someone fixed netcode and even hitreg and desync just by ram timings...

Everything about latency. Tips, testing methods, mouse lag, display lag, game engine lag, network lag, whole input lag chain, VSYNC OFF vs VSYNC ON, and more! Input Lag Articles on Blur Busters.
joseph_from_pilsen
Posts: 166
Joined: 01 Apr 2022, 23:51

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by joseph_from_pilsen » 29 May 2023, 07:03

CSGO has identical effect but not done by ram timings but simply by netcode.
Anything with a non zero ping has hitbox being lagging behind. So if a 100ms player peeks you, he cant be even prefired or 0reflex shotted, because the hitbox is lagging behind him nevertheless. You can try it, simply ask your friend to spoof latency to 100ms and let him peek dust2 doors and aim between them and try to prefire him. Even if you hit the visible target, it will deal 0 dmg as the hitbox is not there yet. Its so frustrating to get punished for outreacting someone who has 100ms lagcomp advantage above you and still losing. And if you wait a while for a hitbox approach, you cant win the duel neither, at this time you are already dead by his shot which will come always first because the 100ms peekers advantage.

User avatar
imprecise
Posts: 277
Joined: 16 Nov 2022, 13:47

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by imprecise » 31 May 2023, 14:04

winkli wrote:
29 May 2023, 01:57
Nice to see that people have found this.

Some addresses to your questions and skepticism:

Personally this is what I run at this time:
5.1GHz Unified core and cache on a 12900KF with 1.35V LLC3, IO LLC2 and AVX disabled entirely
5100MHz DDR4 @ 20-30-30-70-100 1T command Rate RAM, 2 sticks of 8gb Team Xtreem Black RGB 4800MHz
GALAX 3070ti with +850 or +1700MHz VRAM and locked core at first voltage point on boost (do not try +1700MHz on other GPUs, it will crash) Either keep your Vram OC on 0 or 850, (haven't tested 425 or anything else inbetween) [GPUs are known to change memory timings depending on vram speed and load - but NVIDIA does not specify or grant control on it, only AMD does]

PCIE 3 on all devices except the chipset since I'm trying out my PCIE wifi which is immune to USB priority delays
VL805 3.0 for mouse and keyboard USB controller card by FOVORE

I just set the priority of my USB controllers and WIFI to high, and all else to low and linebased, the AX210 Wifi onboard that I have will only operate on MSI mode enabled for that particular device.

When running single channel mode everything is a tad snapper and frame pacing is well smoother. At speeds this high Dual channel hardly matters to hit around 400 in CS and 240+ in valorant. These timings and placement train my ram at RTLs of 50/51, extremely low and on every single boot.

It matters that the external FPS cap is a huge improvement to hitreg over the in game cap, and the Ram timings ensure that the game is interacting with data storage and retrieval properly. Obviously this afects all games, but since Valorant's bandwidth is comically large on 128 tick interpolated, data I theorise would fire at a higher rate than CS:GO or any other shooter.

Oh yeah do not use Rawinput buffer if you use USB internet. It will buffer all USB devices on that including the Wifi/Ethernet adapter. Riot spaghetcode.

Now when I play there is 0 desync. Only you get the minimal peeker's advantage as you go upwards of 3300 hours into LEM-global elite MM, where pros and faceit enjoyers are quick enough and built to the point where they will be faster than you by training - and in the realms of human possibility.

If you are in the market to get new graphics cards, go with the 4070ti, 3070ti, and other TI cards. There's something about them that is likely just a better design. My 3080 runs like a brick compared to my 3070ti, galax for both, I've even had Zotac 3090's with reference PCB's, but so far the 70ti cards have been doing me good for shielding and good highly consistent input lag. Speculation aside, the Galax 4070ti and 3070ti series overclock the VRAM close to 17% over the stock settings with a simple thermal pad changeover to Gelid/Thermalrights, which is unheard of on any other card. I've mastered the 3070ti, but only had hands on the 4070ti for my customer recently -- the memory chips clock far better, beyond the 12000MHz mark in Afterburner.
Thanks for joining us in this discussion.

The biggest problems I see ingame from this sync issue is a feeling of unresponsive control, unpredictable collisions and inaccurate positioning of players. Mostly, it feels like my game is slightly behind and if a player affects me via the server it causes a slight visible desync and reposition. This means if I get aimpunched in CSGO even with full armor, it's up to ~400ms of lag/stun/reposition where the opponent gains full advantage, my footing and xhair position are adjusted, movement speed slow to recover, etc. In Rocket league, it causes opponents to hit the ball earlier than possible from their position and the ball will often change trajectory up to ~200ms after another player hits it.

Testing on single channel mode: In Unreal games and CSGO it seems like the sync to server is more accurate, but also seems a bit slow. Peeker advantage can be achieved but seems to be less than the expected 50-100ms, and getting hit by other players seems to cause a netcode stun where my client falls behind the action momentarily, but only about half a strong as before. Recovery from the stun is usually in time to adjust and win a fight instead of being stunned/unable to react for 400ms. I looked into it some more and found many tests show dual channel gives better performance and higher 1% lows. Have you tested your Intel setup in dual channel mode? I think you upgraded from an AMD, right? It looks like AMD suffers when outside the infinity fabric range (3600-4000mhz). I've been testing higher RAM clocks with looser timings and have found much more consistent performance so far, including 10-20% increased 2D benchmark scores. I didn't find any improvement with single channel over dual channel, found evidence that single channel would decrease FPS in most games, and it felt like sync was falling behind the server if a network event interrupted my actions. I did find an improvement by increasing RAM to 3600 or 3800mhz with very loose timings and now am thinking about whether it would be worth buying fresh DDR4 just for better timings.

Some relevant test videos:
Which Memory Kit for AMD Ryzen 5000? 4000MHz C15 Tested https://www.youtube.com/watch?v=LPTbx4fk3rw
Ryzen 5000 DDR4 Memory Performance, XMP vs Manual Timings, Single vs Dual Rank https://www.youtube.com/watch?v=AGux0pANft0
AMD Ryzen: 4 vs. 2 Sticks of RAM on R5 5600X for Up to 10% Better Performance https://www.youtube.com/watch?v=-UkGu6A-6sQ

I've been testing different MSI priorities but they seem to cause stutter lags when one high priority device interrupts the overall processing order. I read that having variables other than undefined add a layer to the drivers, which could definitely cause some hidden problems if the device doesn't support that kind of adjustment. In my testing, network card on high priority would cause full movement stun when getting hit during full sprint in CSGO, so it seemed like it was freezing the input/output to prioritize the network packets targeting my character ingame. I also tested USB devices on high only, but maybe still seeing some prioritization stutter lags and am currently testing undefined on all MSI devices.

winkli
Posts: 16
Joined: 05 May 2023, 21:52

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by winkli » 24 Jun 2023, 02:42

imprecise wrote:
31 May 2023, 14:04
winkli wrote:
29 May 2023, 01:57
Nice to see that people have found this.

Some addresses to your questions and skepticism:

Personally this is what I run at this time:
5.1GHz Unified core and cache on a 12900KF with 1.35V LLC3, IO LLC2 and AVX disabled entirely
5100MHz DDR4 @ 20-30-30-70-100 1T command Rate RAM, 2 sticks of 8gb Team Xtreem Black RGB 4800MHz
GALAX 3070ti with +850 or +1700MHz VRAM and locked core at first voltage point on boost (do not try +1700MHz on other GPUs, it will crash) Either keep your Vram OC on 0 or 850, (haven't tested 425 or anything else inbetween) [GPUs are known to change memory timings depending on vram speed and load - but NVIDIA does not specify or grant control on it, only AMD does]

PCIE 3 on all devices except the chipset since I'm trying out my PCIE wifi which is immune to USB priority delays
VL805 3.0 for mouse and keyboard USB controller card by FOVORE

I just set the priority of my USB controllers and WIFI to high, and all else to low and linebased, the AX210 Wifi onboard that I have will only operate on MSI mode enabled for that particular device.

When running single channel mode everything is a tad snapper and frame pacing is well smoother. At speeds this high Dual channel hardly matters to hit around 400 in CS and 240+ in valorant. These timings and placement train my ram at RTLs of 50/51, extremely low and on every single boot.

It matters that the external FPS cap is a huge improvement to hitreg over the in game cap, and the Ram timings ensure that the game is interacting with data storage and retrieval properly. Obviously this afects all games, but since Valorant's bandwidth is comically large on 128 tick interpolated, data I theorise would fire at a higher rate than CS:GO or any other shooter.

Oh yeah do not use Rawinput buffer if you use USB internet. It will buffer all USB devices on that including the Wifi/Ethernet adapter. Riot spaghetcode.

Now when I play there is 0 desync. Only you get the minimal peeker's advantage as you go upwards of 3300 hours into LEM-global elite MM, where pros and faceit enjoyers are quick enough and built to the point where they will be faster than you by training - and in the realms of human possibility.

If you are in the market to get new graphics cards, go with the 4070ti, 3070ti, and other TI cards. There's something about them that is likely just a better design. My 3080 runs like a brick compared to my 3070ti, galax for both, I've even had Zotac 3090's with reference PCB's, but so far the 70ti cards have been doing me good for shielding and good highly consistent input lag. Speculation aside, the Galax 4070ti and 3070ti series overclock the VRAM close to 17% over the stock settings with a simple thermal pad changeover to Gelid/Thermalrights, which is unheard of on any other card. I've mastered the 3070ti, but only had hands on the 4070ti for my customer recently -- the memory chips clock far better, beyond the 12000MHz mark in Afterburner.
Thanks for joining us in this discussion.

The biggest problems I see ingame from this sync issue is a feeling of unresponsive control, unpredictable collisions and inaccurate positioning of players. Mostly, it feels like my game is slightly behind and if a player affects me via the server it causes a slight visible desync and reposition. This means if I get aimpunched in CSGO even with full armor, it's up to ~400ms of lag/stun/reposition where the opponent gains full advantage, my footing and xhair position are adjusted, movement speed slow to recover, etc. In Rocket league, it causes opponents to hit the ball earlier than possible from their position and the ball will often change trajectory up to ~200ms after another player hits it.

Testing on single channel mode: In Unreal games and CSGO it seems like the sync to server is more accurate, but also seems a bit slow. Peeker advantage can be achieved but seems to be less than the expected 50-100ms, and getting hit by other players seems to cause a netcode stun where my client falls behind the action momentarily, but only about half a strong as before. Recovery from the stun is usually in time to adjust and win a fight instead of being stunned/unable to react for 400ms. I looked into it some more and found many tests show dual channel gives better performance and higher 1% lows. Have you tested your Intel setup in dual channel mode? I think you upgraded from an AMD, right? It looks like AMD suffers when outside the infinity fabric range (3600-4000mhz). I've been testing higher RAM clocks with looser timings and have found much more consistent performance so far, including 10-20% increased 2D benchmark scores. I didn't find any improvement with single channel over dual channel, found evidence that single channel would decrease FPS in most games, and it felt like sync was falling behind the server if a network event interrupted my actions. I did find an improvement by increasing RAM to 3600 or 3800mhz with very loose timings and now am thinking about whether it would be worth buying fresh DDR4 just for better timings.

Some relevant test videos:
Which Memory Kit for AMD Ryzen 5000? 4000MHz C15 Tested https://www.youtube.com/watch?v=LPTbx4fk3rw
Ryzen 5000 DDR4 Memory Performance, XMP vs Manual Timings, Single vs Dual Rank https://www.youtube.com/watch?v=AGux0pANft0
AMD Ryzen: 4 vs. 2 Sticks of RAM on R5 5600X for Up to 10% Better Performance https://www.youtube.com/watch?v=-UkGu6A-6sQ

I've been testing different MSI priorities but they seem to cause stutter lags when one high priority device interrupts the overall processing order. I read that having variables other than undefined add a layer to the drivers, which could definitely cause some hidden problems if the device doesn't support that kind of adjustment. In my testing, network card on high priority would cause full movement stun when getting hit during full sprint in CSGO, so it seemed like it was freezing the input/output to prioritize the network packets targeting my character ingame. I also tested USB devices on high only, but maybe still seeing some prioritization stutter lags and am currently testing undefined on all MSI devices.
I did some confirmation testing last night on my intel 4 dimm system, 12900kf/b660m mortar max, There are no other sets of timings under 1-1.5-3.5 that seem to be stable for netcode in valorant. Either the server data is processing 'too quickly' or 'too slowly' with the perfect balance there being between 1-1.5-3.5 and 1-2-4

On AMD systems with 2 slots of dimms I can manage 1-1.25-3.25 without issues.

I run my ram fast enough that the additional 20% of FPS gained by dual channel is pointless to me. At 400 FPS stable in CSGO and above, going higher just causes inconsistent input.

Single channel appears to stutter far less than dual channel so long as the ram is not potato slow. The inputs are more smooth and immediate than fast but poorly paced.

I have further MSI mode findings probably finishing that up entirely, with setting all devices that aren't the mouse and keyboard controller to low, including PCIE and PCI root hubs, GPU, and other USB controllers.

tehele
Posts: 1
Joined: 21 Aug 2023, 13:46

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by tehele » 30 Aug 2023, 03:45

winkli wrote:
24 Jun 2023, 02:42
imprecise wrote:
31 May 2023, 14:04
winkli wrote:
29 May 2023, 01:57
Nice to see that people have found this.

Some addresses to your questions and skepticism:

Personally this is what I run at this time:
5.1GHz Unified core and cache on a 12900KF with 1.35V LLC3, IO LLC2 and AVX disabled entirely
5100MHz DDR4 @ 20-30-30-70-100 1T command Rate RAM, 2 sticks of 8gb Team Xtreem Black RGB 4800MHz
GALAX 3070ti with +850 or +1700MHz VRAM and locked core at first voltage point on boost (do not try +1700MHz on other GPUs, it will crash) Either keep your Vram OC on 0 or 850, (haven't tested 425 or anything else inbetween) [GPUs are known to change memory timings depending on vram speed and load - but NVIDIA does not specify or grant control on it, only AMD does]

PCIE 3 on all devices except the chipset since I'm trying out my PCIE wifi which is immune to USB priority delays
VL805 3.0 for mouse and keyboard USB controller card by FOVORE

I just set the priority of my USB controllers and WIFI to high, and all else to low and linebased, the AX210 Wifi onboard that I have will only operate on MSI mode enabled for that particular device.

When running single channel mode everything is a tad snapper and frame pacing is well smoother. At speeds this high Dual channel hardly matters to hit around 400 in CS and 240+ in valorant. These timings and placement train my ram at RTLs of 50/51, extremely low and on every single boot.

It matters that the external FPS cap is a huge improvement to hitreg over the in game cap, and the Ram timings ensure that the game is interacting with data storage and retrieval properly. Obviously this afects all games, but since Valorant's bandwidth is comically large on 128 tick interpolated, data I theorise would fire at a higher rate than CS:GO or any other shooter.

Oh yeah do not use Rawinput buffer if you use USB internet. It will buffer all USB devices on that including the Wifi/Ethernet adapter. Riot spaghetcode.

Now when I play there is 0 desync. Only you get the minimal peeker's advantage as you go upwards of 3300 hours into LEM-global elite MM, where pros and faceit enjoyers are quick enough and built to the point where they will be faster than you by training - and in the realms of human possibility.

If you are in the market to get new graphics cards, go with the 4070ti, 3070ti, and other TI cards. There's something about them that is likely just a better design. My 3080 runs like a brick compared to my 3070ti, galax for both, I've even had Zotac 3090's with reference PCB's, but so far the 70ti cards have been doing me good for shielding and good highly consistent input lag. Speculation aside, the Galax 4070ti and 3070ti series overclock the VRAM close to 17% over the stock settings with a simple thermal pad changeover to Gelid/Thermalrights, which is unheard of on any other card. I've mastered the 3070ti, but only had hands on the 4070ti for my customer recently -- the memory chips clock far better, beyond the 12000MHz mark in Afterburner.
Thanks for joining us in this discussion.

The biggest problems I see ingame from this sync issue is a feeling of unresponsive control, unpredictable collisions and inaccurate positioning of players. Mostly, it feels like my game is slightly behind and if a player affects me via the server it causes a slight visible desync and reposition. This means if I get aimpunched in CSGO even with full armor, it's up to ~400ms of lag/stun/reposition where the opponent gains full advantage, my footing and xhair position are adjusted, movement speed slow to recover, etc. In Rocket league, it causes opponents to hit the ball earlier than possible from their position and the ball will often change trajectory up to ~200ms after another player hits it.

Testing on single channel mode: In Unreal games and CSGO it seems like the sync to server is more accurate, but also seems a bit slow. Peeker advantage can be achieved but seems to be less than the expected 50-100ms, and getting hit by other players seems to cause a netcode stun where my client falls behind the action momentarily, but only about half a strong as before. Recovery from the stun is usually in time to adjust and win a fight instead of being stunned/unable to react for 400ms. I looked into it some more and found many tests show dual channel gives better performance and higher 1% lows. Have you tested your Intel setup in dual channel mode? I think you upgraded from an AMD, right? It looks like AMD suffers when outside the infinity fabric range (3600-4000mhz). I've been testing higher RAM clocks with looser timings and have found much more consistent performance so far, including 10-20% increased 2D benchmark scores. I didn't find any improvement with single channel over dual channel, found evidence that single channel would decrease FPS in most games, and it felt like sync was falling behind the server if a network event interrupted my actions. I did find an improvement by increasing RAM to 3600 or 3800mhz with very loose timings and now am thinking about whether it would be worth buying fresh DDR4 just for better timings.

Some relevant test videos:
Which Memory Kit for AMD Ryzen 5000? 4000MHz C15 Tested https://www.youtube.com/watch?v=LPTbx4fk3rw
Ryzen 5000 DDR4 Memory Performance, XMP vs Manual Timings, Single vs Dual Rank https://www.youtube.com/watch?v=AGux0pANft0
AMD Ryzen: 4 vs. 2 Sticks of RAM on R5 5600X for Up to 10% Better Performance https://www.youtube.com/watch?v=-UkGu6A-6sQ

I've been testing different MSI priorities but they seem to cause stutter lags when one high priority device interrupts the overall processing order. I read that having variables other than undefined add a layer to the drivers, which could definitely cause some hidden problems if the device doesn't support that kind of adjustment. In my testing, network card on high priority would cause full movement stun when getting hit during full sprint in CSGO, so it seemed like it was freezing the input/output to prioritize the network packets targeting my character ingame. I also tested USB devices on high only, but maybe still seeing some prioritization stutter lags and am currently testing undefined on all MSI devices.
I did some confirmation testing last night on my intel 4 dimm system, 12900kf/b660m mortar max, There are no other sets of timings under 1-1.5-3.5 that seem to be stable for netcode in valorant. Either the server data is processing 'too quickly' or 'too slowly' with the perfect balance there being between 1-1.5-3.5 and 1-2-4

On AMD systems with 2 slots of dimms I can manage 1-1.25-3.25 without issues.

I run my ram fast enough that the additional 20% of FPS gained by dual channel is pointless to me. At 400 FPS stable in CSGO and above, going higher just causes inconsistent input.

Single channel appears to stutter far less than dual channel so long as the ram is not potato slow. The inputs are more smooth and immediate than fast but poorly paced.

I have further MSI mode findings probably finishing that up entirely, with setting all devices that aren't the mouse and keyboard controller to low, including PCIE and PCI root hubs, GPU, and other USB controllers.
What would you recommend for 5800x3D, G.Skill DIMM 16 GB DDR4-3200 (2x 8 GB) Dual-Kit (14-14-14-34) ?
If I calculate with 1.25 I would end up at 14 17.5 17.5 45.5 - should I round up or down?
Or better use 14 21 21 49?

Thanks in advance :-)

masneb
Posts: 239
Joined: 15 Apr 2019, 03:04

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by masneb » 02 Sep 2023, 21:41

Another it depends. Memory kits are all made for specific timings to run at and run best at. One kit will have different timings then another kit, so duplicating his timings on another kit not only probably wont work, it's a good way to make instability in your system.

If your kit isn't completely stable it definitely can cause issues.

User avatar
imprecise
Posts: 277
Joined: 16 Nov 2022, 13:47

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by imprecise » 02 Sep 2023, 22:51

tehele wrote:
30 Aug 2023, 03:45

What would you recommend for 5800x3D, G.Skill DIMM 16 GB DDR4-3200 (2x 8 GB) Dual-Kit (14-14-14-34) ?
If I calculate with 1.25 I would end up at 14 17.5 17.5 45.5 - should I round up or down?
Or better use 14 21 21 49?

Thanks in advance :-)
I'd say check hardwareinfo (https://www.hwinfo.com/) and tune based on the specs of your RAM kit.

Here is mine. I used the chart on the bottom and compared to several kits for sale.
image_2023-09-02_203802160.png
image_2023-09-02_203802160.png (6.56 KiB) Viewed 6171 times
Warning: Setting voltage too high can fry your components, be careful.
Generally it takes 1.35V-1.4V to run at 3600+ frequency and low timings. I tried 3800mhz for a while but it was choppy feeling and the timings were something like 27 on auto. I dropped back to 3600mhz, set to 1.4V and shaved a few points off the automatic 25 timings, and my ingame Rocket League sync is almost perfect, even online. The RAM is still ranked about 10-15% lower than the rest of my components so I would likely still benefit from upgrading.

Based on your AMD CPU with infinity fabric, the RAM should be running between 3600-4000mhz. Based on your 3200CL14 timings, your RAM can probably run 3600 at CL16 (maybe even 14 at higher voltage). I would try 1.35V or less at 3600 and see if you can find stable settings. I doubt CL14 is worth running if requires a higher voltage, as you will only lose 4ns latency by going up to 16.

These are generalized settings for RAM optimization, not the proposed settings from the video.
Warning: Setting voltage too high can fry your components, be careful.

I've been experimenting at lower frequencies and trying to match the ratios from this thread, but haven't had any success.
In the video there was mention of being able to hold up to faceit lvl 8 with these settings, which makes me think it's more in-sync at the cost of higher overall latency, similar to how you can get better sync by capping at 60FPS because it's much less demanding on the network and system, but also a disadvantage against players with higher FPS. If the memory can run 20% faster while staying in sync, that could bring the level 8 to level 9.6 by itself, and I do think this is in line with the video as it sounded like a system rating and not a skill rating.

Update:
Experimenting with 400mhz (dual channel @800) with 12-12-12-36-360 in CS2 gave me 50-100FPS but the input felt almost perfect. The frames made the game look like a slideshow, but the responsive feeling didn't go away. I think this is how laggy players are still able to play, they have RAM that allows much faster response latency for inputs even though the game is running very choppy.

Doubling the RAM frequency effectively doubled the FPS, but I wasn't able to test much because steamwebhelper keeps trying to use 10-20% CPU nonstop, and I've resorted to hard throttling it in Process Lasso after trying different steam settings.

It does appear there is a tradeoff between optimal latency and maximum frames happening here.
Last edited by imprecise on 11 Sep 2023, 02:28, edited 1 time in total.

Jackpot71
Posts: 9
Joined: 13 Jun 2023, 14:50

Re: someone fixed netcode and even hitreg and desync just by ram timings...

Post by Jackpot71 » 05 Sep 2023, 00:55

I fixed my hitreg/desync problems with some new settings in MSI Utility AND RAM Settings.
How good are the last few frames and the absolute lowest DPC latency if for some reason it doesn't run in sync with the game?
I then thought to myself, there are probably good frames and bad frames. I wanted only the good frames to remain.

I set the most settings in msi tool to undefined and low but have all devices that support msi mode in msi mode
no msi mode for the ethernet adapter 219V.
Interrupt affinity settings only for gpu and mouse (GPU set core to best low 1%)

The biggest impact was change the ram timings. I allways had XMP Profile active. After i overlock the ram a little and
try to set up some timings by myself it changed everything. Need a little bit time and cmos clears to find the working timings for crash-free play but then i found some and it works well. The timings are also a lot lower than those specified in xmp.

I think now the deeper you dive into optimizing, the more important it becomes to know 100% exactly which setting does what in the system or which settings change related things. If, like me, you simply try to optimize everything with the best settings, there is a risk that things will not harmonize properly and it will lead to desynchronization and resulting Hitreg problems.
I also tried to ask a few professionals about their settings. Some just know where the PC turns on and still play like gods...wtf?!!!

you don't turn a gtx970 and a gen7 cpu into a 4090 with gen13 close to the server...i or we have to deal with it.

Post Reply