Frametimes - the quest for perfection (discussion)

[GanjaSMK]

I've been trying to dial fametimes in for high-FPS play without the intermittent spikes which in turn create 'stutter' effects. 
 

There's a lot to it - hardware, software and the tweaking of all of it, not to mention drivers and their own effects.

 

I am using AMD's driver-reporting frametime for generic visual comparison versus plotting graphs through other software.  I'm interesting diving deeper; and I know RTSS/Afterburner can help here.  I'm using COD:Warzone as test bed since it's what I play mostly (due to friends playing and my gross interest in the game, that's for another discussion; the game has it's own quirks).  It's also highly popular so if anyone reading can benefit, great.

 

Quick reference snip:

 

 

The quest for perfection would be a 'flat' line on the left - the right side shows 1% lows - but depending on the spikes and how high they go - can lead to that 'stutter' or 'hitching' effect.  I'm (for whatever reason) highly sensitive visually and audibly to these spikes, as would be most people who have played first-person-shooters at high enough levels to perceive the effects of low refresh rates, low FPS, etc.  I'm not at all any professional nor do I have a bunch of tournament championships behind me, but I have competed and I have played with pro's casually - always seeking to improve my own game. 

 

I'd love anyone's input into this topic if you're inclined.  Directly affecting frametimes are things like:

 

RAM bandwidth and latency

CPU

Graphics card

 

But, depending on software, Windows services/core residencies, the game itself, and server-side implementation (out of our hands) - your frame times can vary wildly. 

 

System specs:

 

10850K "Stock"

MSI Gaming Edge Wifi

MSI Trio 6800XT "Stock"

32GB 3200/C14 RAM (14-14-14-34-560-2T)

NVME OS Drive & SATA SSD Game Drive

 

I have found my best settings with this particular game to be:

 

AMD driver optimizations completely off (no Anti-lag, no Boost, no Chill, no image sharpening)

adv_options file set to: VideoMemoryScale = 0.5 RendererWorkerCount = 6

All settings in-game are highest possible without ray-tracing enabled (because why not push the hardware?)

 

Anything anyone wants to add that can explicitly show any reduction in those spikes, much appreciated.  I'm on a quest to find the best settings with the hardware above as well as any software/OS related tweaks. Perfectly flat graphs won't ever exist - kind of a pipe dream - but reducing those spikes to lower levels or mitigating them outright is my priority for this topic.

 

 

 

 

Edited April 8, 2021 by GanjaSMK

[UltraMega]

What vsync setting do you use? 

 

It's pretty hard to get rid of all spikes, your graph looks pretty good tbh.

Edited April 8, 2021 by UltraMega

[J7SC_Orion]

One thing I can add to the list as a 'sub-theme' is certain memory timings on the main system RAM, specifically tFAW and perhaps tRAS. If tFAW is too tight - even with passing all memory stress tests, you can experience some serious stutter.

 

Another development to keep an eye on re. frame times is the release of resizable-BAR. I'm tied into knots right now re. time, but hope to test that out later next week on my 3090.  

[GanjaSMK]

36 minutes ago, UltraMega said:

What vsync setting do you use? 

 

It's pretty hard to get rid of all spikes, your graph looks pretty good tbh.

 

With my intention to keep input lag as low as possible, none. I have mixed results using VRR/Freesync options (in particular with this game).  I have spent a bit of time reading up and trying Blur Busters 'low input lag' VSYNC, but without great results (I think simply due to my personal high sensitivity).  I've tried as well running Freesync + VSYNC, but also have not had great results there either.  Frame tearing is hugely visible for me but the input lag reduction trumps the frame tearing in terms of gaming.  As for frame pacing, I see that frame times are generally a lot more 'spiked' and it feels 'spiked' when played with VSYNC/VRR Freesync. 

I think I need to test again though.  Seems like Freesync can be buggy to implement correctly and/or my system just doesn't keep up pace for a 'perfectly smooth' implementation.

 

How's your experience been?

 

 

32 minutes ago, J7SC_Orion said:

One thing I can add to the list as a 'sub-theme' is certain memory timings on the main system RAM, specifically tFAW and perhaps tRAS. If tFAW is too tight - even with passing all memory stress tests, you can experience some serious stutter.

 

Another development to keep an eye on re. frame times is the release of resizable-BAR. I'm tied into knots right now re. time, but hope to test that out later next week on my 3090.  

 

OH man when your time becomes more available please feel free to put out anything you can!  Re. RAM timing and resizable-BAR I have not tried digging further.  RAM timing/tuning has always been beyond my scope, though truth be told simply because I haven't spent enough time reading/tinkering/testing it. 

 

I'm interested in buying higher binned, higher bandwidth RAM; but it's expensive atm and I haven't truly tested my current kit for it's maximum potential.  I've run 3800-15-15-15-36-560 with no subtiming optimization without any particular errors showing up. However I (admittedly) did not test long enough to omit errors production - I just slapped in the timings, reset, and off to the games/usage after quickly testing for immediate errors.

 

[J7SC_Orion]

55 minutes ago, GanjaSMK said:

 

With my intention to keep input lag as low as possible, none. I have mixed results using VRR/Freesync options (in particular with this game).  I have spent a bit of time reading up and trying Blur Busters 'low input lag' VSYNC, but without great results (I think simply due to my personal high sensitivity).  I've tried as well running Freesync + VSYNC, but also have not had great results there either.  Frame tearing is hugely visible for me but the input lag reduction trumps the frame tearing in terms of gaming.  As for frame pacing, I see that frame times are generally a lot more 'spiked' and it feels 'spiked' when played with VSYNC/VRR Freesync. 

I think I need to test again though.  Seems like Freesync can be buggy to implement correctly and/or my system just doesn't keep up pace for a 'perfectly smooth' implementation.

 

How's your experience been?

 

 

 

OH man when your time becomes more available please feel free to put out anything you can!  Re. RAM timing and resizable-BAR I have not tried digging further.  RAM timing/tuning has always been beyond my scope, though truth be told simply because I haven't spent enough time reading/tinkering/testing it. 

 

I'm interested in buying higher binned, higher bandwidth RAM; but it's expensive atm and I haven't truly tested my current kit for it's maximum potential.  I've run 3800-15-15-15-36-560 with no subtiming optimization without any particular errors showing up. However I (admittedly) did not test long enough to omit errors production - I just slapped in the timings, reset, and off to the games/usage after quickly testing for immediate errors.

 

 

...on the resizable_BAR, apart from my own timetable, I'm also waiting for a a final mobo bios update (currently only in beta). Re. your primary memory settings, if you feel like it (but no guarantees...), you might want to try 15-15-15-32-374 . 

 

I'm pretty certain about the 32 for tRAS with the other primary settings you showed as tRAS is best set as the sum of the first two (CL and tRCD) PLUS add '2'. The tRFC at 374 might work, but you can always play around with it...and before tRFC changes, have you tried command rate at 1T instead of 2T ? That usually provides a nice boost, even with GPU

[UltraMega]

You might wanna try v-sync with triple buffering enabled just to see if it feels smoother.

[GanjaSMK]

I need to test more, but subjectively I decided to push the RAM.  Clocked at 4000mhz @ 15-15-15-36-560 I get a really nice graph:

 

 

I tried @J7SC_Orion 15-15-15-32-374 and it wouldn't boot.  So I tried 15-15-15-34-374; no boot.  15-15-15-34-560 @ 3800.  At <- the graph was:

 

 

Pretty clear RAM is best place to start IMHO.  Is it fully stable?  I'll have to truly test but now I want to see how far I can push.  I don't even think aiming for CPU overlcock is worth the effort considering initial results.  4.8 and really fast RAM would seem to 'do the trick'. 

 

@UltraMega

 

I may have to try Freesync/VSYNC now if I can get RAM fast enough to maintain 1% lows above 48 'dips' (monitor does 48-165 VRR).  I don't know if triple buffering will work as supposedly Radeon drivers note it is applicable to OpenGL only.  It's like half of classic settings in graphics options for drivers are all DX9,10 and no real benefit outside of older games.  Dunno though, Nvidia side seems to have pertinent settings that can/do affect performance in specific titles. 

 

RAM wise I dunno how far I'll get in bandwidth; they're DR b-die sticks at 16GB each. 

[J7SC_Orion]

6 hours ago, GanjaSMK said:

I need to test more, but subjectively I decided to push the RAM.  Clocked at 4000mhz @ 15-15-15-36-560 I get a really nice graph:

 

 

I tried @J7SC_Orion 15-15-15-32-374 and it wouldn't boot.  So I tried 15-15-15-34-374; no boot.  15-15-15-34-560 @ 3800.  At <- the graph was:

 

 

Pretty clear RAM is best place to start IMHO.  Is it fully stable?  I'll have to truly test but now I want to see how far I can push.  I don't even think aiming for CPU overlcock is worth the effort considering initial results.  4.8 and really fast RAM would seem to 'do the trick'. 

 

@UltraMega

 

I may have to try Freesync/VSYNC now if I can get RAM fast enough to maintain 1% lows above 48 'dips' (monitor does 48-165 VRR).  I don't know if triple buffering will work as supposedly Radeon drivers note it is applicable to OpenGL only.  It's like half of classic settings in graphics options for drivers are all DX9,10 and no real benefit outside of older games.  Dunno though, Nvidia side seems to have pertinent settings that can/do affect performance in specific titles. 

 

RAM wise I dunno how far I'll get in bandwidth; they're DR b-die sticks at 16GB each. 

 

...what is your RAM vendor and model number, ie. GSkill F4-3866C18-8GTZR ? As you stated, I think RAM is indeed a good place to start in your quest for smooth and low frametimes, but at the end of the day, it will be a combination of that and other parameters, such as GPU settings. Your Radeon 6800XT GPU having a nice 16GB is a good thing, also.

[Avacado]

There is some really good RAM nerd talk going on in this thread and I dig it. 

[J7SC_Orion]

15 minutes ago, Avacado said:

There is some really good RAM nerd talk going on in this thread and I dig it. 

 

RAM optimization is a rabbit hole that is fun falling into, but frustrating to crawl out of...

 

 

[Avacado]

I choose the blue pill. 

 

[mouacyk]

RAM optimization is the ultimate rabbit hole you will have to go in.  However, have you trimmed the fat first so that you can fit.  Absolute minimal background apps and services for the OS?  Check your system with a resplendence Latency Monitor?

[GanjaSMK]

Almost forgot. I'm low on RGB. Might ding performance a little bit.

 

 

Edited April 9, 2021 by GanjaSMK

[GanjaSMK]

3 hours ago, mouacyk said:

RAM optimization is the ultimate rabbit hole you will have to go in.  However, have you trimmed the fat first so that you can fit.  Absolute minimal background apps and services for the OS?  Check your system with a resplendence Latency Monitor?

 

I have not; I mean I'm OCD like the next guy but I have several other things running while gaming sometimes. For years I'd do all that nipping of services and stuff and I can't say it ever really did much.  These parts may not be the Queen's jewels, but they're the highest end parts I've ever owned.  They should be capable of dealing with all the little digital bits IMHO.

 

4 hours ago, J7SC_Orion said:

 

...what is your RAM vendor and model number, ie. GSkill F4-3866C18-8GTZR ? As you stated, I think RAM is indeed a good place to start in your quest for smooth and low frametimes, but at the end of the day, it will be a combination of that and other parameters, such as GPU settings. Your Radeon 6800XT GPU having a nice 16GB is a good thing, also.

 

F4-3200C14-16GVK - 2x16GB.  With all my nerd research, these are dual rank samsung b-die sticks that should be *decent* enough to rack up some hertz n timings.  Any insight?

 

3 hours ago, Avacado said:

There is some really good RAM nerd talk going on in this thread and I dig it. 

 

[UltraMega]

23 minutes ago, GanjaSMK said:

Almost forgot. I'm low on RGB. Might ding performance a little bit.

 

 

I posted this a while back but it's perfect again here now:

[J7SC_Orion]

1 hour ago, GanjaSMK said:

 

I have not; I mean I'm OCD like the next guy but I have several other things running while gaming sometimes. For years I'd do all that nipping of services and stuff and I can't say it ever really did much.  These parts may not be the Queen's jewels, but they're the highest end parts I've ever owned.  They should be capable of dealing with all the little digital bits IMHO.

 

 

F4-3200C14-16GVK - 2x16GB.  With all my nerd research, these are dual rank samsung b-die sticks that should be *decent* enough to rack up some hertz n timings.  Any insight?

 

 

  

...yeah, according to this handy Samsung 'B-die' finder > here, you're right ...dual rank has advantages and disadvantages, but overall great choice RAM.

 

The first thing I would try is switch to CmdRate 1T from 2T, even at a lower speed, say 3400MHz or 3600MHz, and less 'max tight' timings. You can then iterate your speed up and tighten timings afterwards. One thing to keep in mind is all the testing you have to do in-between changing RAM timings - some of them are interrelated, but generally, it is best to change one thing at a time - that way, you don't end up with 'one equation with 27 unknowns'.

 

...another tip: check > this out  ...I realize it is for Ryzen, but it will still help as at least rough guidance for Intel. Apart from my Intel machines, I run a X399/2950X and X570/3950X...both have identical 'Intel marketed' 4x8GB B-die kits, yet the linked software was great for double-checking once I finished my own optimizations...the only thing I would disagree on is tRAS (per above post) but that's a minor thing. I even use that Ryzen DRAM calculator to check my settings for Intel Skylake. It will provide a great template to get you going into the rabbit hole, even if some Intel-specific values might differ slightly... 

 

...final tip: Once you have made progress with your RAM optimization but feel there's more to be had, save 'intermediate' profiles in the Bios - you can always return to those and build on them for the next iteration...

 

 

[GanjaSMK]

I got a lotta reading to do on RAM timings.  I've read maybe two or three equations for calculating timings - but I known full well beyond primary that secondary and tertiary tuning can add up for equitable results. 

 

I tried 1T a couple times when I first built the machine but only with the XMP timings from the kit.  It wouldn't boot.  I suppose I'll have to try again.  I know 1T can produce excellent snappy results, I remember running 1T on AMD 1055T/i7-960 days. That was Samsung RAM too back then. Dunno if anyone remembers but it was the 'magic' RAM the 'ECO' branded no-heatspreader DDR3 1600 kits that would push 1866 @ 7-7-7 with like 1.35v...

 

Still even then I never dove too far into secondary/tertiary. 

 

Any recommended reading for calculations?  I've seen some Intel / AMD spec sheets with references but never equations...

[GanjaSMK]

Well RAM timings above were not resulting in error-free testing with Karhu; but currently hoping 16-18-18 @ 4200Mhz is (1.5v) as coverage so far about 1/5 way through 6500% coverage.

 

Will test 4K @UltraMega once I can see these stable enough.  Probably most of the night into tomorrow.  Gonna move the TV lol way easier.