Hello,
This is a pattern that I’ve recently discovered in our project as (much like anyone else), we utilize the hardware benchmark functionality for detecting recommended settings.
However it appears that due to what we assume is a freak interaction between the engine, driver and newest 5000 series cards from nvidia, the benchmark will never finish.
Depending on circumstances (primarily driver version) it will either infinitely sit on the benchmark and eventually log the 5 second timeout on D3D12 instruction and not do anything with it further, rendering the application inoperable OR instantly cause BSOD crash.
Log Cmd SynthBenchmark
Display LogSynthBenchmark FSynthBenchmark (V0.95): requested WorkScale=10.00
Display LogSynthBenchmark ===============
Display LogSynthBenchmark Main Processor:
Display LogSynthBenchmark ... 0.008761 s/Run 'RayIntersect'
Display LogSynthBenchmark ... 0.008571 s/Run 'Fractal'
Display LogSynthBenchmark CompiledTarget_x_Bits: 64
Display LogSynthBenchmark UE_BUILD_SHIPPING: 0
Display LogSynthBenchmark UE_BUILD_TEST: 0
Display LogSynthBenchmark UE_BUILD_DEBUG: 0
Display LogSynthBenchmark TotalPhysicalGBRam: 128
Display LogSynthBenchmark NumberOfCores (physical): 16
Display LogSynthBenchmark NumberOfCores (logical): 32
Display LogSynthBenchmark CPU Perf Index 0: 292.3 (weight 1.00)
Display LogSynthBenchmark CPU Perf Index 1: 333.7 (weight 1.50)
Display LogSynthBenchmark
Display LogSynthBenchmark Graphics:
Display LogSynthBenchmark Adapter Name: 'NVIDIA GeForce RTX 5090'
Display LogSynthBenchmark (On Optimus the name might be wrong, memory should be ok)
Display LogSynthBenchmark Vendor Id: 0x10DE
Display LogSynthBenchmark Device Id: 0x2B85
Display LogSynthBenchmark Device Revision: 0xA1
Display LogSynthBenchmark GPU Memory: 31867/0/65333 MB
Display LogSynthBenchmark GPU warmup (round 1) test: 0.05s
Display LogSynthBenchmark ... 49.899 s/GigaPix, Confidence=100% 'ALUHeavyNoise' (likely to be inaccurate)
Display LogSynthBenchmark ... 52.350 s/GigaPix, Confidence=100% 'TexHeavy' (likely to be inaccurate)
Display LogSynthBenchmark ... 50.559 s/GigaPix, Confidence=100% 'DepTexHeavy' (likely to be inaccurate)
Display LogSynthBenchmark ... 58.493 s/GigaPix, Confidence=100% 'FillOnly' (likely to be inaccurate)
Display LogSynthBenchmark ... 6.492 s/GigaPix, Confidence=100% 'Bandwidth' (likely to be inaccurate)
Display LogSynthBenchmark ... 25.307 s/GigaVert, Confidence=100% 'VertThroughPut1' (likely to be inaccurate)
Display LogSynthBenchmark ... 50.190 s/GigaVert, Confidence=100% 'VertThroughPut2' (likely to be inaccurate)
I’ve tried it on both 5.5.4 and 5.6 releases and it’s a consistent occurrence, is anyone else able to confirm this?
To clarify, 100% reproduction method so far seems to be calling the benchmark functionality from editor.