X570 Realtek vs Intel LAN

Realtek vs Intel LAN


  • Total voters
    50

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
Which is better for gaming performance / latency etc? The Intel 1GbE (Intel® I211-AT Gb LAN)
or Realtek 2.5 GbE?
 
Which is better for gaming performance / latency etc? The Intel 1GbE (Intel® I211-AT Gb LAN)
or Realtek 2.5 GbE?
If I could get the 2.5 to work I'd let you know, might be a driver issue on my side.
 
I did a speed test on both of mine for fun with a 10gb file. They almost didn’t differ at all. On a gigabit wired network.

also what about cpu usage between the two? i'm also more interested in latency than raw bandwidth
 
You should get the most out of your network by tuning, using qos and make sure your MTU/MRU sizing is good on your router (to prevent fragmentation and set small packet size optimal). I also set steam to grab the even cores since i multitask and have hyperthreading enabled. Disable Steam browser.

This is for 6 core 12 thread,

batch code:
C:\WINDOWS\System32\cmd.exe /c start /affinity 555 F:\Games\Steam\Steam.exe -no-browser +open steam://open/minigameslist
 
Last edited:
The difference in cpu use between a Realtek NIC and an Intel NIC is inconsequential on a modern system.

Right, this is a pointless discussion; processors are so powerful now that you won't notice any performance difference on what is now such a small system load.

The only real reason he might get the 2.5gbs Realtek is if he's planning on upgrading his network switch in the near future (it will work on old Cat5e).

2.5gbps is a lot faster than 1Gbps (if you have a local NAS that's pushing 120MB/s or more).
 
Last edited:
Can't find a Latency Benchmark.. yet.



Intel

index.png


Realtek

index.png
 
  • Like
Reactions: erek
like this
TheFpsReview - LAN

"The ASUS ROG Crosshair VIII Hero WiFi integrates two adapters. The first is the standard Intel i211-AT. This is a ubiquitous adapter that can be found on a variety of motherboards. It is the option for Intel and non-Intel based motherboards alike. It is the gold standard for reliability and compatibility. The second is the Realtek RTL8125-CG 2.5G LAN adapter. This is a controller I don’t have a lot of experience with. Historically, I haven’t been a huge fan of Realtek adapters, but again, this isn’t a common controller, so I don’t have much to say about it outside of actual testing.

RTL-8125-CG @ 1Gb
In our first network controller showed an average transfer rate of 69.49MB/s in the write test and 45.46MB/s in the read test. The maximum transfer rate was 78.26MB/s in the write test and 50.90MB/s in the read test. Our minimum transfer rates were 66.40MB/s in the write test and 40.83MB/s in the read test. CPU usage was 2% in the write test and 1% in the read test.

i211-AT @ 1Gb
In the write test, our average transfer rate was 54.29MB/s. In the read test, we saw a result of 47.83MB/s. The maximum transfer rates were on the low side at 55.68MB/s for the write test and 48.56MB/s for the read test. Our minimum transfer rates were 51.96MB/s and 41.02MB/s in the write and read tests respectively. CPU usage is a modest 2% in the write test and 1% in the read test."\

From Intel site:
Product Brief
IPv4 and IPv6 checksum offload, TCP/UDP checksum offload, extended Tx descriptors for more offload capabilities, up to 256 KB TCP segmentation (TSO v2), header splitting, 40 KB packet buffer size, and 9.5 KB Jumbo Frame support.
 
Last edited:
  • Like
Reactions: erek
like this
TheFpsReview - LAN

"The ASUS ROG Crosshair VIII Hero WiFi integrates two adapters. The first is the standard Intel i211-AT. This is a ubiquitous adapter that can be found on a variety of motherboards. It is the option for Intel and non-Intel based motherboards alike. It is the gold standard for reliability and compatibility. The second is the Realtek RTL8125-CG 2.5G LAN adapter. This is a controller I don’t have a lot of experience with. Historically, I haven’t been a huge fan of Realtek adapters, but again, this isn’t a common controller, so I don’t have much to say about it outside of actual testing.

RTL-8125-CG @ 1Gb
In our first network controller showed an average transfer rate of 69.49MB/s in the write test and 45.46MB/s in the read test. The maximum transfer rate was 78.26MB/s in the write test and 50.90MB/s in the read test. Our minimum transfer rates were 66.40MB/s in the write test and 40.83MB/s in the read test. CPU usage was 2% in the write test and 1% in the read test.

i211-AT @ 1Gb
In the write test, our average transfer rate was 54.29MB/s. In the read test, we saw a result of 47.83MB/s. The maximum transfer rates were on the low side at 55.68MB/s for the write test and 48.56MB/s for the read test. Our minimum transfer rates were 51.96MB/s and 41.02MB/s in the write and read tests respectively. CPU usage is a modest 2% in the write test and 1% in the read test."\

From Intel site:
Product Brief
IPv4 and IPv6 checksum offload, TCP/UDP checksum offload, extended Tx descriptors for more offload capabilities, up to 256 KB TCP segmentation (TSO v2), header splitting, 40 KB packet buffer size, and 9.5 KB Jumbo Frame support.
Absolutely junk test.

throw it out.

He didn't use a 2.5gbps controller and both NICs at 1 Gigabit LAN should absolutely max out at a little over 112MB/s on a typical network both upload and download with real world scenarios if things were configured even with default settings. Who knows what was wrong in his tests, but something was desperately wrong.

That reviewer clearly knows not one thing about networking.
 
Absolutely junk test.

throw it out.

He didn't use a 2.5gbps controller and both NICs at 1 Gigabit LAN should absolutely max out at a little over 112MB/s on a typical network both upload and download with real world scenarios if things were configured even with default settings. Who knows what was wrong in his tests, but something was desperately wrong.

That reviewer clearly knows not one thing about networking.

This may be a *whoosh* moment but:

I'm quite sure that guy knows a thing or two about testing and you may come across him around here one time or another.

I'm also guessing said reviewer isn't using one big 10GB file to test max throughput of 1gigabit links.
 
  • Like
Reactions: Mega6
like this
Absolutely junk test.

throw it out.

He didn't use a 2.5gbps controller and both NICs at 1 Gigabit LAN should absolutely max out at a little over 112MB/s on a typical network both upload and download with real world scenarios if things were configured even with default settings. Who knows what was wrong in his tests, but something was desperately wrong.

That reviewer clearly knows not one thing about networking.


ok.. in gaming, throughput really doesn't matter since it's so low and the reviewer only had a gigabit lan - which is good enough for gaming by a mile. Anyways can't find a latency test. The Intel nic has a lot of good hardware, offloading features. My suspicion is that the latency is lower on the Intel.

The OP can just ping his router, gameserver, gateway,, ect. and other tests with both nics configured and find the difference - if there is any on latency.
 
Last edited:
This may be a *whoosh* moment but:

I'm quite sure that guy knows a thing or two about testing and you may come across him around here one time or another.

I'm also guessing said reviewer isn't using one big 10GB file to test max throughput of 1gigabit links.
I hate to break your heart but he's wrong as wrong can be on this topic.

Here's my 1GB Intel NIC on a ~700MB file on my 1GB LAN from a source and destination that are capable fo maxing out typical real world transfer rates.
upload_2020-1-13_20-58-4.png


from my PC Intel NIC to NAS
upload_2020-1-13_20-43-23.png


from NAS to my PC Intel NIC
upload_2020-1-13_20-49-14.png


NIC properties (all default Windows 10)
upload_2020-1-13_20-44-6.png


Simple unmanaged Trendnet 8 port 1Gb switch makes the connection between my X99 based motherboard's Intel NIC to my cheap $150 QNAP 231P NAS.

He shows/says he uses a 800MB zip file. So very similar test.
Again, he doesn't know how setup or test network equipment if he's using an 800MB zip file and receiving results like that and thinks that's to be expected.


Switching NICs - going to do it again with my Killer NIC for comparison.

Killer NIC - from NAS to PC
upload_2020-1-13_20-56-13.png


From PC's Killer NIC to NAS
upload_2020-1-13_20-57-25.png


Killer NIC properties (all default) - same exact network structure - just flipped the cable from one NIC to the other.
upload_2020-1-13_20-59-0.png
 
Last edited:
I hate to break your heart but he's wrong as wrong can be on this topic.

Here's my 1GB Intel NIC on a ~750 mb file on my 1GB LAN from a source and destination that are capable fo maxing out typical real world transfer rates.

from my PC to NAS
View attachment 215670
from NAS to my PC
View attachment 215682
NIC properties (all default Windows 10)
View attachment 215673

Simple unmanaged Trendnet 8 port switch bridges the connection between my X99 based motherboard's Intel NIC to my cheap $150 QNAP 231P NAS.

He shows/says he uses a 800MB zip file. So very similar test.
Again, he doesn't know how setup or test network equipment.


Switching NICs - going to do it again with my Killer NIC for comparison.



112MB/s == 896Mb/s

Pretty close, what's the problem exactly?
 
112MB/s == 896Mb/s

Pretty close, what's the problem exactly?
This is the problem: His 1Gb LAN test are showing WAY short of max real world practical results, and he's trying to compare two NICs with benchmarks using invalid results.

RTL-8125-CG @ 1Gb
In our first network controller showed an average transfer rate of 69.49MB/s in the write test and 45.46MB/s in the read test. The maximum transfer rate was 78.26MB/s in the write test and 50.90MB/s in the read test. Our minimum transfer rates were 66.40MB/s in the write test and 40.83MB/s in the read test. CPU usage was 2% in the write test and 1% in the read test.

i211-AT @ 1Gb
In the write test, our average transfer rate was 54.29MB/s. In the read test, we saw a result of 47.83MB/s. The maximum transfer rates were on the low side at 55.68MB/s for the write test and 48.56MB/s for the read test. Our minimum transfer rates were 51.96MB/s and 41.02MB/s in the write and read tests respectively. CPU usage is a modest 2% in the write test and 1% in the read test."
 
Last edited:
This is the problem: His 1Gb LAN test are showing WAY short of max real world practical results, and he's trying to compare two NICs with benchmarks using invalid results.

RTL-8125-CG @ 1Gb
In our first network controller showed an average transfer rate of 69.49MB/s in the write test and 45.46MB/s in the read test. The maximum transfer rate was 78.26MB/s in the write test and 50.90MB/s in the read test. Our minimum transfer rates were 66.40MB/s in the write test and 40.83MB/s in the read test. CPU usage was 2% in the write test and 1% in the read test.

i211-AT @ 1Gb
In the write test, our average transfer rate was 54.29MB/s. In the read test, we saw a result of 47.83MB/s. The maximum transfer rates were on the low side at 55.68MB/s for the write test and 48.56MB/s for the read test. Our minimum transfer rates were 51.96MB/s and 41.02MB/s in the write and read tests respectively. CPU usage is a modest 2% in the write test and 1% in the read test."


Listen, I know what I posted. It's a MOTHERBOARD review and not intended as a FULL NIC REVIEW. If you actually read the review..

"this controller is capable of performing much faster than our tests here would indicate. This is due to a lack of 2.5GbE capable controllers in our test environment."

What exactly was the point of your windows throughput test, because it matches the review results and you appear irritated by that for some reason.
 
Last edited:
Listen, I know what I posted. It's a MOTHERBOARD review and not intended as a FULL NIC REVIEW. If you actually read the review..

"this controller is capable of performing much faster than our tests here would indicate. This is due to a lack of 2.5GbE capable controllers in our test environment."
No, you too don't understand networking.

1Gb LAN network is capable of 112MB per second real world transfer speeds or a little better. UN-optimized. He didn't get that. Something's wrong with that test result if it's ANYTHING but right around that transfer speed on a single large file.

It's NOT because he doesn't have a 2.5Gb LAN network. If he had 2.5Gb LAN network then he should see the 1Gb Intel card run at about this standard 112MB/s and the 2.5Gb Realtek card run significantly faster. The 2.5Gb card was negotiated down to a 1Gb speed, but should still get a full 1Gb NIC speed on a 1Gb LAN. It didn't. Neither Intel or Realtek card ran at rated speed, and if the author knew anything about networking throughput he would know - without question that something was wrong with his testing - and he would never have published those numbers. They were both running about half speed.. Something was wrong, clearly. His test on that particular matter is clearly bunk - thus those results should be tossed. I didn't say his whole review was bunk - I said those test results are bunk. I don't know who that author is. It doesn't matter. On that test he is wrong.

Both those NICs should have gotten 112MB/s on a 800MB zip file with similar CPU usage on a 1Gb capable network segment. Period. End of story.
 
Last edited:
TheFpsReview - LAN

"The ASUS ROG Crosshair VIII Hero WiFi integrates two adapters. The first is the standard Intel i211-AT. This is a ubiquitous adapter that can be found on a variety of motherboards. It is the option for Intel and non-Intel based motherboards alike. It is the gold standard for reliability and compatibility. The second is the Realtek RTL8125-CG 2.5G LAN adapter. This is a controller I don’t have a lot of experience with. Historically, I haven’t been a huge fan of Realtek adapters, but again, this isn’t a common controller, so I don’t have much to say about it outside of actual testing.

RTL-8125-CG @ 1Gb
In our first network controller showed an average transfer rate of 69.49MB/s in the write test and 45.46MB/s in the read test. The maximum transfer rate was 78.26MB/s in the write test and 50.90MB/s in the read test. Our minimum transfer rates were 66.40MB/s in the write test and 40.83MB/s in the read test. CPU usage was 2% in the write test and 1% in the read test.

i211-AT @ 1Gb
In the write test, our average transfer rate was 54.29MB/s. In the read test, we saw a result of 47.83MB/s. The maximum transfer rates were on the low side at 55.68MB/s for the write test and 48.56MB/s for the read test. Our minimum transfer rates were 51.96MB/s and 41.02MB/s in the write and read tests respectively. CPU usage is a modest 2% in the write test and 1% in the read test."\

From Intel site:
Product Brief
IPv4 and IPv6 checksum offload, TCP/UDP checksum offload, extended Tx descriptors for more offload capabilities, up to 256 KB TCP segmentation (TSO v2), header splitting, 40 KB packet buffer size, and 9.5 KB Jumbo Frame support.

I could get faster speeds with a 12$ NIC. How can this reviewer be getting such slow speeds
 
No, you too don't understand networking.

1Gb LAN network is capable of 112MB per second real world transfer speeds or a little better. UN-optimized. He didn't get that. Something's wrong with that test result if it's ANYTHING but right around that transfer speed on a single large file.

It's NOT because he doesn't have a 2.5Gb LAN network. If he had 2.5Gb LAN network then he should see the 1Gb Intel card run at about this standard 112MB/s and the 2.5Gb Realtek card run significantly faster. The 2.5Gb card was negotiated down to a 1Gb speed, but should still get a full 1Gb NIC speed on a 1Gb LAN. It didn't. Neither Intel or Realtek card ran at rated speed, and if the author knew anything about networking throughput he would know - without question that something was wrong with his testing - and he would never have published those numbers. They were both running about half speed.. Something was wrong, clearly. His test on that particular matter is clearly bunk - thus those results should be tossed. I didn't say his whole review was bunk - I said those test results are bunk. I don't know who that author is. It doesn't matter. On that test he is wrong.

Both those NICs should have gotten 112MB/s on a 800MB zip file with similar CPU usage. Period. End of story.

No, I understand networking. Don't be condescending. We don't now the details of the test(s) and exactly how they were performed. I know the intel nic matches your test @ 1G. Looks like the 2.5G was running full speed as well. So what. It was just a full speed throughput test.
 
Okay - I apologize for being heavy handed. It was unnecessary. I see this thread as unnecessary at the premise of performance differences on modern day consumer NICs, and to see data that on it's face looks blatantly wrong being used to qualify a tech preference decision riled my feathers. My friend and I were just talking recently about the YouTube expert generation who because they have a website, or a youtube channel, publish themselves as experts and pretend they navigate fully experienced in all the subject matter they chance to cover. I'm an audio geek, and it's one of those hobbies that the more you know, the more you realize you don't know. Same with Information Security, my career. I see these popular audio review guys pop onto YouTube or review sites and make ridiculously novice level reviews and claim something as fact, vs opinion, without the experience levels and product comparison experience to even back up their claim as opinion. I'm only 40, but I'm finding I'm becoming a grumpy old man early - especially on the internet. I admit this very thread and conversation touched on that nerve.

Aside --

I'm trying to figure out what he did to get those numbers. I downloaded that software he used. The "lite" version (non-paid) does not allow me to pick a file to test with, and doesn't look the same as what he has in his screen prints.. When I chose the size of the testing - it gives no mention of it being a single contiguous file. That was my assertion in my replies, because that is what the review said he used for testing.

If he wasn't using a single file of large size, and was using 800MB's worth of multiple/various files then all bets are off on maintaining a steady speed. However, his review states he used a 800MB zipped file for the tests.

I'm not going to buy this software to test with but I'd be half-interested to see it's performance metrics on my network to compare with what that reviewer sees.
 
Last edited:
  • Like
Reactions: Mega6
like this
I don't think it matters at all for gaming. For data transfer, if you're just running a single Ethernet cable to a gigabit switch, then I would choose the Intel NIC. If you are actually connecting to a 2.5Gb capable switch, I'd at least give the realtek a shot.

It's also worth noting that SMB 3.0 automatically combines separate network connections when doing file transfers without having to do NIC teaming or anything like that. I run 3 Intel Gigabit adapters in my file-server (Windows Server 2019) and 3 Gigabit adapters in my main PC (Windows 10) and I can see file transfer rates nearing 3Gbps, with speeds almost always limited by the drive and not the network at that point. So there could be benefit to simply using both network connections in the OP's case.
 
It's also worth noting that SMB 3.0 automatically combines separate network connections when doing file transfers without having to do NIC teaming or anything like that. I run 3 Intel Gigabit adapters in my file-server (Windows Server 2019) and 3 Gigabit adapters in my main PC (Windows 10) and I can see file transfer rates nearing 3Gbps, with speeds almost always limited by the drive and not the network at that point. So there could be benefit to simply using both network connections in the OP's case.
I didn't know that! Pretty cool!
 
Should be a third option for the Aquantia 10G

They're cheap enough to add as an add-on card and they do come built-in on several of the higher end boards...
 
Absolutely junk test.

throw it out.

He didn't use a 2.5gbps controller and both NICs at 1 Gigabit LAN should absolutely max out at a little over 112MB/s on a typical network both upload and download with real world scenarios if things were configured even with default settings. Who knows what was wrong in his tests, but something was desperately wrong.

That reviewer clearly knows not one thing about networking.
You could ask Dan_D for a comment on the specifics.
 
The argument is not about saturating the I/O of GbE. This has been long capable, since much of the interfaces and hardware can push easily 100's of MB/s and even 10's of GB/s now. But knowing how to utilize multi-core performance with Receive Side Scaling and mutli-transit queues, virtualization queues, non-bridged logics, MSI, and offloading for the many Gb interfaces, it's entirely a different argument.

Intel will always be better than Realtek. Realtek will always be cheaper, some Intel logics are cheaper and limit your over performance of your LAN for extensive usage, but not to sub early Y2k days. With WAN'ing the interactions are entirely different, and much of what you all argue is not even relevant. You start talking about maxing out your muti-transit queues with PPS and multi-core performance, with say, FreeBSD/PFsense (https://bsdrp.net/ and https://bsdrp.net/documentation/technical_docs/performance) then you will be on the correct path in which the core logics used are very effective, or not.

And to note, most do not tune the settings of the NIC.....and another note as to a preference. It is well documented how drivers from Realtek were broken, and had many performance issues with the *nix community. It is the very fact that Intel's drivers worked, and on many platforms. Intel is not the standard as much as it used to be, but still is with SOHO.
 
Last edited by a moderator:
PsPing is a simple tool to test latency. Comparing the Intel vs Realtek adapters on my x570 board with a quick test pinging the router didn't seem to differ between the two. They both fluctuated between 0.50 to 1.75ish ms. Latest drivers for both. Disabling all the power saving and interrupt moderation settings didn't make a difference either. For what it's worth, the Intel driver has more bloat. It installs an NT service and network adapter diagnostics driver while Realtek has just the normal driver installed.
 
does Intel vs Realtek matter for the average home user who only uses the port for internet usage?...Intel seems to be the better LAN but I'm guessing it only matters if you're transferring a lot of files or running some sort of server or home network?

I don't run any servers or home network and I've narrowed my motherboard choices for my new build down to 2- one uses the Intel GbE LAN chip and the other Realtek 8125B...should the LAN port play a factor in my decision?
 
Last edited:
does Intel vs Realtek matter for the average home user who only uses the port for internet usage?...Intel seems to be the better LAN but I'm guessing it only matters if you're transferring a lot of files or running some sort of server or home network?

I don't run any servers or home network and I've narrowed my motherboard choices for my new build down to 2- one uses the Intel GbE LAN chip and the other Realtek 8125B...should the LAN port play a factor in my decision?

personal preference really.. is there a performance difference? probably but i don't care enough to check, i just had enough bad experiences with realtek's lan chips years ago that i just refuse to ever use them again even though i know that it doesn't matter anymore these days for general lan usage.
 
I've had issues with Intel and realtek drivers, gone are the days of CPU utilization and latency differences it appears for stock settings, so really both are same in my books. Use whatever.
 
does Intel vs Realtek matter for the average home user who only uses the port for internet usage?...Intel seems to be the better LAN but I'm guessing it only matters if you're transferring a lot of files or running some sort of server or home network?

I don't run any servers or home network and I've narrowed my motherboard choices for my new build down to 2- one uses the Intel GbE LAN chip and the other Realtek 8125B...should the LAN port play a factor in my decision?

Nope. Realtek is good enough; at the same speed Intel is probably better, but it's going to be hard to measure it. It would be different if you were running a server trying to push 40Gbps+, but otherwise not very significant. If you had a 2.5Gbps network (or were planning on it), might as well get a Realtek 2.5g card over a Intel 1G, otherwise, pick boards based on something else.
 
Nope. Realtek is good enough; at the same speed Intel is probably better, but it's going to be hard to measure it. It would be different if you were running a server trying to push 40Gbps+, but otherwise not very significant. If you had a 2.5Gbps network (or were planning on it), might as well get a Realtek 2.5g card over a Intel 1G, otherwise, pick boards based on something else.

is it worth it to get a separate Intel networking card?
 
Right, this is a pointless discussion; processors are so powerful now that you won't notice any performance difference on what is now such a small system load.

The only real reason he might get the 2.5gbs Realtek is if he's planning on upgrading his network switch in the near future (it will work on old Cat5e).

2.5gbps is a lot faster than 1Gbps (if you have a local NAS that's pushing 120MB/s or more).
This is my issue, I can sustain about 950MB/s with my RAID 1+0... but I only get ~120MB/s transfers :(. It's such a hassle to upgrade everything though (and I don't do to large of files that often) that I just live with it. I probably won't be buying anything new with less than 2.5G (preferably 10G if it's not to costly) for this reason though. I would prefer not to have to have an add on card for networking (especially in my 2 ITX desktops, but even my mATX desktops don't really have the space!)

Edit: Hassle as in costs, probably new cabling, 2 switches, and 3 of 5 of my PC's I don't have a free pcie slot.
 
is it worth it to get a separate Intel networking card?
For your use case? Only if you get the board with the Realtek NIC and you run into some issue. (but if you ran into an issue with onboard intel, I'd tell you to get a Realtek add-in card).
 
The real bargain for fast file transfers IMO comes from all of the cheap Intel dual-port and quad-port gigabit adapters on places like eBay. SMB 3.0 and above supports SMB multichannel. It will utilize multiple network adapters automatically to increase transfer speed. In my case, I have 3x Gigabit in my main computer and 3x Gigabit in my fileserver (Intel Dual-Port + Intel motherboard gigabit in both cases), all hooked to the same switch. Using those 3x Gigabit connections, I get speeds above 2.5GbE when transferring files. All using nothing more than a cheap 16-port gigabit switch and cheap used Intel dual-port gigabit adapters. It's very versatile also. For example, you could have your fileserver hooked up using a single 10GbE, and then a client using a cheap Intel quad-port gigabit card and be able to get 4x GbE speeds when transferring files. That makes those cheap switches with ~24 GbE ports but only 2 10GbE ports a lot more handy.
 
Back
Top