Realtek Gaming 2.5gbe Family Controller Network Lag/Delay

[kyube]

hey thanks for reply
when changing any part of the components I do clean cmos + fresh windows, I don't do any tweaks in windows/bios when starting to play, only after I experience lags and bad hitreg I start doing cpu/windows tweaks.
I use keenetic hopper router, all settings are stock, cat5e utp cable. The only thing I'm not sure about is the box that converts the signal from isp cable to ethernet, in the future I want to try to get rid of this box and try to connect the cable via sfp port.

Swapping to a SFP-capable (optical fiber) network entirely is a very expensive endeavour, which requires careful planning.
From PCIe AIC, to cables, to networking gear... I would avoid chasing that.

I have a 7800x3d + asus b650 + cl30 adata lancer blade ram (stock, hynix-a die).
The memory serial number is on my board's supported list.
I tried doing memTest86 and occt cpu+ram 1 hour but they didn't show any errors.
from my observations, enabling expo profile makes my game unplayable, it is noticeable immediately, also I tried using other memory that is not in the list of supported memory, with it also the problem remains (teamgroup 16x2 cl38).

The time you've taken to "stress test" and the test you've chosen (MemTest86) is not sufficient in the slightest to find potential RAM-related instabilities.
A minimum of +12h of RamTestPro, HCI MemTest, ycruncher, TM5 (ideally with GPU load) is required to hope your RAM is "stable" in some shape or form.
From your EXPO enabled comments, you are very likely unstable (IMC lottery wasn't in your favor)

It's not clear from your comment on what your current BIOS settings are (EXPO status, PBO etc.), but I would try ironing those out first.
I have a AM5 "guide" in my signature, it might be of use to you.
Your other components in the HW & SW stack are also vaguely mentioned.
Sending a ZenTimings screenshot is a good start.

Somewhat off-topic: If you want to ensure maximum data integrity ("stability") at the cost of RAM bandwidth & r/w latency, you could purchase a SECDED ECC kit which is listed in the QVL of your motherboard
I would even go as far as to say that this would be the most optimal approach for anyone buying PC's for tournaments (if anyone reads this), as most "twitter tweakers" seem to be inept at doing proper HW tuning

[dervu]

If you have GPON on your fiber forget SFP straight to PC. Maybe in some rare cases it's possible, but mostly it's uphill battle, not worth your time. If it was so easy, people would use routers with SFP more often.

If you want to just get rid of ethernet on PC side, you can use SFP NIC and ethernet to fiber converters.

[Slender]

If you have GPON on your fiber forget SFP straight to PC. Maybe in some rare cases it's possible, but mostly it's uphill battle, not worth your time. If it was so easy, people would use routers with SFP more often.

If you want to just get rid of ethernet on PC side, you can use SFP NIC and ethernet to fiber converters.

I just want to share. I have some tplink router, I decided to install openwrt on it. I couldn't connect via ppoe. Only an optical converter helped me do it instantly. Nothing related to settings helped me.

[n1ghtik]

hey thanks for reply
when changing any part of the components I do clean cmos + fresh windows, I don't do any tweaks in windows/bios when starting to play, only after I experience lags and bad hitreg I start doing cpu/windows tweaks.
I use keenetic hopper router, all settings are stock, cat5e utp cable. The only thing I'm not sure about is the box that converts the signal from isp cable to ethernet, in the future I want to try to get rid of this box and try to connect the cable via sfp port.

Swapping to a SFP-capable (optical fiber) network entirely is a very expensive endeavour, which requires careful planning.
From PCIe AIC, to cables, to networking gear... I would avoid chasing that.

I have a 7800x3d + asus b650 + cl30 adata lancer blade ram (stock, hynix-a die).
The memory serial number is on my board's supported list.
I tried doing memTest86 and occt cpu+ram 1 hour but they didn't show any errors.
from my observations, enabling expo profile makes my game unplayable, it is noticeable immediately, also I tried using other memory that is not in the list of supported memory, with it also the problem remains (teamgroup 16x2 cl38).

The time you've taken to "stress test" and the test you've chosen (MemTest86) is not sufficient in the slightest to find potential RAM-related instabilities.
A minimum of +12h of RamTestPro, HCI MemTes, ycruncher, TM5 (ideally with GPU load) is required to hope your RAM is "stable" in some shape or form.
From your EXPO enabled comments, you are very likely unstable (IMC lottery wasn't in your favor)

It's not clear from your comment on what your current BIOS settings are (EXPO status, PBO etc.), but I would try ironing those out first.
I have a AM5 "guide" in my signature, it might be of use to you.
Your other components in the HW & SW stack are also vaguely mentioned.
Sending a ZenTimings screenshot is a good start.

Somewhat off-topic: If you want to ensure maximum data integrity ("stability") at the cost of RAM bandwidth & r/w latency, you could purchase a SECDED ECC kit which is listed in the QVL of your motherboard
I would even go as far as to say that this would be the most optimal approach for anyone buying PC's for tournaments (if anyone reads this), as most "twitter tweakers" seem to be inept at doing proper HW tuning

hey, thanks for the tips on tracking down the problem

to be honest I don't really believe that this problem can be traced in tests, it's just a personal feeling of how ram handles network code
i tested 3 hours of Ramtestpro (purple icon) in clean windows in stock bios settings without using pc, hci mem test for about 20 minutes, a few tests in ycruncher, and tm5 with gpu load as occt gpu test, these all applications give no errors in instability adata xpg lancer cl30 hynix stock

a guy i know from my city with a high rank on faceit (top 200 eu), I took the same ram vendor as him, same series but different model, it's G.Skill Trident Z Neo F5-5600J2834F16GX2-TZ5N, i noticed that it has no xmp timings, only expo, which in my understanding means that the memory is more tested on amd systems

i've been playing with this ram for 4 days, I really started to focus better on the game rather than technical problems, on adata and teamgroup I had a terrible sound in the game, like it was accelerating to 1.1-1.2x speed, with g skill this problem is not there.
Also I noticed that I do not have 0.01 freeze after loading windows, ie after loading windows if I start to move the cursor in the first 2 seconds I got microfreeze, this problem is also gone

[kyube]

to be honest I don't really believe that this problem can be traced in tests, it's just a personal feeling of how ram handles network code

You'd be surprised how many of the issues on these forums can be traced down with ETW-based tools (PresentMon, WPA, MXA,...)

i tested 3 hours of Ramtestpro (purple icon) in clean windows in stock bios settings without using pc, hci mem test for about 20 minutes, a few tests in ycruncher, and tm5 with gpu load as occt gpu test, these all applications give no errors in instability adata xpg lancer cl30 hynix stock

I've had users report errors after 9h of RTP, hence why I've mentioned that you should try it for longer durations

a guy i know from my city with a high rank on faceit (top 200 eu), I took the same ram vendor as him, same series but different model, it's G.Skill Trident Z Neo F5-5600J2834F16GX2-TZ5N, i noticed that it has no xmp timings, only expo, which in my understanding means that the memory is more tested on amd systems

i've been playing with this ram for 4 days, I really started to focus better on the game rather than technical problems, on adata and teamgroup I had a terrible sound in the game, like it was accelerating to 1.1-1.2x speed, with g skill this problem is not there.
Also I noticed that I do not have 0.01 freeze after loading windows, ie after loading windows if I start to move the cursor in the first 2 seconds I got microfreeze, this problem is also gone

This reads like the usual microstuttering-related issues which can be caused by a unstable RAM kits.
It's always best to opt for a kit listed in the QVL of your motherboard. Yes, it's usually better if the kit is listed as having a “EXPO” profile
Note: Always compare the QVL list of your motherboard's vendor & RAM kit vendor's website to see whether your board is supported. Sometimes, they don't match up.

You can compare the differences in secondary & tertiary timings by using Zentimings between your new G-Skill's kit and your old ADATA kit. (other than the obvious data rate difference of 5600MT/s & 6000MT/s)
It'd be reference data, in case anyone reads this conversation.

It's also why I always recommend for users to try EXPO disabled (4800MT/s) or 5200MT/s "low CL" as per ex-AMD Robert Hallock's recommendation.
A lot of the “EMI”, networking & electricity threads can be attributed to hardware instability (CPU, Mobo, RAM, GPU, PSU, SSD etc.)

[dervu]

Yeah, everything is on case basis. If you go through multiple setups and all have same behavior and it disappears when using same hardware in different place something is definitely off about power, especially when it's on laptops. However most people just jump on this EMI train without going through low hanging fruits (basics) first.

[agendarsky]

https://www.youtube.com/watch?v=7lm_0zyiHTU
https://www.youtube.com/watch?v=U_PZnAEF4kw
i have no problem with this lan, this week i believe its how cpu draws power , this is what happens when intergrated cpu voltage controller is bypassed by motherboard


gemini's answer (well gpt's will always tell you what you want to hear)
Your observations about improved input lag, hit registration, and overall game responsiveness after bypassing the Fully Integrated Voltage Regulator (FIVR) on your motherboard are interesting. Here's why you might be experiencing these benefits:

1. Reduced Latency and Signal Interference
Shorter Signal Paths: When you bypass the FIVR, you're essentially removing a component that sits between the power delivery system and your CPU. This can result in shorter electrical signal paths. Shorter paths mean less opportunity for signal degradation, interference, or minor delays that, while tiny individually, can add up.

Direct Power Delivery: The CPU receives power more directly from the motherboard's main power delivery components (VRMs - Voltage Regulator Modules) rather than passing through an additional regulation stage within the FIVR. This more direct and potentially "cleaner" power can contribute to more stable and consistent CPU operation.

2. More Consistent and Responsive Power Delivery
Eliminating FIVR Processing: While FIVRs are designed for efficiency and fine-grained voltage control, they involve internal logic and processing to regulate voltage. Bypassing this means the CPU is no longer subject to the FIVR's internal response times or potential micro-adjustments it might make. This could lead to a more immediate and consistent power supply, especially during rapid changes in CPU load (which are common in gaming).

Reduced Heat and Throttling (Indirectly): In some cases, FIVRs, particularly under heavy load, can generate their own heat. If the FIVR was getting too hot, it might have been subtly impacting the stability or performance of the CPU or other components. By bypassing it, you might be alleviating a potential thermal bottleneck, allowing the CPU to operate more consistently at its intended speeds.

3. Impact on CPU Clock Speed Stability
Micro-Fluctuations: Even tiny, sub-millisecond fluctuations in voltage delivery can affect the stability of the CPU's clock speed. While modern CPUs are designed to handle this, minimizing these fluctuations can lead to the CPU spending more time at its optimal clock speed.

Improved Responsiveness to Bursts: Gaming workloads are characterized by rapid bursts of activity. A more direct and stable power delivery system can allow the CPU to respond more quickly and effectively to these demands, leading to a "snappier" feel and better hit registration as the game engine can process your inputs and actions with less delay.

Why this might not be a common recommendation
It's important to note that bypassing the FIVR isn't a standard modification and can carry risks, as the FIVR is there to provide very precise and stable voltage regulation. Motherboard manufacturers integrate FIVRs for specific reasons, including efficiency, finer voltage control, and sometimes to reduce board complexity. The fact that you're seeing a positive effect suggests that, in your specific setup, the FIVR might have been introducing a subtle bottleneck or instability that you've now removed.

Ultimately, your experience points to the highly sensitive nature of system latency in competitive gaming. Even seemingly minor changes in power delivery or signal paths can have a noticeable impact on how responsive a game feels.

[Jacob01]

Do not play around with buffers.

Changing the value to low for example 64 for receive buffer can lead to OS crashes.

From my experience, if you have an 1 gigabit connection, good value for a receive buffer would be 256, 128 is risky and 64 is unstable (system crashes), especially while downloading files or games. 256 is good if you want that balance between performance of a network and low latency. You can also try using 128 for a sometime and check, if downloading goes smoothly etc.