Gigabit NICs

[CBuxton]

I'm having a hard time figuring out which Intel Gigabit NIC would be the fastest. There are several available:
Desktop:
Pro/1000 MT - Intel® 82540EM chip
Pro/1000 T - Intel® 82544GC chip
Server:
Pro/1000 MT - Intel® 82545EM chip
Pro/1000 XT - Intel® 82544EI chip

I found a couple of tests on etestinglabs.com that show the XT as the "latest and greatest" beating 3Com and Broadcom, but then there's a Broadcom test that beats the Intel and if you go to the 3Com test it also wins! Currently I have a couple of the Desktop Pro/1000MT cards, and they are very slow (only about 120Mbps). The machines are 1.7Ghz so it shouldn't be a cpu limitation. I also would consider a NIC with a Broadcom chipset (Compaq is using the 5701 on the NC7770), but again I don't know how to figure out which is the fastest of their chipsets (5701,5702,5703 (Tested on ETestinglabs), or 5705). Can anybody shed any light on the situation?? Thanks

 

[SaigonK]

A funny thing is that the new DELL servers have both! =) (The intel and the broadcomm)
I like the Broadcomm, it runs very nicely and I have yet to see one die on me. **knock on wood!**
Do you have a gigabit backbone you will be connecting to to use this LAN card? If not why bother with one....it wont do you any good.

 

[ScottMac]

The last test I saw (University of Nebraska, I think), the SysKonnect board smoked 'em all, and was more configurable (for best performance for a given environment, MTU, etc).

Check 'em out.

FWIW

Scott

 

[CBuxton]

I should have given more info! Currently the nics are in my home machines (each is a 1.7Ghz machine) and I have been using netcps to test. It maxes out at 20MB/sec which works out to be somewhere around 160Mbps. File transfers using windows give about the same results. I've run them through my 3Com Baseline gigabit switch as well as a cross-over cable. All get the same results, so I'm pretty sure it's a NIC issue. I'm testing these at home before I put some in at work. The servers have raid, dual processors, etc, but I don't want to take them offline for testing purposes, so I'm trying the nics at home. The network runs fiber to multiple buildings all coming back to a switch @ gigabit speeds. The switch has 1000BaseT ports. Anyway, I still feel I should be getting more throughput. Thanks. I'll check out the Syskonnects

 

[n0cmonkey]

Originally posted by: CBuxton
I should have given more info! Currently the nics are in my home machines (each is a 1.7Ghz machine) and I have been using netcps to test. It maxes out at 20MB/sec which works out to be somewhere around 160Mbps. File transfers using windows give about the same results. I've run them through my 3Com Baseline gigabit switch as well as a cross-over cable. All get the same results, so I'm pretty sure it's a NIC issue. I'm testing these at home before I put some in at work. The servers have raid, dual processors, etc, but I don't want to take them offline for testing purposes, so I'm trying the nics at home. The network runs fiber to multiple buildings all coming back to a switch @ gigabit speeds. The switch has 1000BaseT ports. Anyway, I still feel I should be getting more throughput. Thanks. I'll check out the Syskonnects

Do your home machines have decent I/O or are you using some old IDE drives and no raid? 20MB/s sounds about right for a lot of the IDE drives out there as far as sustained transfers go, but I havent looked into speeds recently.

 

[Rilescat]

Originally posted by: CBuxton
I should have given more info! Currently the nics are in my home machines (each is a 1.7Ghz machine) and I have been using netcps to test. It maxes out at 20MB/sec which works out to be somewhere around 160Mbps. File transfers using windows give about the same results. I've run them through my 3Com Baseline gigabit switch as well as a cross-over cable. All get the same results, so I'm pretty sure it's a NIC issue. I'm testing these at home before I put some in at work. The servers have raid, dual processors, etc, but I don't want to take them offline for testing purposes, so I'm trying the nics at home. The network runs fiber to multiple buildings all coming back to a switch @ gigabit speeds. The switch has 1000BaseT ports. Anyway, I still feel I should be getting more throughput. Thanks. I'll check out the Syskonnects

See, that is odd. We only run 10/100 on most clients, and those are almost the same speeds the clients get. We are attached to a very fast backplane, on a Cisco Catalyst 2948G (1000) but that won't help clients that top at 100.

 

[CBuxton]

One machine is a RAID0, the other is a Seagate X15-36LP 15k scsi drive, so throughput should be pretty good. I'm also using throuput testing programs as opposed to copying files and timing it. I still think it's a nic issue. Thanks for the comments.

 

[lobadobadingdong]

Anyone think of the pci bandwith issues? Are you using pci 33MHz 32bit, or speedy pci 64bit 66MHz, you could be flooding your available bandwith on your pci buss, this of course would limit your transfer speeds no matter how fast your disk i/o performed. at 33MHz with 32bit your maximum bandwidth is such that you would never see anything near gigabit transfers, this is why gigabite is pretty much only used for hi-bandwidth backbones and hasn't really migrated to desktops, there is no need(marginal performance gains that don't justify the cost) for gigabit use on standard workstations. 10/100 3com 3c905-c is still my choice for anything not needing direct backbone connections.

 

[owensdj]

I agree with lobadobadingdong. Unless you know why you need Gigabit Ethernet, you probably don't.

 

[Hanpan]

Just to chime in. I have some 1000mt's and sawa bit more performance. About 200-250 mbit, however in my case this is close to the hdd limit. Remeber with some programs that test your network that if transfers are very small (say 1 meg or less) you might not get an accurate reading. Using the network testing tool JackMDS suggested I got about 400 mbit or so which is still not close to 1000mbit but certainly better than you are seeing. Another puzzeling thing is according to intel the pro1000t and mt are the same card just the mt has the 3 chips of the t integrated into one but otherwise performance is the same. THe pro 1000t has been shown to perform very well so I see no reason for the MT's to performe this poorely.