Introduction
This article launches Tom’s Networking Guide, a new section for the site. Feel free to give us feedback and suggestions on ways to improve and expand this section. Send your comments to omid@tomshardware.com.
Does The Choice Of Network Interface Cards Really Impact System Performance?
As Internet broadband technology continues to grow, many users now have reasons beyond standard network uses (print and file sharing) to require a Network Interface Card (NIC) as standard equipment on a PC. Most types of DSL and Cable modems require an Ethernet NIC card in order to connect to a user’s PC. While many people might think that examining Ethernet cards performance is about as fun as watching paint dry, we’re going to examine the performance of several NIC cards and see if there is any real difference.
When one asks a PC user which NIC card is installed in their PC and why, the answers are amazing! Some of these answers include: “I used to use brand X, but I had problems with the drivers, so I switched to brand Y.” / “I only use brand X because that is what our company uses.” / “I use brand Z because that brand is what came with the PC or was built into my computer.” / “I use brand X because it is cheap, and there is no difference between brands because Ethernet is a standard.”
As most readers know, Ethernet technology is a standard. Beyond this, is there a difference between NIC cards? Will one NIC card yield better performance results than another? What good is building a top of the line system when the system’s network performance is being crippled by a slow performing NIC? Is there a real feature difference between NIC cards? Can your choice of NIC degrade the performance of your PC? In this article we look at the performance of several 10/100mbit NIC cards to answer these questions.
It is a fact that Ethernet technology is a standard. Early 10 Base-T standards provided 10mbit half duplex performance, and in fact this technology is still in use today on the vast majority of DSL and Cable modems that require an Ethernet connection. With the cost of Ethernet technology falling, many users have moved from simple 10 Base-T technology to 100 Base-TX, which provides 100mbit half or full duplex performance depending on the device that connects with the 100mbit card. In order to achieve 100mbit full duplex performance, the card needs to be attached to a switch that supports the full duplex option. Additionally, 100 Base-TX does require the use of category 5 cabling.
Many home users are now using Ethernet home gateways. These have a hub/switch that provides local network connections of 100mbit full duplex and a 10 Base-T connection for a DSL or Cable modem solution to allow Internet connection sharing. In many cases, this is a good solution, because it provides local network connections of 100mbit full duplex, while still allowing all of the networked home PCs to have access to the Internet.
On the horizon, we are starting to see Gigabit-over-copper Ethernet, which will require category 5e cabling. Yes, Gigabit-over-copper Ethernet is available, but currently it is reserved for high-end server applications. It will be some time before we see Gigabit-over-copper Ethernet in use in non-corporate environments. However, if you want to be prepared for the future, which will likely be Gigabit-over-Copper, it is worth it to spend the few extra cents per foot for category 5e cabling rather than slightly cheaper category 5.
Does The Choice Of Network Interface Cards Really Impact System Performance?, Continued
There are two different types of NIC cards – Business/Enterprise and Home/SOHO (Small Office/Home Office) cards. The main difference is that Business/Enterprise cards often support management features like WOL, DMI and PXE (see below), whereas Home/SOHO cards tend to be stripped down versions of the Business/Enterprise level cards. Many of the management features will most likely never be used by Home/SOHO users and are therefore left unsupported to reduce the cost of the cards made for this market. If the NIC card connects the PC to a large network such features can on the other hand become very convenient.
Wake On LAN (WOL) is a technology that enables you to remotely manage the clients in your network even if they have been turned off. With the need to provide consistent and timely updates, such a feature brings new levels of control, simplicity and savings to LAN-based networks.
The Wired for Management (WfM) specification is an Intel-led, industry-supported effort to make Intel Architecture-based systems universally manageable and universally managed, without sacrificing agility or performance. Through the WfM Initiative, Intel has worked with others in the industry to develop guidelines for a new generation of platforms that can be centrally managed over networks to reduce Total Cost of Ownership (TCO). Systems based on these guidelines provide key technologies that, combined with management software applications, deliver capabilities that enable “down-the-wire management” and benefits in five critical areas of managing and controlling the computing environment: Asset Management, Preboot Execution Environment (PXE), Off-Hours Maintenance (Power Savings), System Diagnosis and Repair, and Investment Protection.
Pre-boot Execution Environment (PXE – pronounced PiXiE) is one of the components of Intel’s Wired For Management (WfM) specification. The PXE model provides computers the ability to load and execute a Network Bootstrap Program (NBP) from a server on the network prior to booting the OS on the local hard drive. It therefore eliminates the need to visit machines and manually boot them with floppy disks.
Desktop Management Interface (DMI) is an architecture that includes a service layer, a management information format database, a management interface, and a component interface. The DMI Service Layer acts as an information broker between the NIC card and the management applications. The MIF database defines the standard manageable attributes of PC and server products.
Does The Choice Of Network Interface Cards Really Impact System Performance?, Continued
Another important feature that most of the Business/Enterprise NICs provide is design stability. Most Business/Enterprise NICs have a longer “no change” life cycle than the Home/SOHO NICs. Business/Enterprise NIC manufacturers might assure you that they will have no design changes to the card that would require a driver change for 12 to 24 months. Home/SOHO NIC card designs are on the other hand price driven, and a manufacturer of Home/SOHO NICs might not make any design changes during the life of the card. More often than not changes are made to these NICs due to pricing changes in the MAC and PHY chips that are required on every NIC card. Deal of the day pricing, in some cases, causes many design changes that require a driver change. Of course, this means an image change if you are using these cards at the Business/Enterprise level. If you are going to use or replace NIC cards at the Business/Enterprise level, you should consider paying more for the Business/Enterprise NICs as you will have less revision headaches down the road. Some manufactures do use a unified driver technology which helps overcome some of these issues, in many, but not all cases.
Since all of the NICs we are evaluating use the PCI bus, the performance of these NICs is, to some degree, dictated by the performance of the PCI bus. The bus performance needs to be measured, not only in total throughput, but also in peak CPU utilization during the throughput benchmark. A NIC card that hogs the CPU and the PCI bus capacity while in heavy use can lead to degraded PC system performance. For example, even though the performance of many network games is capped to help with latency issues, a NIC card that requires too many CPU cycles will slow the game as well as hamper the overall system performance.
This is not the only problem to consider. Differences in the IP stack among different Windows based Operating Systems can, in some cases, restrict performance. I consider Ethernet and NIC technology to be mature, thus the IP stack performance differences have not persisted due to issues common to new technology. Performance of some NIC cards did differ between operating systems. One card in particular had average scores under our Windows 98 and DOS based NDIS driver tests, but had better scores under Windows 2000. Some of these differences can be contributed to driver optimizations between the operating systems, but many feel that this is due in part to a more robust IP stack included with Windows 2000.
NIC card manufacturers use various techniques to manage CPU and PCI bus utilization. Most of these involve a type of processor, or core logic chipset, on the NIC to help the NIC control traffic, as well as to allow the NIC to optimize data throughput by performing the read in/transmit out data transfers simultaneously. These companies often use different names or branding for this technology process. Additionally, optimized driver performance will impact the performance of any device, and NIC cards are no exception.
Due to the amount of various optimization variables that can be changed on each card, we didn’t attempt to optimize each card. Each card was tested in the default install mode. This it seemed benefited some cards, while hindered others. Based on experience and testing, tweaking the settings on the card where possible can increase the card’s performance by a few percent. Of course, some companies used a more conservative approach when choosing default settings of the NIC card. I think it is important to note that none of the companies go into much detail on what effect changing these options will have.
In the case of Windows itself, some performance can be gained by using a registry editor to change the TCP Receive Window size. Although, other myths exist about other changes that you can make, the only thing that seems to really help is changing this setting. With a connection of normal latency, that I consider to be less than 100ms, a size of 32K will yield some performance increase.
We will now look at several NIC cards, the feature set of each NIC card, and conclude with the evaluation of the tested NIC cards.
3Com 3C905C-TX
If one company can be labeled the “Grandfather” of NIC technology, it is 3Com. The 3C905C-TX is the latest generation of the famous 3Com 3C905 NIC card. Each generation of the 3C905 series has brought us more features and somewhat lower prices. This card is a Business/Enterprise level NIC card with one of the most complete feature sets currently available. Due to its extensive features, this card has earned its reputation for reliability and functionality, and has thus become one of the most commonly chosen NIC cards in IT Departments worldwide.
The 3C905C-TX has the following features:
- Media: 10Base-T/100Base-TX
- Connector: RJ-45
- Bus: 32-bit PCI
- IEEE Standards Supported: 802.3, 802.3u, 802.2, 802.1p, 802.1Q, 802.1 GMRP
- Standards Compliance: PCI 2.1/2.2, WfM 2.0, ACPI 1.0, RWU, PXE
- Driver Support: Windows 95/98/ME/NT/2000 – Linux, Netware, OS/2, Packet Driver
- Special Features: 3Com Parallel Tasking II
- Warranty: Limited Lifetime
Installation of the 3C905C-TX NIC card is straightforward: insert the card into a PCI slot and use the included EtherLink driver CD. This driver CD also provides installation options to install 3Com’s DynamicAccess technology, which adds a host of new management features to the NIC card. Additionally, this card provides one of the best visual diagnostic and configuration tools available today with a DOS based configuration tool, which allows you to set the NICs media type, duplex, enable/disable boot ROM, I/O Address, IRQ, and the card’s Network Optimization Mode. The DOS based configuration utility also allows you to run basic tests on the NIC card to help isolate a failure. This testing includes a unique Echo Server feature that allows testing the card and the cabling while looking for faults.
Setting the card’s “Network Optimization Mode” allows you to choose from these modes: “Maximized Network Performance”, “Minimized CPU Utilization”, or “Normal.” These options affect the overall performance of the card and can greatly assist in tweaking the NIC card’s configuration to a mode that best suits the user environment. In testing the card, varying the mode setting did make a difference in our benchmarking results, so we recommend trying different settings in order to find out the optimum mode for your use.
Living up to its 3Com reputation, we found the performance of the 3C905C-TX card to rank at the high end in Network throughput and toward the low end in CPU utilization. Additionally, we found this card and the drivers to be very stable. Overall, the 3C905C-TX card is an excellent choice. Although most users will not use all of the extensive management features of the card, this card is still an outstanding choice for most users. The price might scare some users away from this card. However, since this card is an IT favorite, it is fairly easy to find inexpensive used 3C905/3C905B/3C905C NIC cards at local computer shows.
3Com 3CR990-TX-97
The 3CR990-TX-97 is the next generation in NIC card technology from 3Com. This card is clearly a Business/Enterprise level NIC card with a new feature set built around 3Com’s newest on board 3XP processor. Some may wonder why 3Com has taken the approach to place an on board processor on this NIC card and the answer is simple: the need for additional processing power to handle one of this NIC card’s main features, full hardware based IP SEC off load support and possible future features.
IP SEC secures sensitive IP traffic by using 3DES, DES, MD, or SHA-1 encryption. With the 3CR990-TX-97 under Windows 2000, IP SEC is completely off loaded to the NIC card. This allows the NIC to share a greater load of network traffic processing with the result that the host system can focus on servicing applications, providing better network and system performance. In addition, the 3XP processor on the 3CR990-TX-97 off loads both TCP/IP Checksum and TCP Segmentation, which reduces host CPU load.
The 3CR990-TX-97 has the following features:
- Media: 10Base-T/100Base-TX
- Connector: RJ-45
- Bus: 32-bit PCI
- IEEE Standards Supported: 802.3, 802.3u, 802.2, 802.1p, 802.1Q, 802.1 GMRP
- Standards Compliance: PCI 2.1/2.2, WfM 2.0, ACPI 1.0, RWU, PXE
- Driver Support: Windows 95/98/ME/NT/2000 – Linux, Netware, OS/2, Packet Driver
- Special Features: 3XP Processor, IP SEC Technology, 128K SRAM, Off Loads
- Warranty: Limited Lifetime
A specific Server version of the 3CR990SRV97 goes one step beyond these features above and adds larger 256K SRAM, Self Healing Drivers, Resilient Server Link Technology, Bi-Directional Load Balancing, Multiple VLANs, and Hot Plug PCI. While the 3CR990SRV97 adds these additional features, it is still built on the same base platform as the 3CR990-TX-97.
Even if you do not foresee a use for the IP SEC that this card offers, do not be so quick to dismiss this card as an option. Because of the 3XP processor that this card provides, 3Com has been working with Secure Computing to add Embedded Firewall technology to the 3CR990-TX-97 NIC card. This is definitely something to be excited about! The Embedded Firewall technology is not vulnerable to software or operating system security holes. Although the Embedded Firewall technology has been announced by 3Com, it was unavailable for testing in this article, which we hope to review once it is released. At this time, the real limits of the 3XP processor are not known, but the ability to upgrade the feature set of the NIC card is something that has been needed for a long time.
The installation of the 3CR990-TX-97 NIC card went off without a hitch. Again, the 3Com drivers for the 3CR990-TX-97 provide a nice set of options. In this driver set, we found the ability to test and manage all of the card’s settings in one convenient place. We found the drivers to be extremely stable and caused no problems in our test system.
In testing procedures, we found that this NIC card has no equal in our Windows 2000 testing. Under Windows 2000 it was able to best each of the other cards by a comfortable margin. If you are running Windows 2000 and want maximum NIC performance, this is the card to buy. The 3CR990-TX-97’s performance under Windows 98 and DOS NDIS performance was average, but considering that this card is optimized for Windows 2000, this was to be expected. The card’s hefty price tag will scare some folks away, but if you are looking for the best, this card is it. It also has the upgrade capability that all other NIC cards lack. The added features like the new Embedded Firewall technology are interesting, but you have to wonder if this will result in a performance hit on the NIC.
SMC 1211TX
If a company can cram a lot of performance into an affordable package, it would have to be SMC. The SMC 1211TX NIC card can also be found boxed in retail packaging under the name “EZNET 10/100.” The 1211TX provides a mix of features that will meet the needs of most users. The 1211TX avoids the more advanced features, leaving those for SMC’s EtherPower II series NIC cards. The 1211TX is not heavy on management features, but does support [WOL].
The 1211TX has the following features:
- Media: 10Base-T/100Base-TX
- Connector: RJ-45
- Bus: 32-bit PCI
- IEEE Standards Supported: 802.3, 802.3u
- Standards Compliance: PCI 2.2, WOL, ACPI 1.0
- Driver Support: Windows 95/98/ME/NT/2000 – Linux, Netware, OS/2, Packet Driver
- Special Features: Early Receive & Transmit, Empty Boot ROM Socket
- Warranty: Limited Lifetime
The install of drivers was painless with no issues arising. Unlike many of the other cards, the 1211TX doesn’t offer the nice Windows GUI configuration and setup utility, rather it offers a DOS based setup and configuration utility along with some options that can be configured from within network properties. The only options that you can change on the 1211TX are: Early TX Threshold, Link Speed, Link Wake Up, Network Address, and PME Enable.
Many advanced users would like to have some additional control over the configuration of the card, but due to the card’s low price and target market you will not find such settings. This is as bare bones as NIC cards get and if you are looking for bells and whistles, you are going to be disappointed.
I found the SMC to be an average performer with stable drivers and a slick package. If you purchase the retail box version you even get 7ft Cat-5 Patch cable and a copy of Turbo Linux in the box. For the value segment of the market, this card represents solid performance at a reasonable price, but concerns about product changes during the card’s life cycle don’t make it a good choice for IT shops.
Intel Pro/100+ Management Adapter
Intel has been building Ethernet chipsets for a long time. At one time, Intel’s Ethernet chipsets were among some of the most popular for many 3rd party companies. Now however, many companies are using “cheaper” alternatives to the Intel’s Ethernet chipsets. This is not to say that Intel hasn’t continued to make vast strides in the development of their Ethernet chipset, it is just that few companies are taking advantage of these developments. As with all chipset products, the components become more intergraded and the size of the board shrinks further reducing cost.
The Intel Pro/100+ Management Adapter is the latest in a long line of Intel NIC cards. As in previous versions Intel continues to stress the over all manageability of the NIC card, but this time they go as far as adding “management” into the card’s name. The first thing that you will notice about the Pro/100+ after taking it out of the box is how small the card is. When compared to other products in review, the Intel Pro/100+ is between 10% to as much as 25% smaller than other NIC cards. It not only does this allow the card to be installed in the newer slim form factor PCs, but it can help increase the air flow within the case. Intel uses a unified driver package that covers almost the entire line of Intel NIC cards. Although, the drivers might be different depending on which card you install, the installation and utilities are almost the same. As with many cards, Windows does provide a default driver for the Pro/100+, but it is not as full featured and doesn’t have some of the performance gains that you may realize by using the Intel driver package.
The Intel Pro/100+ Management Adapter is a Business/Enterprise level NIC. It provides support of all of the major management standards. The card is priced fairly reasonable, but still costs more than some Home/SOHO NIC cards.
The Pro/100+ has the following features:
- Media: 10Base-T/100Base-TX
- Connector: RJ-45
- Bus: 32-bit PCI
- IEEE Standards Supported: 802.3, 802.3u, 802.2, 802.1p, 802.1Q, 802.1 GMRP
- Standards Compliance: PCI 2.2, WfM 2.0, ACPI 1.0, RWU, PXE
- Driver Support: Windows 95/98/ME/NT/2000 – Linux, Netware, OS/2, Packet Driver
- Special Features: 6K On Board Cache, PROSet Utility, Boot Agent, DCREATE.EXE
- Warranty: Limited Lifetime
The installation of the Intel Pro/100+ Management Adapter was a little more difficult than other card due to the fact that Windows 98SE and Windows 2000 automatically loaded a default driver for the device. Most users are not going to want to use these drivers. Although it might prove convenient, you will be missing the best feature of this card, the PROSet II Utilities. Installing under the control Panel in Windows 98SE or Windows 2000, this little utility not only gives you vast control over the configuration of the NIC card itself, but the performance of the NIC as well. For our testing purposes, we tested this card with the factory default settings. Like some of the other card utilities in this review, PROSet II gives you the ability to tweak the performance of the card to suit your needs or application. We had no problem uninstalling the default Windows provided drivers and replacing them with the Intel version 5.0 drivers with PROSet II. In fact, Intel includes a nice utility to automate the process of upgrading the drivers. This is handy if you want to upgrade a large group of users by using Microsoft SMS or a Server Login script.
We found Intel’s PROSet II utility and drivers to be very complete. The graphics based utility leads the user by the hand in tweaking the settings of the card.
We found the card’s performance to be near the top in throughput and in CPU Utilization. During our QCheck UDP testing the CPU utilization of the card ranged from 9% to as high as 15%. We did find it to be a consistent performer with very stable drivers. I can’t say much about the enclosed manual. Like many vendors, Intel has chosen to make most of the advanced documentation available only in electronic form. Intel does include some additional utilities for DMI, SNMP, 802.1p QoS I and II, as well additional utilities for network-based management.
3Com 3CSOHO100-TX
Since 3COM is one of the market leaders in NIC technology and didn’t have an off-the-shelf NIC with a price that vendors would touch, they introduced the 3CSOHO100-TX.
The 3CSOHO100-TX is, at first glance, very similar to the 3C905C-TX. The card uses a Lucent chip rather than the Broadcom chip that is found on the 3C905C-TX series cards. Also, gone is the WOL connector. Many people have suggested that this is nothing more than a stripped down version of the 3C905-C-TX. According to 3Com, nothing is farther from the truth. You will find the same quality that you find in other 3Com cards, yet this card is strictly targeted at the home consumer who wants the 3Com name at a cheaper price.
As with most NIC cards that are targeted to the home market, you will not have a guarantee of there being no design changes that will not require a driver change during the card’s life cycle. You also give up many of the management features that most SOHO or home users don’t need or use. One other interesting thing of note is that this card only supports the following operating systems: Windows 95/98/98SE/ME/NT 3.51/NT 4.0. Currently the card has no additional driver support for any other operating systems. 3Com offers no direct Linux driver support for the 3CSOHO100-TX, but the Tulip driver in most Linux distributions appear to support the 3CSOHO100-TX. 3Com has released an unsupported Beta driver for Windows 2000. I expect 3Com to add Windows 2000 and XP driver support the 3CSOHO100-TX, once XP ships later this year.
The 3CSOHO100-TX has the following features:
- Media: 10Base-T/100Base-TX
- Connector: RJ-45
- Bus: 32-bit PCI
- IEEE Standards Supported: 802.3, 802.3u
- Standards Compliance: PCI 2.2, ACPI 1.0
- Driver Support: Windows 95/98/ME/NT
- Special Features: Early Receive & Transmit
- Warranty: Limited Lifetime
We didn’t experience any problems installing the drivers. Since the card is not yet supported under Windows 2000, we were forced to use the 3Com’s beta driver for the 3CSOHO100-TX, which is posted on the web site. We were unable to run the GHOST re-clone test, because 3Com does not have a DOS NDIS available for this card. Of course if you are in an IT shop that uses a DOS based re-clone process that requires a DOS NDIS driver, then you are our of luck if you choose this card.
The performance of the card was better than we expected. The card always had low CPU utilization, which is a real plus. The throughput of the card was about average. The driver only allows you to change three options, which were Network Driver Optimization, Duplex, and Media Type. As were have covered in the other reviews of the 3Com cards, using the driver’s optimization setting can improve the performance of the card. In doing so, we really didn’t see much of a gain on the 3CSOHO100-TX’s performance.
We were impressed with the included 3Com Network Assistant CD that is included with the card. This multimedia presentation gives you a basic overview on the configuration and setup of a basic peer-to-peer network. We found the CD informative and a good addition to the package for the novice network user.
The card is impressive when compared to other budget cards, but the lack of a long life cycle makes this card a bad choice for IT shops. Although we found no issues with the Beta Windows 2000 drivers, we would not suggest the purchase of the card until these drivers are a released and supported. Whether you are connecting one system to a DSL or Cable modem or just networking a few Windows 95/98 systems in a home or small office, this card presents an attractive choice, as long as one is aware of the limitations of this card.
The Benchmark Questions – How do we measure each card’s performance?
In order to get a real-world performance measurement of these NIC cards, we tried to use a system that is more reflective of what many people are still using today. Our test system was configured as follows:
- Asus P3B-F
- Pentium III 600mhz Coppermine Slot 1
- 256MB Micron CL2 PC 100 RAM
- Asus 7100 Pure Geforce2 MX
- Quantum 10.2GB Fireball EX Series
- Asus 50X CD-ROM Drive
- Hercules Gamesurround Fortissimo II Sound Card
- Enlight 7237 Mid-Tower Case w/300watt P/S
Using the configuration above, we built standard base line Ghost Images for both Windows 98SE and Windows 2000 Professional. Each of these images are built without a NIC card installed. At the start of each test run, we re-clone the test system with a fresh image and follow the standard NIC installation process that was required for each card. Each NIC card in this test used the same PCI slot. Each NIC card was set to auto-detect the line speed. We confirmed, unless otherwise noted, that each of these NIC cards were connecting to the server in 100mbit full duplex mode. We used the default optimization settings for the NIC card. We didn’t attempt to alter or customize the setting to better optimize the NIC’s performance. It is possible with some of the cards to customize settings, which could yield additional increases in performance. We felt that these options were best left to the user to customize and experiment with.
To eliminate the possibility of testing results that could be skewed by the use of a hub or switch, we used a CAT5 crossover cable that was tested and certified good by a Fluke Cable Test Device. The test system was connected to a Windows NT 4.0 Server running Windows NT 4.0 with Service Pack 6a. This server was using the 3Com 3C905C-TX NIC card. In our testing, we changed the server’s NIC card and found no change in performance on our test machine.
To measure the performance and CPU utilization of each NIC card, we ran three different tests, both under Windows 98SE and Windows 2000.
NetIQ’s QCheck Utility
This is a free utility that you can download from http://www.netiq.com/qcheck/default.asp – QCheck is a free network performance utility that is part of the larger NetIQ Chariot Network Performance Suite. We ran this test five times, rebooting between each run under both Windows 98SE and Windows 2000.
Using QCheck we look at three things: TCP Response Time, TCP Throughput, and UDP Streaming. Although QCheck uses simulated data, most manufacturers use this program in combination with the Chariot Network Performance Suite to test their own cards. Here is a little more background on what we hope to find out from each of these tests.
TCP Response Time
This test is very similar to “ping” and tells you how long it takes another machine to respond. This measurement is commonly referred to as “lag” or “latency”. We use 10 iterations of 100 byte data size unless otherwise noted.
TCP Throughput
Tests the time it takes to transfer a block of data from one computer to another. Sometimes called “bandwidth”, it is a measure of a network’s raw transfer speed. We use 1000kByte test data size unless otherwise noted.
UDP Streaming: [description from Qcheck Help file]
“Streaming tests, like multimedia applications, send data without acknowledgments. In Qcheck, streaming uses the connectionless protocols IPX or UDP. Qcheck’s streaming tests emulate the behavior of applications that use streaming, such as Voice over IP and video broadcasts. This test indicates how much bandwidth the multimedia traffic might need, and makes it easy to compare the speed of the network hardware with the actual data rate the network is achieving.” We used Qcheck’s default setting in the UDP mode and report Qcheck’s numbers for throughput as well as CPU utilization.
During the QCheck phase of the testing in both the TCP and UDP tests we did look for possible packet loss. All five cards had a 0.0% packet loss during both the TCP and UDP tests. Because of our testing setup using a cross over cable to connect direct to the server, this is what we would have expected. In a real word network or Internet connection, packet loss is more common during heavy traffic periods and is to sometimes be expected. The CPU utilization was zero percent for all cards when running Windows 2000.
Command Line FTP Throughput and CPU Utilization
After rebooting each system, we launch the Windows CPU Performance Monitor, start recording, and then open a DOS window and complete a FTP transfer of Microsoft’s DirectX 8a Redistribution package, which has a file size of 25,939kb. (You can download the file at: http://www.microsoft.com/downloads/release.asp?ReleaseID=27398 .) We set the FTP file type to binary and sent the file to the Windows NT 4 Server running the FTP service. The values reported for throughput and time are the values shown in the graph. Once the transfer is complete, we then close the DOS window and look at the maximum/peak CPU value in the Windows Performance Monitor that is given on the bottom status bar. This is the figure that we use for our CPU peak utilization number. We run this test five times under both Windows 98SE and Windows 2000, rebooting between each testing cycle. This test attempts to better simulate expected real-world performance of the NIC card using non-simulated data.
Ghost Image TCP/IP DOS NDIS Performance
For fun more than anything else, we measure the time it takes to re-clone the system using Symantec’s Ghost Enterprise Version 6.51 using the manufacturer’s NDIS driver for DOS and very basic DOS TCP/IP boot disk. Since we ran each test five times, we wrote down the time and throughput numbers as reported by Ghost. If you are cloning a lot of machines in your business using a process similar to this, these numbers might mean something to you, if not, these numbers will mean nothing. The size of the Windows 98SE image file was 242MB. It was dumped to the server using Ghost’s high compression setting.
Conclusion
Well, there is some difference between Network cards. The difference was not that great. Some cards did have better performance than others, but over all most of the cards performed within acceptable ranges.
For business and IT shops who are running Windows 95/98/98SE/ME the choice seems clear with the 3C905C-TX. This card offers great performance and good management features with a very stable and proven driver set.
For business and IT shops that are running Windows 2000 the choice seems clear with the 3CR990-TX-97. With the additional of IP Sec support for later this card seems to have no equal under Windows 2000. Although we found the card to be expensive, we still found the extra performance worth the money. The 3C905C-TX also proved to have good scores under Windows 2000 and would be the best second choice for Windows 2000 users.
For home users who want the best all around performance we would suggest the 3C905C-TX. Although, you are not likely to use all of the management features of the card, it still is a solid performer. You also get a nice little tweaking utility to improve this cards performance.