Need help configuring Intel 82574L NIC for better performance

[Hardcore]

On my all other PCs for last ~8 years I always played on external PCI network cards, they always worked better than integrated ones, even considering integrated worked 'just fine'.

But now I bought motherboard without PCI slots, so I had to buy new PCI-e network card. So I bought Intel 82574L CT Gigabit Desktop Network Adapter(PCI\VEN_8086&DEV_10D3&SUBSYS_A01F8086&REV_00) to play video games. Integrated NIC on this new motherboard works just fine, but you can always do better than default or integrated.

Here's screens of technologies available to me in NIC's control panel:
http://i.imgur.com/qmfuiTc.png
http://i.imgur.com/n4w9t8y.png

I don't use any kinds of VPNs, routers, modems, etc. Cable goes directly into network commutator of ISP(I believe it's Cisco unless they changed it). If model will make a difference on settings, tell me and I'll get the exact model of the network commutator of ISP. My ISP using DHCP to configure network. I get dedicated dynamic(it's different every time I reconnect) white IP address, so I can setup a server on my PC.

Values of properties:

Adaptive Inter-Frame Spacing: Disabled
Flow Control: Rx & Tx Enabled
Gigabit Master Slave Mode: Auto Detect
Interrupt Moderation: Enabled
Interrupt Moderation Rate: Adaptive(other possible values: off, minimal, low, medium, high, extreme)
IPv4 Checksum Offload: Rx & Tx Enabled
Jumbo Packet: Disabled
Large Send Offload V2(IPv4): Enabled
Large Send Offload V2(IPv6): Enabled
Locally Administered Address: Not Present
Log Link State Event: Enabled
Maximum Number of RSS Queues: 2 Queues
Packet Priority & VLAN: Packet Priority & VLAN Enabled
Receiver Buffers: 256(minimum value is 80, max - 2048)
Receive Side Scaling: Enabled
Speed & Duplex: Auto Negotiation(works in 100 MB Full Duplex mode)
TCP Checksum Offload (IPv4): Rx & Tx Enabled
TCP Checksum Offload (IPv6): Rx & Tx Enabled
Transmit Buffers: 512(minimum value is 80, max - 2048)
UDP Checksum Offload (IPv4): Rx & Tx Enabled
UDP Checksum Offload (IPv6): Rx & Tx Enabled
Wait for Link: Auto Detect

And there are more properties in TCP Optimizer(and who knows how many more you can tweak in registry manually):
http://i.imgur.com/PyzTZCx.png
http://i.imgur.com/kEwc4ss.png

Screenshoted values above are defaults for all of the properties.

I tried to read descriptions about these technologies but unfortunately I am unable to set proper values for all of these properties. I don't know how to test those values and when I tweak all of them, sometimes it works faster than default, sometimes slower. I measure it by how fast I get confirmation of a kill in video games such as CS:GO or Insurgency. I also tried to read different guides on the internet and different people suggest to turn off or turn on the exact same properties of NIC, they also explain and both explanations make sense, so I am completely lost what values should I set for these properties.

It would be nice if there was some program to monitor packets and then by seeing how many packets there are, what are sizes of these packets, etc., one could determine proper settings for these properties.

Could someone help me to set proper values for these properties, guys? If you need any other info, don't hesitate to ask, I'll provide as many info as I can. If you know some articles and programs which can help me to determine proper values, please post them here.

Thanks in advance, gurus!

 

[MichaelCG]

"Personally" I think you are way over thinking this....it is highly unlikely your tweaks of any of these settings will have as much impact as the quality of your Internet connection, Internet latency, Internet routing, and server performance. Unless you are pushing near line-rate, it is unlikely any modern network card is going to need to be "tweaked" for performance purposes that will make any noticeable difference.

Now keep in mind...I am not a tuner/tweaker of settings anymore. I was in my younger years for sure. I changed any setting I could trying to get as much performance I could. These days, I prefer simple and stable.

 

[Hardcore]

"Personally" I think you are way over thinking this....it is highly unlikely your tweaks of any of these settings will have as much impact as the quality of your Internet connection, Internet latency, Internet routing, and server performance. Unless you are pushing near line-rate, it is unlikely any modern network card is going to need to be "tweaked" for performance purposes that will make any noticeable difference.

Now keep in mind...I am not a tuner/tweaker of settings anymore. I was in my younger years for sure. I changed any setting I could trying to get as much performance I could. These days, I prefer simple and stable.

Maybe I didn't specify it. I'm trying to set it up for games like CS:GO(not for this particular game), where these 2 milliseconds actually can and do make all the difference. And when I tweak them, they make a difference. When I twek all of them, sometime it's faster, sometime it's slower. But I don't know how to determine best value for every setting

 

[MichaelCG]

I understand what you are saying....but, 2ms of latency is well within the margins of latency fluctuations across the Internet. I still say none of these settings 'should' change your latency by 2ms. Most of these settings are more for CPU offload when under high traffic volumes.

Your Interrupt Moderation settings would probably have the most impact...it is a trade off between host CPU usage and response times...although we are generally talking 'ns' here instead of 'ms'. Unless your host CPU is already pegged out and/or your network connection is buried, most of these can all be disabled without much difference. I remember these settings back in the 10Mbps days when your CPU was always the bottleneck and all of these offload functions/features were a life-saver. A couple of Mbps of traffic then could easily chew through 50% or more of your CPU without having the proper NIC and drivers to help with offload.

If you were talking about gaming on a local LAN only and you were able to test and confirm 2ms difference, I would be on-board with you. But if you are talking about Internet gaming...all bets are off since there are way too many other variables that are outside of your control. If you really want to test and validate that a change improved or degraded performance, you must have a repeatable and stable test environment that you control all variables.

I'm sure there will be other responses coming from others that will be more supportive of your efforts than mine. Good luck and have fun....I do miss the days of doing what you are doing...just no time for it anymore.

 

[MichaelCG]

If you haven't seen this already, it somewhat talks about some of the options...especially the Interrupt Moderation one I was talking about:

http://blog.serverfault.com/2011/03/23/performance-tuning-intel-nics/

 

[Hardcore]

If you were talking about gaming on a local LAN only and you were able to test and confirm 2ms difference, I would be on-board with you. But if you are talking about Internet gaming...all bets are off since there are way too many other variables that are outside of your control. If you really want to test and validate that a change improved or degraded performance, you must have a repeatable and stable test environment that you control all variables.

I was thinking of this kind of tests:

Combined with some kind of packet monitoring software.

I understand what you are saying....but, 2ms of latency is well within the margins of latency fluctuations across the Internet. I still say none of these settings 'should' change your latency by 2ms. Most of these settings are more for CPU offload when under high traffic volumes.

I'm not sure, but I think last time I have 'turned off' everything(like all offloading options) or just changed every settings I could just for the sake of changing it and seeing if I can see something with my eye. So last time I did it, I could say that after I shoot and hit other player in a head, his model would start falling with a very big delay(compared to default settings) which can be caught with an eye. And that's on the same server with ~30 ms ping. What I would like to have is some kind of tracking system, so I can actually see it in numbers. But I don't even know what I have to monitor(definitely not ping)

If you haven't seen this already, it somewhat talks about some of the options...especially the Interrupt Moderation one I was talking about:
http://blog.serverfault.com/2011/03/23/performance-tuning-intel-nics/

Thanks, I never saw this one. I'll read it through.

 

[System Error Message]

that depends. Many of these games now host their own servers so you cant host your own server. There are also other factors in play such as latency of the input system which was measured on linus tech tips to compare various sync settings.

Hit detection in games can vary as guns can also have deviance in accuracy too,. A lot of math goes into these games. The best way to measure the performance isnt like this. For example in space engineers if you have 2 people in the same ship and it goes at maximum speed and both players are on LAN and both the players and server arent maxed out on hardware than it becomes a test of latency. Most of the time you'll just be measuring latency over the net rather than NIC performance.

NIC performance is measured by how much less CPU is used and how many packets it can handle before being overloaded (such as filling a 1Gb/s NIC with 64 byte packets is extremely taxing on network hardware). There are various ways to measure the performance and bufferbloat is also a factor too.