of the clouds is the difference from the first row.
Clear water
| Part | NVIDIA GeForce RTX 5080, ms | Note |
|---|---|---|
| Pond, from the eyes | 0.30 | 3.9 % of the trace |
| Open sea | 0.69 | 8 % of the trace |
Measured 2026-11-08. rt_bench, trace stage, the setting on against off. terrain3's pond from the eyes, and the open sea; 2560x1440, DLSS Ray Reconstruction. Settings: Clear water on against off. Build: development build. NVIDIA GeForce RTX 5080: driver not recorded. One more ray where the camera looks into water, so it costs nothing elsewhere.
Water at 4K, in the test build
| Part | NVIDIA GeForce RTX 5080, ms | Note |
|---|---|---|
| Clear water, pond from the eyes | 0.55 | |
| Clear water, open sea to the horizon | 1.10 | |
| Clear water, open sea looked down on | 1.58 | |
| Light in the water, pond from the eyes | 0.72 | 2.07 in 0.3.0 |
| Light in the water, open sea to the horizon | 1.42 | 3.65 in 0.3.0 |
| Light in the water, open sea looked down on | 2.30 | 5.75 in 0.3.0 |
| Near and Far water wave resolution | under 0.2 | nothing that can be measured; from a second run, of five rounds |
Measured 2026-11-12. rt_bench, trace stage, one build switched at run time, the median of three rounds. a pond from the eyes, and the open sea; 3840x2160, DLSS Ray Reconstruction, Performance (a 1920x1080 trace). Settings: Each part on against off. Build: 0.4.0-dev.7 test build. NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). Clear water is one more ray where the camera looks into water, so it costs nothing elsewhere. The trace here is at 1920x1080, 2.25 times the pixels of the guide's other tables.
The test build against 0.3.0, in the game
| Setting | NVIDIA GeForce RTX 5080, fps | Note |
|---|---|---|
| Test build, standing | 61.7 | |
| Test build, flying | 61.3 | |
| Test build, settled | 67.8 | |
| 0.3.0, standing | 61.3 | one flight |
| 0.3.0, flying | 59.9 | one flight |
| 0.3.0, settled | 68.1 | one flight |
Measured 2026-11-12. The game's frame log, flying over a generated village; traced frames a second, the median of two rounds. A generated village, standing, flying over it, and settled after the flight; 3840x2160, DLSS Ray Reconstruction, Performance (a 1920x1080 trace). Settings: The High preset's defaults as 0.3.0 has them, view distance 832, Classic clouds. Build: 0.4.0-dev.7 test build, and VSPT 0.3.0. NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). 0.3.0 was flown once, so the two read as alike, not as a gain or a loss. With a Ryzen 7 5800X3D.
NRD and Detail resolve
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | Note |
|---|---|---|---|
| NRD, in the cave | 1.11 | 0.94 | its own pass; the whole frame is 1.12 ms longer on the RTX 5080 and 0.89 on the RX 9070 XT |
| NRD, on the grassland | 1.15 | 0.94 | its own pass; the whole frame is 1.13 ms longer on the RTX 5080 and 0.94 on the RX 9070 XT |
| Detail resolve on top, in the cave | 0.35 | 0.28 | the whole frame; 0.33 of it in NRD's pass on the RTX 5080, 0.22 on the RX 9070 XT |
| Detail resolve on top, on the grassland | 0.33 | 0.27 | the whole frame on the RX 9070 XT, 0.21 of it in NRD's pass. On the RTX 5080, NRD's pass alone: its trace there varies from round to round by more than the setting adds |
Measured 2026-11-12. trace kit V7's NRD cases: each case against the case before it, the median of three rounds of 240 frames. The RTX 5080 ran them from the development tree, the RX 9070 XT from the packed kit. a torch-lit cave, and open grassland; 1280x720 trace (1440p with Performance upscaling), no upscaler after NRD. Settings: The game's defaults, traced inline as AMD cards trace; NRD REBLUR. Build: the code of the test build 0.4.0-dev.8. NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). NRD's own textures take 61 MB at this size, and 96 MB with Detail resolve on, on both cards. In the game an RTX card traces through the ray tracing pipeline; these cases trace inline, as AMD cards do, so they say what NRD adds on each card and not how fast one card is beside the other. The RX 9070 XT's cave figures are read from the kit's logs: its report leaves out a run whose slowest frames take over 1.5 times the median, as they did in every round of the cave on that card, with and without NRD (about twice). Those runs' medians agree to 0.03 ms across the three rounds.
VSPT 4.0 against 0.3.0, in the game
| Setting | NVIDIA GeForce RTX 5080, fps | Note |
|---|---|---|
| 0.3.0, standing | 59.2 | |
| Release candidate at 0.3.0's water settings, standing | 59.0 | |
| Release candidate at its defaults, standing | 58.4 | |
| 0.3.0, flying | 57.1 | two rounds |
| Release candidate at 0.3.0's water settings, flying | 57.4 | two rounds, one of them the slower first round |
| Release candidate at its defaults, flying | 56.6 | two rounds, one of them the slower first round |
| 0.3.0, settled | 65.0 | |
| Release candidate at 0.3.0's water settings, settled | 64.6 | |
| Release candidate at its defaults, settled | 64.2 |
Measured 2026-11-15. The game's frame log, flying over a generated village; traced frames a second, the median of up to three rounds flown in turn, the versions minutes apart in each. A phase in which shaders were still compiling is left out. A generated village, standing, flying over it, and settled after the flight; 3840x2160, DLSS Ray Reconstruction, Performance (a 1920x1080 trace). Settings: The High preset's defaults, view distance 832, Classic clouds. "At 0.3.0's water settings" is VSPT 4.0 with Clear water off and water stepped everywhere, as 0.3.0 has it; "at its defaults" has Clear water on and water smooth near you. Build: the code of the release candidate 0.4.0-rc.1, and VSPT 0.3.0. NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). At the same settings the two versions came within 0.6 of a frame a second of each other standing in every round, and within 1.5 flying and settled, neither ahead throughout: they read as alike. The first round was flown while something else used the graphics card and reads about 3 frames a second lower for both. The whole evening reads about 2 frames a second below the flights of 12 November for both versions, so compare the rows of this table with each other and not with that one. With a Ryzen 7 5800X3D.
NRD with FSR 3.1 beside Ray Reconstruction, in the game, on an RTX card
| Setting | NVIDIA GeForce RTX 5080, fps | Note |
|---|---|---|
| Ray Reconstruction, standing | 59.5 | |
| NRD with FSR 3.1, standing | 60.6 | |
| Ray Reconstruction, flying | 59.0 | from the next round: shaders were still compiling in this one |
| NRD with FSR 3.1, flying | 61.0 | |
| Ray Reconstruction, settled | 64.6 | |
| NRD with FSR 3.1, settled | 67.0 |
Measured 2026-11-15. The game's frame log, flying over a generated village; traced frames a second, one round, the two flown minutes apart. A generated village, standing, flying over it, and settled after the flight; 3840x2160, Performance upscaling (a 1920x1080 trace). Settings: The High preset's defaults with Clear water off and water stepped everywhere, view distance 832, Classic clouds; DLSS Ray Reconstruction, or NRD REBLUR at its defaults with FSR 3.1. Build: the code of the release candidate 0.4.0-rc.1. NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). NRD and FSR 3.1 together took about 2.4 ms of a frame here and Ray Reconstruction about 3.4, and the trace was about 0.6 ms longer with NRD. This is NRD on an NVIDIA card, whose rays go through the ray tracing pipeline. An AMD card traces inline and uses FSR 4.1 where it is offered, so this says nothing of a Radeon card's frame rate. With a Ryzen 7 5800X3D.
The ray tracing pipeline against inline, RX 9070 XT, trace kit V5
| Part | AMD Radeon RX 9070 XT (a tester's card), ms |
|---|---|
| Grassland, default settings | 13.69 / 8.65 → 11.37 / 6.46 |
| Grassland, light reuse off | 10.34 / 8.26 → 8.72 / 6.65 |
| Grassland, the game case | 22.38 / 16.73 → 16.74 / 11.13 |
| Grassland, the game case without light reuse | 15.82 / 13.59 → 13.73 / 11.21 |
| Grassland, the game case with Clear water | 29.02 / 23.01 → 16.98 / 11.27 |
| Cave, default settings | 9.46 / 6.05 → 4.92 / 1.82 |
| Cave, light reuse off | 7.41 / 6.04 → 3.45 / 2.18 |
| Cave, the game case | 14.62 / 11.12 → 5.93 / 2.62 |
| Cave, the game case without light reuse | 11.10 / 9.46 → 4.44 / 2.99 |
| Torch-lit room, default settings | 9.08 / 6.17 → 4.38 / 1.65 |
| Torch-lit room, light reuse off | 6.77 / 5.45 → 3.05 / 1.81 |
Measured 2026-11-09. Trace kit V5 (rt_bench, the renderer without the game), each case beside its inline twin. Three scenes: the same tester's dusk grassland as the other kit sets, a dark cave, and an enclosed room lit by torches; 1280x720 trace (what 1440p traces at Performance upscaling). Settings: Each cell is the ray tracing pipeline's whole trace / main pass, then inline's. Default settings: three bounces at three quarters of the pixels, light reuse through RTXDI, the radiance cache, 4 light candidates; the game case adds NRD's split signal, everyday fog traced the game's way and classic clouds. 240 frames, median of 3 rounds, every run kept. Build: The V5 kit as sent to testers (b0195d17e). Only the main path tracing pass was inline in it; light reuse's passes and the radiance cache's update stayed on the ray tracing pipeline. AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). Inline, the main pass was 3.3 to 4.2 times as fast in the cave and the torch-lit room with light reuse on (2.8 to 3.2 times with it off), 1.2 to 1.5 times on the grassland, and 2 times with Clear water, whose cost to the main pass went from 6.28 ms to 0.14. This is why Radeon cards on AMD's Windows driver trace inline. On an RTX 5080 the pipeline is the faster, so NVIDIA cards keep it.
Every pass inline against the ray tracing pipeline, trace kit V7
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 9060 XT (a tester's card), ms |
|---|---|---|---|
| Default settings, inline | 7.55 / 4.38 | 10.13 / 6.58 | 19.47 / 12.42 |
| Default settings, the other passes on the pipeline | 7.73 / 4.36 | 11.39 / 6.54 | 21.88 / 12.39 |
| Default settings, on the pipeline | 5.80 / 2.70 | 15.33 / 10.45 | 31.06 / 21.41 |
| The game case, inline | 15.42 / 11.79 | 15.06 / 11.27 | 29.20 / 21.53 |
| The game case, the other passes on the pipeline | 15.63 / 12.07 | 16.71 / 11.22 | 32.84 / 21.55 |
| The game case, on the pipeline | 8.44 / 5.22 | 24.70 / 19.22 | 50.58 / 39.02 |
| The game case with Clear water, inline | 15.99 / 12.24 | 15.21 / 11.41 | 29.58 / 21.90 |
| The game case without light reuse, inline | 14.38 / 12.11 | 13.27 / 11.44 | 25.84 / 21.87 |
Measured 2026-11-12. Trace kit V7 (rt_bench, the renderer without the game). The same tester's dusk grassland as the other kit sets. The kit's cave runs are not here: in every round their slowest frames took over 1.5 times the median, and the kit's report leaves such runs out; 1280x720 trace (what 1440p traces at Performance upscaling). Settings: Each cell is the whole trace / the main pass. Inline: every pass that traces as ray queries, as VSPT 4.0 does on AMD's Windows driver. The other passes on the pipeline: the main pass alone inline, light reuse's passes and the radiance cache's update on the ray tracing pipeline. On the pipeline: everything on the ray tracing pipeline (VS_RTX_INLINE=0). Cases as in V5. 240 frames, median of 3 rounds, every run kept. Build: The V7 kit, packed with the code of the test build 0.4.0-dev.8 (6b4c3e61b). NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). AMD Radeon RX 9060 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). With every pass inline the game case's whole trace took 15.06 ms against 24.70 on the pipeline, 1.6 times as fast; moving light reuse's and the cache's passes inline saved 1.65 ms of it. Inline, light reuse costs this card 1.79 ms of the game case and Clear water 0.15. What NRD itself costs the card, from the same kit, is in the set "NRD and Detail resolve". The RTX 5080 and a tester’s RX 9060 XT ran the same kit on 22 November. The RTX 5080 traces on the pipeline in the game, so its "on the pipeline" rows are its own; the set "Three cards as VSPT 4.0 traces on each" puts each card on its own path.
Three cards as VSPT 4.0 traces on each, trace kit V7
| Part | NVIDIA GeForce RTX 5080, ms | AMD Radeon RX 9070 XT (a tester's card), ms | AMD Radeon RX 9060 XT (a tester's card), ms | RX 9070 XT against RTX 5080 | RX 9060 XT against RTX 5080 |
|---|---|---|---|---|---|
| Default settings, in the cave | 3.72 | 4.63 | 8.30 | 1.2 times | 2.2 times |
| Default settings, on the grassland | 5.80 | 10.1 | 19.5 | 1.7 times | 3.4 times |
| The game case, in the cave | 4.25 | 5.31 | 10.0 | 1.2 times | 2.4 times |
| The game case, on the grassland | 8.44 | 15.1 | 29.2 | 1.8 times | 3.5 times |
Measured 2026-11-22. Trace kit V7 (rt_bench, the renderer without the game), the same packed kit on every card. The RX 9070 XT ran it on 12 November, the RTX 5080 and the RX 9060 XT on 22 November. The kit's two scenes: a torch-lit cave, and a tester's open grassland at dusk; 1280x720 trace (what 1440p traces at Performance upscaling). Settings: Each card on its own path, as in the game: the RTX 5080 on the ray tracing pipeline, the Radeons inline (AMD's Windows driver). The whole trace; 240 frames, median of 3 rounds, every run kept. Build: The V7 kit, packed with the code of the test build 0.4.0-dev.8 (6b4c3e61b). NVIDIA GeForce RTX 5080: driver 621.01 (Windows driver version string 32.0.16.2101). AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). AMD Radeon RX 9060 XT (a tester's card): driver AMD Software 26.10.1 (Windows driver version 32.0.32015.7029). Over the two scenes’ game cases the RX 9070 XT takes 1.5 times as long as the RTX 5080 and the RX 9060 XT 2.9 times. The RX 9060 XT takes 1.8 to 2.1 times as long as the RX 9070 XT in every case of the kit. Traced inline, as the Radeons trace, the RTX 5080 is no faster than the RX 9070 XT on the grassland (15.42 ms against 15.06): on its own pipeline it reorders its rays, which a ray query cannot. No RX 7900 XTX has run this kit.
Radeon cards in the game on Windows, tracing inline
| Setting | AMD Radeon RX 9070 XT (a tester's card), fps | AMD Radeon RX 9060 XT (a tester's card), fps | Note |
|---|---|---|---|
| Standing, nothing swaying in the wind, Performance | 88.2 | Frame time 11.16 ms median, 12.28 ms at the 99th percentile; GPU busy 10.71 ms a frame; 60 s | |
| The same place, Quality | 60.9 | Frame time 15.87 ms median, 17.56 ms at the 99th percentile; GPU busy 15.42 ms; 59 s | |
| Moving about, plants swaying in the wind, Performance | 47.6 | The release candidate. Frame time 20.07 ms median, 32.34 ms at the 99th percentile; GPU busy 21.92 ms; 59 s | |
| A world with about 70 other mods, Balanced | 40.6 | The release candidate. Frame time 24.70 ms median, 30.06 ms at the 99th percentile; GPU busy 21.21 ms; 25 s. Frame generation showed about twice as many frames |
Measured 2026-11-16. Testers' frame logs, from the report zips they sent: one stretch of steady play each, every traced frame in it kept. The testers' own worlds, view distance 320; RX 9070 XT: 2560x1440 through AMD FSR 4.1, Performance (1280x720 traced) or Quality (1706x960). RX 9060 XT: 1920x1027 through FSR 4.1 Balanced (1129x604 traced), with FSR 3 frame generation. Settings: NRD REBLUR; light reuse on (direct light), the radiance cache, fog and Clear water on, three bounces at three quarters of the pixels: VSPT's defaults, with each tester's own changes. On the RX 9070 XT lamps are found to 320 blocks (64 by default) and terrain detail by distance is on; on the RX 9060 XT the denoiser's history is 8 frames, light reuse's temporal pass is off and frame generation is on. Frames a second are traced frames. Build: Test builds before VSPT 4.0: 0.4.0-dev.8 for the RX 9070 XT's first two rows, the release candidates 0.4.0-rc.1 and 0.4.0-rc.4 for the others. All traced inline, as VSPT 4.0 does on these cards by default. AMD Radeon RX 9070 XT (a tester's card): driver AMD Software 26.10.1 (the test build's rows), 26.11.1 (the release candidate's). AMD Radeon RX 9060 XT (a tester's card): driver AMD Software 26.10.1. Each row is one stretch of a tester's frame log. No frame in them was traced while a pipeline was being made, and no frame cap held them back: the game's own frames were capped at 160 and 200 a second, the traced view not, and the GPU was busy for most of each frame. The rows say what each card ran at there, and are not comparable with each other or with the RTX 5080's sets: worlds, places, resolutions, settings and builds all differ. Taken from 12 to 16 November 2026.
Inline against the ray tracing pipeline on Linux, RX 7900 XTX
| Setting | AMD Radeon RX 7900 XTX (a tester's card), fps | Note |
|---|---|---|
| Traced inline (the release candidate's default) | 11.8 | Frame time 84.57 ms median, 87.99 ms at the 99th percentile; the main trace pass 74.22 ms; 60 s |
| On the ray tracing pipeline (VSPT 4.0's default on RADV) | 52.8 | Frame time 18.87 ms median, 19.73 ms at the 99th percentile; the main trace pass 9.72 ms, NRD and FSR 5.02 ms; 30 s |
Measured 2026-11-15. A tester's two frame logs from the same place with the same settings, one traced inline and one on the ray tracing pipeline (VS_RTX_INLINE=0). A tester's own world, view distance 256; 3440x1368, traced at full size: NRD REBLUR, then AMD FSR 3.1 at native resolution. Settings: The tester's own, below the defaults: one bounce at half the pixels; the radiance cache, fog and water caustics off; Low clouds; light reuse on (direct light); NRD's Detail resolve on. Build: The release candidate 0.4.0-rc.1, which traced inline on every AMD card. VSPT 4.0 keeps the ray tracing pipeline on Mesa's RADV driver. AMD Radeon RX 7900 XTX (a tester's card): driver Mesa 26.2.6 (RADV), on Fedora. On this driver the main trace pass took 7.6 times as long inline, where on AMD's own Windows driver inline is the faster (trace kit V5): so VSPT 4.0 traces on the ray tracing pipeline on RADV and inline on AMD's Windows driver. The game's own frames were capped at 175 a second, the display's rate, which the traced view never reached. At 3440x1368 traced the card traced about five times the pixels of 2560x1440 at Performance.
