Best Gigabit NiC for Server (Win XP x64)

S

Stereodude

2[H]4U
Joined
Oct 20, 2000
Messages
3,285
I've read on another forum that add in NiC's offer much better performance than the Realtek stuff soldering on most motherboards. They pointed to this article. I have a server in my basement running Windows XP x64 SP2. It has two Realtek 8111C onboard PCIe x1 NiC's. I'm only using one of them.

Will I really see an improvement by going to something like an Intel Pro/1000 PT Server Adapter or are the drivers for Windows much more optimized than Linux for the Realtek chips and any difference will be minimal?
 
If you're looking for best performance you can't go wrong with the intel NIC's.
 
You're using a desktop operating system, don't sweat it much, it cannot even fully utilize a decent desktop grade NIC. Just get a decent Intel desktop NIC....total waste of money investing a few hundred bucks in a high end server grade NIC that a desktop OS cannot even utilize 10% of.
 
good point. You can save some money by picking up the desktop version, http://www.intel.com/products/desktop/adapters/gigabit-ct/gigabit-ct-overview.htm

It's not available on newegg though, so you'd have to go with their GT line but that's PCI only. I disagree with the desktop operating system not being able to fully utilize a NIC comment, I've been able to push gigabit to it's limits with mere "desktop operating systems" ;) On a different note, some of the worse onboard NICs bog down quite a bit when stressed.
 
YeOldeStonecat said:
You're using a desktop operating system, don't sweat it much, it cannot even fully utilize a decent desktop grade NIC. Just get a decent Intel desktop NIC....total waste of money investing a few hundred bucks in a high end server grade NIC that a desktop OS cannot even utilize 10% of.
Click to expand...

XP x64 is basically just Server 2003 x64 renamed. They're much more similar than the x86 versions of XP and Server 2003. For example, the recent TCP/IP issue that Microsoft said they wouldn't be patching in XP... They said that they wouldn't patch the x86 or x64 versions of XP, but the download for 2003 x64 is actually named WindowsServer2003.WindowsXP-KB967723-x64-ENU.exe and works great on XP x64.

If any "desktop" OS could take advantage of server grade stuff, I'd say it would be XP x64. However, outside an enterprise environment, I don't think most people will notice a difference between the $30 1000 CT and the $85 1000 PT.
 
InvisiBill said:
XP x64 is basically just Server 2003 x64 renamed. They're much more similar than the x86 versions of XP and Server 2003. For example, the recent TCP/IP issue that Microsoft said they wouldn't be patching in XP... They said that they wouldn't patch the x86 or x64 versions of XP, but the download for 2003 x64 is actually named WindowsServer2003.WindowsXP-KB967723-x64-ENU.exe and works great on XP x64.

If any "desktop" OS could take advantage of server grade stuff, I'd say it would be XP x64. However, outside an enterprise environment, I don't think most people will notice a difference between the $30 1000 CT and the $85 1000 PT.
Click to expand...

The focus on background services, notably for networking services, is substantially different. A desktop OS isn't designed to be efficient with more than 10 concurrent sessions, beyond that..and the performance curve takes a good drop. Server OS's are tuned differently, they maintain efficiency well beyond that...way way beyond that.
 
YeOldeStonecat said:
total waste of money investing a few hundred bucks in a high end server grade NIC that a desktop OS cannot even utilize 10% of.
Click to expand...
Well, a single port Intel Pro/1000 PT is like $45 shipped on ebay, so it's not like it's much more than a "desktop" PT or CT card.
 
If this is in your home, I would think the realtek nic is fine until you start to max it out IMO...

That being said, used intel nics go fairly reasonable off of ebay used.
 
As far as I'm concerned there's very little difference between the server and desktop versions other than price. The biggest difference I can find is that some features are disabled in the driver on some desktop cards (like aggregation). Performance seems to be good with either, and they both support similar CPU offloading features. I just don't buy it as much more than marketing.

And definitely ditch the Realteks for just about anything else.
 
FWIW, the Realtek's also support the same checksum CPU offloading features as the Intel cards.

Maybe I will buy an Intel and benchmark the two but admittedly I am skeptical of there being much difference myself.
 
Stereodude said:
FWIW, the Realtek's also support the same checksum CPU offloading features as the Intel cards.
Click to expand...
They do, but it apparently doesn't do much good, since they still use a lot more CPU time.

Maybe I will buy an Intel and benchmark the two but admittedly I am skeptical of there being much difference myself.
Click to expand...

Most modern machines should get decent throughput out of either NIC, but you'll find significantly increased CPU load from the Realtek solution, as well as CPU load-dependent latency. These effects get much worse when you're dealing with small packets. I just tested it out on my Q9550 box, saturating GigE on the Intel PCIe 1x NIC uses about 6% CPU, the Realtek doing the same uses 12-13%. Throughput is nearly identical at ~935mbps.

Edit: I just re-ran the same thread with full-duplex traffic instead. Both cards again reach similar throughput (~1700mbps total - with Intel leading by 50mbps or so) in my test, however while the Intel stays at around 7-8% CPU, the Realtek peaks at 25%, averaging about 17%.

Edit2: 1 final test. I loaded up Prime95 and started 4 blended threads, then performed the bandwidth test again. Realtek plummeted to about 1275mbps while Intel maintained about 1650mbps.
 
Last edited:
keenan said:
As far as I'm concerned there's very little difference between the server and desktop versions other than price. The biggest difference I can find is that some features are disabled in the driver on some desktop cards (like aggregation). Performance seems to be good with either, and they both support similar CPU offloading features. I just don't buy it as much more than marketing.
Click to expand...

Substantially better hardware and driver support on the server grade versions. I've seen the differences myself even going back years when I used to build lots of public gaming servers, and game servers for big LAN parties. Believe me, desktop NICs...even the mighty 3COM 905 series back then, have a LAN with two exact same servers, one running a 905, and the other server running the server grade 3COM 990svr NIC I used to often use..have a 100 person LAN party and you'll see the performance difference.

Saying its marketing hype is like trying to say a USR Courier V.Everything modem is really the same as a Winmodem.
 
Well, I bought a used Intel Pro/1000 PT Server Adapter off ebay and a new Intel Gigabit CT Desktop Adapter to do some testing with.

I did a little experimenting with iPerf today and I was surprised at how high the CPU usage was with the on-board Realtek 8111C [PCIe] ~48% (server) / ~67% (client) with an E5200 CPU. My other server with an on-board Intel 82547GI [CSA] was on the other side of the iPerf test had this CPU usage ~28% (server) / ~70% (client) with a 3.0gHz P4 w/ HT (Prescott). Bandwidth was >900Mbit/sec each way (sequential not simultaneous).

I wouldn't have expected the P4 to have lower CPU usage than the E5200 in the server role. I'm curious now to see how the two new Intel PCIe cards will compare in the server to the on-board Realtek 8111C.
 
I haven't done any CPU utilization testing yet, but the Intel Pro/1000 PT Server didn't set the world on fire in terms of transfer speeds when compared to the Realtek 8111C.

Realtek 8111C in Server1:

Case1:
Desktop push to Server1 - 66.07MB/sec
Server1 pull from Desktop - 51.72MB/sec

Desktop pull from Server1 - 59.34MB/sec
Server1 push to Desktop - 22.73MB/sec

Case2:
Server2 push to Server1 - 59.14MB/sec
Server1 pull from Server2 - 52.85MB/sec

Server2 pull from Server1 - 47.51MB/sec
Server1 push to Server2 - 14.87MB/sec


Intel Pro/1000 PT Server in Server1:

Case1:
Desktop push to Server1 - 63.11MB/sec
Server1 pull from Desktop - 56.19MB/sec

Desktop pull from Server1 - 56.56MB/sec
Server1 push to Desktop - 22.76MB/sec

Case2:
Server2 push to Server1 - 59.34MB/sec
Server1 pull from Server2 - 53.18MB/sec

Server2 pull from Server1 - 48.32MB/sec
Server1 push to Server2 - 14.74MB/sec

Summary:
Using the Intel... in Case1 two of the four tests slowed down by 3MB/sec, one test speed up by 3MB/sec, and one test stayed the same. There was basically no change in Case2 with Server2 (which also has an Intel in it).
 
Here are my iperf results comparing the RT8111C (in Server1) to the Intel Pro/1000 PT Server (also in Server1) when connecting to Desktop (RT8111B):



The numbers are an average of 3 30 second runs. Here is the iperf command line used:
Desktop: iperf.exe -c Server1 -r -t 30 -w xxk
Server1: iperf.exe -s -w xxk

The Intel Pro/1000 PT card in Server1 gave more symmetrical results than the RT8111C. That translated to improved bandwidth when it was sending as compared to the RT8111C.

Now if we compare that to Windows File Sharing (SMB) performance:


We see very little correlation. :confused:

Next up CPU usage...
 
I don't believe you are doing the tests that matter, or show off the differences. The analogy of a hardware/ASIC based USR Courier modem vs a Winmodem is a good one.

I suspect you'll find more differences with the CPU usage, but you may not be able to tell unless you have multiple simultaneous connections from different computers. You also want to test UDP drops like with video streaming. I've found Intel NIC's to be far superior to anyone else in that respect.

Another test to try.... spike your CPU with Prime95 or something else and give the process a NORMAL priority in Task Manager, not BelowNormal or Low. Then run the tests with an onboard Realtek vs an Intel Server grade NIC. That Realtek will then have to fight for CPU cycles while it packetizes the data.
 
Also SMB performance is a poor benchmark. It's pretty terrible in WinXP no matter what NIC you use.
 
I monitored CPU utilization while running iperf (with 5 parallel threads to max out the connection) with the two NiC's and the Intel had lower CPU usage for all TCP window sizes except for the 8k client number.

RT8111C

Server:
4k - 67%
8k - 66%
16k - 64%
32k - 64%
64k - 64%
128k - 62%

Client:
4k - 45%
8k - 33%
16k - 51%
32k - 51%
64k - 48%
128k - 48%

Intel Pro/1000 PT Server

Server:
4k - 45%
8k - 47%
16k - 49%
32k - 52%
64k - 53%
128k - 52%

Client:
4k - 35%
8k - 34%
16k - 37%
32k - 34%
64k - 35%
128k - 34%
 
Intel Gigabit CT Desktop

Server:
4k - 59%
8k - 59%
16k - 48%
32k - 48%
64k - 55%
128k - 52%

Client:
4k - 37%
8k - 33%
16k - 31%
32k - 36%
64k - 31%
128k - 38%
 
Updated iperf results with the Intel Gigabit CT Desktop adapter in Server1. There's also an additional test with the Intel Gigabit CT Desktop adapter in my Desktop in place of the Realtek 8111B.



I have another Intel Pro/1000 PT Server adapter on the way and am curious to see the results of the Intel Pro/1000 PT Server to Intel Pro/1000 PT Server combination.
 
Last edited:
So the fastest NiC in one scenario is also the slowest in the other three scenarios. Classic...



Results are the average of 3 copies of a 5.5GB file using Windows Explorer.

I guess the moral of the story is buy the Intel PT server (or presumably the PT Desktop) card over the CT Desktop model.
 
FrostBite said:
Thanks for the links to Intel NICs on this thread. My motherboard GIGABYTE GA-G33M-DS2R (http://www.newegg.com/Product/Product.aspx?Item=N82E16813128053) only has 2x PCI-E slots which are both being used so I'm left with just PCI. NewEgg has both the PCI-E (http://www.newegg.com/Product/Product.aspx?Item=N82E16833106010) and PCI version (http://www.newegg.com/Product/Product.aspx?Item=N82E16833106121), will both work just fine or will bus be a problem?
Click to expand...
I haven't tested with an Intel MT (PCI) card yet. I have some of them though. They certainly won't have the throughput of a PCIe x1 card being bus limited. They might have lower CPU usage though.
 
Ok, so I tested my Intel PCI MT cards. I'm not entirely sure how they compare to the Intel PCI GT cards, but they certainly don't light the world on fire with their iperf results.



I also tested their Windows File Sharing performance. They don't light the world on fire there either.

 
You must log in or register to reply here.
Back
Top