Skip to main content

Gigabit Ethernet: On-Board Chips Reviewed

Test System

Processor
Socket 478Intel Pentium 4 Northwood, 3.2 GHz, 512 kB Cache, FSB800
Socket 604Dual Intel Pentium 4 Xeon, 2.8 GHz, 512 kB Cache, FSB533
System Memory
DDR-SDRAM2x 512 MB PC3700OCZ Gold Edition Rev. 2 (2.5-3-3-7)
Network/Motherboard
3COM / Marvell 940Asus P4C800, Rev. 1.03Intel 875P chipset & ICH5-R
Broadcom BCM5705Epox EP-4PCA3+, Rev. 1.0Intel 875P chipset & ICH5-R
Intel 82547EI (CSA)DFI LanParty PRO875, Rev. AIntel 875P chipset & ICH5-R
3COM 3C996B-TBroadcom BCM5701Asus PP-DLW, Rev. 1.03Intel E7505 chipset
Realtek RTL8100SGigabyte GA-8S655TX Ultra, Rev. 1.0SiS655TX & SiS964
Other Hardware
Graphics cardnVIDIA GeForce FX 5900, 128 MB
ControllerClient systems: ICH5-R or SiS964 for RAID 0Host system: Raidcore RC4000 für RAID 0
Hard drives (client systems)System drive: Western Digital WD800JBTest drive: RAID-0 array consisting of2x Hitachi Deskstar 7K250, HDS722580VLSA80, 80 GB,7,200 rpm, 8 MB Cache
Hard drives (server system)System drive: Western Digital WD1000BBTest drive: RAID-0 array consisting of4x Western Digital WD360 Raptor, 36 GB,10,000 rpm, 8 MB Cache
Software
Intel ChipsetIntel Chipset Installation Utility 5.1.1.1002Intel Application Accelerator RAID Edition Ver. 3.53
Network drivers3COM / Marvel 940Broadcom BCM5705 & BCM5701: Ver. 6.34.0.0Intel Pro/1000 CT&MT (82547EI & 82541GI): Ver. 7.3.13.0Realtek: Ver. 5.611.1231.2003
DirectX9.0b
OSWindows XP Professional Build 2600 Service Pack 1

Tests: A Detailed Look

Synthetic Benchmarks
TCP Transfer performanceNetIO Ver. 1.23 from Kai Uwe Rommel
TCP Transfer performanceNetperf Ver. 2.1
TCP Transfer performanceNetIQ Chariot 5.0
Application Benchmarks
Real Life Transfer Performancexxcopy 2.85.3 von Pixelab, Inc. idf2002_master.avi - 1.17 GB
Real Life Windows File Copy PerformanceWindows Explorer Drag&Drop Test idf2002_master.avi - 1.17 GB

During our benchmarks under real-life conditions, we copied the said file five times, doing so back to back. All tests were conducted five times in order to compensate for data dispersion while determining the mean.

To avoid a slowdown of the network adapter through an unnecessary bottleneck, we added a RAID 0 array (consisting of two current hard drives) to each system for this test. As a result, transfer rates von 110 MB/s were, therefore, no problem - that's more than most adapters manage as a net output.