Intel NIC vs RealTek NIC - Performance Testing

[joeschmuck]

All my testing was to create a baseline and then results after a change. For test 5, 6, 7 and 8, as I mentioned I used Windows Copy. I opened an explorer window for both the source and destination and dragged the file over.

As for the Intel NAS Speed Test kit, that is true, the results probably could be skewed but I'm measuring the difference between the baseline and the change on the same platform. I'm not coming up with number that I could compare again say your computer because it may skew the results, however I didn't get the feeling that data was cached as I did delete the test data between each test and I rebooted the machine which should have cleared any cached data. Again, I wasn't trying to state how fast or slow my throughput was compared to everyone else, it's a test of before and after hardware changes to justify if the Intel NIC was a valuable gain in my system. It wasn't and I pulled them out of my computers to reduce the power draw and heat created. Those guys did generate more heat than I would have thought.

So if you're going to do some comparative testing, I think I created a reasonable template as a guide and you can use whatever you feel would meet your objectives.

 

[Yatti420]

Awesome thanks for clarifying.. I started doing some speed tests overnight to get some basic info.. Using your testing as a template.. http://forums.freenas.org/threads/freenas-9-1-rc2-initial-speed-testing-slow.13999/

 

[cyberjock]

I'm going to be playing with a small Atom system a friend wants to make into a backup FreeNAS server to store his most critical data in a few days. It has onboard Intel NIC(dual if I remember correctly) and I do happen to have a Realtek PCIe NIC. I was thinking about doing some speed comparisons between them just to see how they do in a low performance CPU environment.

As a forum we've kind of made some assumptions/realizations with Realteks:

1. Some people have weird networking problems, and going Intel often solved them. This could be a customized Realtek NIC not being compatible with the standard Realtek driver.
2. Some people have Realtek NICs that just won't work at all, despite appearing to work based on chipset model.
3. It has been assumed(at least I think assumed and not validated) that Realteks perform poorly in low performing CPU environments.
4. In high performing CPU environments many(but not all) won't seem to go much faster than about 60MB/sec even if you want to get out and push. Other with high performance CPUs are capable of getting 100MB/sec.

The Realtek vs. Intel debate will never end because there's lots of "what ifs" and exceptions. Realtek has dozens of models, some mislabled, some not compatible.

It kind of comes down to some simple rules(especially for new users).

1. If you have Realtek on your hardware and you're happy with the performance and reliability, there's no reason to upgrade.
2. If you are new to FreeNAS and buying purpose-built hardware for a FreeNAS machine and want to ensure compatibility, Intel is a sure bet. Realtek is hit and miss. It's your gamble to take if you wish.
3. If you absolutely want to be sure that you will get the best speeds possible, going Intel NICs are the way to go.

Generally I recommend Intel for newbies because if it doesn't work or doesn't work "quite right" newbies can spend countless hours fighting their machines trying to see if they have a setting wrong. Often when the Realteks have issues its not immediately apparent that the NIC is the problem. It adds a lot of complexity to something that is already probably confusing the heck out of them already. Going from Windows to FreeBSD isn't a walk in the park, and having stuff not work right off the bat is not a good way to start learning FreeBSD. I learn a lot more from seeing working systems than seeing broken ones.

 

[joeschmuck]

I cannot argue with what you're saying. I really think in my situation one other issue is at play, I have Realtek NICs on both machines and they both have high performing CPUs. I did have a slight improvement using Intel NICs but not enough to justify the cost of the new NICs. I would be interested in seeing the results of an Atom system doing similar benchmarks. I don't think you need to do everything I did, I did cover all the bases but to get a general idea I think a few tests would be good. I'm seriously thinking about replacing my FreeNAS system with a new one with ECC RAM. I'm going to hate flipping for the bill! But it is the right way to go. We should be pushing everyone who is planning to build a new system from new parts to buy ECC.

 

[joeschmuck]

I have done some more testing on a different MB, the ASUS M5A78L-M/USB3 (w/AMD FX-4300) both with the internal RealTek NIC and the Intel NIC. Funny, the results are in essence the same, no real difference between the two sets of tests. My results are slower than my original tests only because I'm now using a RAIDZ2 setup, but to be specific just to document it, I originally used four WD Red 2TB drives in a RAIDZ1 configuration and for this testing I an using six WD Red 2TB drives in a RAIDZ2 configuration. Funny that I thought I'd be gaining some read speed, not only an extra level of safety.

Here are the results from the Intel NAS Speed Tests...

Code:

Test                RealTek  Intel NIC
HD Video Playback     62        65
2x HD Playback        82        82
4x HD Playback        95        98
HD Video Record      126       142
HD Playback & Record  71        66
Content Creation       9        10
Office Productivity   47        47
File copy to NAS      69        78
File copy from NAS    75        69
Dir copy to NAS        5        5
Dir copy from NAS     22        21
Photo Album           12        10
(Note: All values are in MB/Sec)

I'm sure if I ran the tests again there would be subtle changes but in general the throughput is virtually identical unless you really just wanted to split hairs about it.

So I still say that a RealTek NIC is just as good as an Intel NIC provided you have enough CPU horsepower.

 

[cyberjock]

Interesting, especially since so many people have commented on how much switching to an Intel has improved performance.

Of course, the other reason to go to Intel is still probably valid. Many people just have random problems with Realteks and as soon as they switch to an Intel their weird unexplained networking problems disappear.

I know in Windows switching from Realtek to Intel was the only way I could find to get above about 60MB/sec, with an Intel card taking me to saturation speeds on Gb. All of my desktops and servers were state of the art circa 2009 when I went to all Intel.

 

[joeschmuck]

Fortunately I haven't experienced any of the weirdness the RealTek NICs have to offer, maybe it's the location I'm at, the dark side of the moon gets a bit of cosmic radiation so it could be boosting performance, like a supercharger. I do like the Intel NICs just because they are not bound by the CPU and have decided to leave the Intel NIC installed in my NAS.

 

[Yatti420]

I do like the Intel NICs just because they are not bound by the CPU and have decided to leave the Intel NIC installed in my NAS.

I noticed this but figured it was standard across any half decent network card add-on card.. I rarely use them anymore.. As this is a good idea can you only use an Intel add-on card? How about a standard gigabit card like a realtek xxx.. ? Are PCIe ethernet tied to CPU aswell - just via PCIe lane (more bandwidth)? Is there a big difference between a pci and pcie gigabit card? If no other components are installed in the computer? Just hdds/add on gigabit card.. Can you point me to this intel feature? Thanks,

 

[joeschmuck]

You have a lot of questions but I might be able to answer one or two... When it comes to CPU utilization, it depends on the LAN chip as Intel chips typically perform the processing to convert the ethernet into data where as many others (realtek included) do not process the data and rely on a CPU to do the work. So this is why people like Intel LAN adapters. As for PCIe attatched to the CPU, you bet ya they are. But I wouldn't worry about this because the data throughput is so slow so in my opinion I doubt there is any significant drawback unless you have a very slow CPU, like an ATOM. Again, my opinion. My best advice is if you are using a low speed platform, use and Intel NIC if you can but if you don't have an Intel NIC, test the one you have for throughput. As in my testing above, with a RAIDZ1 I could saturate my connection, with a RAIDZ2 it's just a bit different just because I have slow drives and it's a different configuration but I can saturate it.

 

[craigyb]

I had 3 NIC's in a lag group, 2x intel and the onboard realtek. it didn't perform as expected, so I dropped the realtek from the lag group and just run the two intel nics and everything is sweet.

 

[ssgoku129]

Thanks a lot for this post man, this is extremely helpful for what I was about to do!!!!!!

 

[ex0r]

Hello!

I bought a intel network card and tested it against the realtek. The difference was huge. I created a PDF with some details. You might want to check it out. Im interessted in your opinions :)

ExOr

 

[jgreco]

Which just goes to show, people have extremely variable experiences with the Realteks. Not so much, usually, with the Intels.

 

[joeschmuck]

Your tests were fine basic tests but to really test your throughput you should run a LAN test suite. It's not going to change your overall results which are you get a better throughput with the Intel NIC, but the question as you pointed out would be if it was worth it. And right now for me it's a stability concern more than the speed issue.