Any known recent performance regressions?

themanuel · 02-16-2015, 02:07 AM #1

I have been using 4.0-4384 and was checking out the latest builds but I'm noticing a ~15% degradation. Has anything significant changed since then?

Below are my settings and my pc specs are in my signature.

Dolphin.ini

Code:
[General]

LastFilename = D:\Disc Images\Gamecube/F-Zero GX (USA) [GFZE01].iso

ShowLag = False

GCMPathes = 2

GCMPath0 = D:\Disc Images\Gamecube

RecursiveGCMPaths = False

NANDRootPath = C:/Users/Arcadius/Documents/Dolphin Emulator/Wii

WirelessMac = 

GCMPath1 = D:\Disc Images\Wii

ShowFrameCount = False

ISOPaths = 3

ISOPath0 = D:\Disc Images\Wii

RecursiveISOPaths = False

ISOPath1 = D:\Disc Images\Gamecube

ISOPath2 = D:\Disc Images\Wii\WiiWare

[Interface]

ConfirmStop = False

UsePanicHandlers = True

OnScreenDisplayMessages = False

HideCursor = True

AutoHideCursor = False

MainWindowPosX = 22

MainWindowPosY = 39

MainWindowWidth = 912

MainWindowHeight = 913

Language = 0

ShowToolbar = True

ShowStatusbar = True

ShowLogWindow = False

ShowLogConfigWindow = False

ShowConsole = False

ThemeName40 = Clean

ExtendedFPSInfo = False

[Hotkeys]

Open = 79

OpenModifier = 2

ChangeDisc = 0

ChangeDiscModifier = 0

RefreshList = 0

RefreshListModifier = 0

PlayPause = 349

PlayPauseModifier = 0

Stop = 27

StopModifier = 0

Reset = 0

ResetModifier = 0

FrameAdvance = 0

FrameAdvanceModifier = 0

StartRecording = 0

StartRecordingModifier = 0

PlayRecording = 0

PlayRecordingModifier = 0

ExportRecording = 0

ExportRecordingModifier = 0

Readonlymode = 0

ReadonlymodeModifier = 0

ToggleFullscreen = 13

ToggleFullscreenModifier = 1

Screenshot = 348

ScreenshotModifier = 0

Exit = 0

ExitModifier = 0

Wiimote1Connect = 344

Wiimote1ConnectModifier = 1

Wiimote2Connect = 345

Wiimote2ConnectModifier = 1

Wiimote3Connect = 346

Wiimote3ConnectModifier = 1

Wiimote4Connect = 347

Wiimote4ConnectModifier = 1

BalanceBoardConnect = 348

BalanceBoardConnectModifier = 1

ToggleIR = 0

ToggleIRModifier = 0

ToggleAspectRatio = 0

ToggleAspectRatioModifier = 0

ToggleEFBCopies = 0

ToggleEFBCopiesModifier = 0

ToggleFog = 0

ToggleFogModifier = 0

IncreaseFrameLimit = 0

IncreaseFrameLimitModifier = 0

DecreaseFrameLimit = 0

DecreaseFrameLimitModifier = 0

LoadStateSlot1 = 340

LoadStateSlot1Modifier = 0

LoadStateSlot2 = 341

LoadStateSlot2Modifier = 0

LoadStateSlot3 = 342

LoadStateSlot3Modifier = 0

LoadStateSlot4 = 343

LoadStateSlot4Modifier = 0

LoadStateSlot5 = 344

LoadStateSlot5Modifier = 0

LoadStateSlot6 = 345

LoadStateSlot6Modifier = 0

LoadStateSlot7 = 346

LoadStateSlot7Modifier = 0

LoadStateSlot8 = 347

LoadStateSlot8Modifier = 0

LoadStateSlot9 = 0

LoadStateSlot9Modifier = 0

LoadStateSlot10 = 0

LoadStateSlot10Modifier = 0

SaveStateSlot1 = 340

SaveStateSlot1Modifier = 4

SaveStateSlot2 = 341

SaveStateSlot2Modifier = 4

SaveStateSlot3 = 342

SaveStateSlot3Modifier = 4

SaveStateSlot4 = 343

SaveStateSlot4Modifier = 4

SaveStateSlot5 = 344

SaveStateSlot5Modifier = 4

SaveStateSlot6 = 345

SaveStateSlot6Modifier = 4

SaveStateSlot7 = 346

SaveStateSlot7Modifier = 4

SaveStateSlot8 = 347

SaveStateSlot8Modifier = 4

SaveStateSlot9 = 0

SaveStateSlot9Modifier = 0

SaveStateSlot10 = 0

SaveStateSlot10Modifier = 0

LoadLastState1 = 0

LoadLastState1Modifier = 0

LoadLastState2 = 0

LoadLastState2Modifier = 0

LoadLastState3 = 0

LoadLastState3Modifier = 0

LoadLastState4 = 0

LoadLastState4Modifier = 0

LoadLastState5 = 0

LoadLastState5Modifier = 0

LoadLastState6 = 0

LoadLastState6Modifier = 0

LoadLastState7 = 0

LoadLastState7Modifier = 0

LoadLastState8 = 0

LoadLastState8Modifier = 0

SaveFirstState = 0

SaveFirstStateModifier = 0

UndoLoadState = 351

UndoLoadStateModifier = 0

UndoSaveState = 351

UndoSaveStateModifier = 4

SaveStateFile = 0

SaveStateFileModifier = 0

LoadStateFile = 0

LoadStateFileModifier = 0

ToggleThrottle = 9

ToggleThrottleModifier = 0

SelectStateSlot1 = 0

SelectStateSlot1Modifier = 0

SelectStateSlot2 = 0

SelectStateSlot2Modifier = 0

SelectStateSlot3 = 0

SelectStateSlot3Modifier = 0

SelectStateSlot4 = 0

SelectStateSlot4Modifier = 0

SelectStateSlot5 = 0

SelectStateSlot5Modifier = 0

SelectStateSlot6 = 0

SelectStateSlot6Modifier = 0

SelectStateSlot7 = 0

SelectStateSlot7Modifier = 0

SelectStateSlot8 = 0

SelectStateSlot8Modifier = 0

SelectStateSlot9 = 0

SelectStateSlot9Modifier = 0

SelectStateSlot10 = 0

SelectStateSlot10Modifier = 0

SaveSelectedSlot = 0

SaveSelectedSlotModifier = 0

LoadSelectedSlot = 0

LoadSelectedSlotModifier = 0

VolumeUp = 0

VolumeUpModifier = 0

VolumeDown = 0

VolumeDownModifier = 0

VolumeToggleMute = 0

VolumeToggleMuteModifier = 0

FreelookDecreaseSpeed = 49

FreelookDecreaseSpeedModifier = 4

FreelookIncreaseSpeed = 50

FreelookIncreaseSpeedModifier = 4

FreelookResetSpeed = 70

FreelookResetSpeedModifier = 4

FreelookUp = 69

FreelookUpModifier = 4

FreelookDown = 81

FreelookDownModifier = 4

FreelookLeft = 65

FreelookLeftModifier = 4

FreelookRight = 68

FreelookRightModifier = 4

FreelookZoomIn = 87

FreelookZoomInModifier = 4

FreelookZoomOut = 83

FreelookZoomOutModifier = 4

FreelookReset = 82

FreelookResetModifier = 4

IncreaseDepth = 0

IncreaseDepthModifier = 0

DecreaseDepth = 0

DecreaseDepthModifier = 0

IncreaseConvergence = 0

IncreaseConvergenceModifier = 0

DecreaseConvergence = 0

DecreaseConvergenceModifier = 0

[Display]

FullscreenResolution = 1920x1080

Fullscreen = True

RenderToMain = False

RenderWindowXPos = -4

RenderWindowYPos = 125

RenderWindowWidth = 1297

RenderWindowHeight = 699

RenderWindowAutoSize = False

KeepWindowOnTop = False

ProgressiveScan = False

DisableScreenSaver = True

ForceNTSCJ = False

[GameList]

ListDrives = False

ListWad = True

ListWii = True

ListGC = True

ListJap = True

ListPal = True

ListUsa = True

ListFrance = True

ListItaly = True

ListKorea = True

ListTaiwan = True

ListUnknown = True

ListSort = 1

ListSortSecondary = 3

ColorCompressed = True

ColumnPlatform = True

ColumnBanner = True

ColumnNotes = True

ColumnID = False

ColumnRegion = True

ColumnSize = True

ColumnState = True

ListAustralia = True

ListGermany = True

ListInternational = True

ListNetherlands = True

ListRussia = True

ListSpain = True

[Core]

HLE_BS2 = True

CPUCore = 1

Fastmem = True

CPUThread = True

DSPThread = False

DSPHLE = True

SkipIdle = True

DefaultGCM = 

DVDRoot = 

Apploader = 

EnableCheats = True

SelectedLanguage = 0

DPL2Decoder = False

Latency = 2

MemcardAPath = C:\Users\Arcadius\Documents\Dolphin Emulator\GC\MemoryCardA.USA.raw

MemcardBPath = C:/Users/Arcadius/Documents/Dolphin Emulator/GC/MemoryCardB.USA.raw

SlotA = 8

SlotB = 255

SerialPort1 = 255

BBA_MAC = 

SIDevice0 = 6

SIDevice1 = 6

SIDevice2 = 6

SIDevice3 = 6

WiiSDCard = False

WiiKeyboard = False

WiimoteContinuousScanning = False

WiimoteEnableSpeaker = False

RunCompareServer = False

RunCompareClient = False

FrameLimit = 0x00000001

UseFPS = False

GFXBackend = D3D

FrameSkip = 0x00000000

GPUDeterminismMode = auto

DefaultISO = 

SyncOnSkipIdle = True

Overclock = 1.000000

OverclockEnable = False

GameCubeAdapter = True

GameCubeAdapterThread = True

AgpCartAPath = 

AgpCartBPath = 

[Movie]

PauseMovie = False

Author = 

DumpFrames = False

ShowInputDisplay = False

DumpFramesSilent = False

[DSP]

EnableJIT = False

DumpAudio = False

Backend = XAudio2

Volume = 100

CaptureLog = False

[FifoPlayer]

LoopReplay = True

[Input]

BackgroundInput = False

gfx_dx11.ini

Code:
[Hardware]

VSync = True

Adapter = 0

[Settings]

AspectRatio = 1

Crop = False

wideScreenHack = False

UseXFB = False

UseRealXFB = False

SafeTextureCacheColorSamples = 128

ShowFPS = True

LogFPSToFile = False

ShowInputDisplay = False

OverlayStats = False

OverlayProjStats = False

DLOptimize = 0

Show = 0

DumpTextures = False

HiresTextures = True

DumpEFBTarget = False

DumpFrames = False

FreeLook = False

UseFFV1 = False

AnaglyphStereo = False

AnaglyphStereoSeparation = 200

AnaglyphFocalAngle = 0

EnablePixelLighting = False

HackedBufferUpload = False

FastDepthCalc = True

ShowEFBCopyRegions = False

MSAA = 6

EFBScale = 0

TexFmtOverlayEnable = False

TexFmtOverlayCenter = False

Wireframe = False

DstAlphaPass = False

DisableFog = False

EnableOpenCL = False

OMPDecoder = False

EnableShaderDebugging = False

LogRenderTimeToFile = False

BorderlessFullscreen = True

ConvertHiresTextures = True

[Enhancements]

ForceFiltering = False

MaxAnisotropy = 4

PostProcessingShader = 

Enable3dVision = False

StereoMode = 0

StereoDepth = 20

StereoConvergence = 70

StereoSwapEyes = False

[Hacks]

EFBAccessEnable = True

DlistCachingEnable = False

EFBCopyEnable = True

EFBToTextureEnable = True

EFBScaledCopy = True

EFBCopyCacheEnable = False

EFBEmulateFormatChanges = False

ForceDualSourceBlend = False

kirbypuff · (This post was last modified: 02-16-2015, 03:48 AM by kirbypuff.)

(02-16-2015, 02:07 AM)themanuel Wrote: I have been using 4.0-4384 and was checking out the latest builds but I'm noticing a ~15% degradation. Has anything significant changed since then?

* The 'VBeam Speed Boost' option was removed : ~5% speed drop if you used this feature before.
* Various perf. regressions which affect AMD GPUs only. One of them changed some of the graphics processing to be in sync instead of async (to improve perf. on NVIDIA GPUs). But AMD GPU drivers just love Async. What's interesting is the recent 'Async Events' PR (not merged yet) gives a nice perf. boost to AMD GPUs.

If you want performance, use 4.0-4575 or 4.0-5429
If you want accuracy, use the latest dev. build.

[EDIT] Fixed. (thanks for the explanation, JMC47).

**JMC47** · 02-16-2015, 03:06 AM #3

@themanuel

There's nothing that should have caused a 15% performance degradation. If you can narrow down to builds that cause large drops in performance, we can always have ti looked at.

@kirbypuff

You are completely wrong on most of your points. Let me clean up your argument here.

"vbeam option" was not removed; accurate vbeam emulation is always on. vbeam speedhack was removed because it was basically useless for normal emulation. Anyone who used it dealt with severe audio options in 99% of titles.

"Block merging" is always on, not always off now. We removed the option to turn it off because there's no conceivable reason for a user to turn it off. How the hell is that a slowdown?

MMU is not always on. We can't do that because it'd be 20% slower. The MMU Speedhack is always on, and we did significant testing to make sure it didn't cause slowdowns in games that didn't use the MMU Speedhack. I'm pretty sure it was added way before 4575.

3 Perf regressions that affect AMD GPUs, and you've reported none of them to the issue tracker, and none of our other AMD users have confirmed them. How convenient. I won't believe those reports until we start getting build numbers + issue reports that are confirmed by our other AMD users. I have an R9 285 mind you, so, if there's a regression, I can help out. I'm not arguing that some perf regressions snuck in (we don't have every game, and we don't have every GPU/CPU combination.)

themanuel · 02-16-2015, 03:26 AM #4

Thanks for clarifying that, JMC47.

I can confirm that on the old version I'm using (4384), disabling vbeam, block merging and MMU on the game properties doesn't make a difference in performance, so that's not it, as JMC47 explained.

Secondly, my report on 15% degradation is inaccurate. I realized can't seem to get stable performance from test to test, even with the same build, as sometimes unexplained slowdowns creep in. However, over several runs, after getting consistent numbers, it does look like the recent version I tried (5487) is a few frames slower than 4384 but not 15%, more like 4-5%, so my apologies for the misinformation. The game I always use as benchmark is F-Zero GX, Sand Ocean: Lateral Shift stage with the mirage effect in the background.

I you have ideas on how to best benchmark the game to provide more meaningful reports, I can certainly give a try.

Thanks!

kirbypuff · (This post was last modified: 02-16-2015, 04:29 AM by kirbypuff.)

(02-16-2015, 03:26 AM)themanuel Wrote: ...the old version I'm using, 4.0-4384...

There's no such build.

I also see a lot of deprecated entries (from very old builds) in those .ini files. They need some cleanup.

themanuel · 02-16-2015, 04:47 AM #6

I'm not helping you help me. Build is 5384 (4384 was a typo).
I'll redo my tests in portable mode as soon as I can.

Thanks.

**JMC47** · 02-16-2015, 04:48 AM #7

That's a poor benchmark for general performance; the lateral shift thing is a CPU - GPU communication. But, if that's the game you play the most and you're seeing a regression in that situation, it narrows it down a lot.

IceStrike256 · 02-16-2015, 01:27 PM #8

I have noticed some hiccups on my 290x playing RS2 used to keep smooth with EFB to RAM on now I have to use texture to keep my constant performance using latest master.

themanuel · 02-16-2015, 11:20 PM #9

(02-16-2015, 04:48 AM)JMC47 Wrote: That's a poor benchmark for general performance; the lateral shift thing is a CPU - GPU communication. But, if that's the game you play the most and you're seeing a regression in that situation, it narrows it down a lot.

For now, I've been playing it mostly when I'm benchmarking Smile

I should start doing benchmarks on games I actually play with any frequency.

spyro · 02-18-2015, 07:54 PM **#10**

For me on my desktop
i5 2500k
460gtx
windows 8
almost every revision has the same fps +-1/2%
new super mario (5738) 297 latest 295

on my laptop i7 720qm
amd 5830 mobility 1gb ddr5
windows 8.1
after 5378 until the most recent version i see a lost in fps
new super mario intro menu 190 to 165/167 (latest revision)

smg2 intro 64fps to 62fps
rlII 137/143 to 127/131