If you wanted to find out what the fastest desktop CPU is, it would just take a few clicks and you would have 10 reviews at your disposal. If you're trying to put together a workstation system, you'll have a bit less to go off of but there is still more than enough information available to help make your decision. But when it comes to servers, the information flow is severely reduced.

Generally speaking, you'll find reviews of individual servers from Dell and HP but you won't find the type of component level comparisons we're used to seeing in the desktop and workstation sectors.

There are a number of reasons for this. First of all, it's very difficult to test the performance of CPUs and platforms in a server environment without actually setting up such an environment. Especially as you get into the 4-way and beyond realms, it becomes very difficult to stress those systems. There are industry standard benchmarks that can be used but again, they are very costly to run and require a great deal of setup. In the desktop and workstation worlds, you simply load up the application you wish to measure performance under and find a benchmark to help you gauge performance.

Putting the difficulty of finding benchmarks aside, other issues need to be taken into account. A huge value in the server market lies in manageability features, which are generally platform/motherboard specific and even system specific. With servers hundreds or thousands of miles away from those that need to work on them, being able to perform simple tasks like flashing a BIOS, power cycling a system or faulting a failed SCSI drive is critical to uptime. If you've ever tried to reboot your crashed home PC while 30 minutes away at work you'll realize that these features must be added by an OEM. These features don't shine through, however, when you're just comparing two CPUs and how they perform in a particular server application. You may have one CPU that is twice as fast as another, but it lacks any platforms with the manageability features necessary for the market.

Then there are issues of reliability and support that are even more difficult to test. In the end, there is good reason for the lack of component level testing for servers. When we have to upgrade our servers, we don't turn to Dell or HP; we configure the platforms ourselves and it's helpful to know what offers the most bang for your buck and what does end up offering the best performance. After all, we manage these servers ourselves and we are in the position to try solutions that aren't necessarily proven but just need a chance to shine. If you ask a Fortune 500 company to move over their servers to AMD's first multiprocessor platform ever, they won't even return your phone calls, but we'll definitely do it (and we have).

That brings us to the current issue at hand; when the Intel Xeon processor launched last May it performed well but was easily beaten by the 1.2GHz Athlon MP processor just one month later. On the desktop end, the new 0.13-micron Northwood based Pentium 4 processors are finally outperforming the Athlon XPs, but will this translate into similar results in the server world?

We asked ourselves the same question as we took the three databases that power AnandTech and benchmarked the latest entry-level 2-way server CPUs. The data ended up being extremely useful to us and thus we decided to share it.

The Contenders: Athlon MP 2000+ vs. Intel Xeon 2.2GHz

Since the launch of the Palomino core in the desktop market, AMD has transitioned all of their processors to their new modeling system. AMD's Athlon MP CPUs are architecturally identical to their Athlon XP CPUs, and we are not sure if they are validated any differently. The modeling system obviously makes much less sense in the server world as the benchmarks used to come up with the model numbers are not server benchmarks at all. Customers in the server markets are also much less likely to buy based on clock speed alone.

Virtually identical to the Athlon XP desktop CPU

AMD's current server CPU release schedule puts the launch of Athlon MP processors around a month after the release of equivalently clocked desktop CPUs. The launch of the Athlon MP 2000+ actually happened on the same day as the launch of the Athlon XP 2100+, meaning that the Athlon MP 2100+ isn't too far away. For more information on AMD's MP architecture be sure to read our review of the 760MP chipset.

The Duron MP processor is still on hold and will most likely never see the light of day. If properly validated, the Duron MP platform would end up costing much more than AMD would like to spend on an extremely low-cost server CPU. The performance of the CPU, as we've illustrated in the past, would simply be too low for any serious server applications (on par with desktop Pentium III servers).

On the Intel side of things, the dual processor Xeon has moved over to Intel's 0.13-micron manufacturing process. Just like the Xeon's desktop counterpart, the CPUs feature a 512KB on-die L2 cache, which definitely comes in handy in server situations. Unlike the Athlon MP/XP relationship, the Xeons plug into a different socket (Socket-603) and use a different processor substrate with additional thermal monitoring circuitry made specifically for the server market.

The additional pins used on the Xeon processors are mainly necessary for delivering power to the larger cache versions of the CPUs

Click to Enlarge

Quite possibly the biggest difference between the Xeon and its otherwise identical desktop counterpart is the enablement of Hyper Threading technology on the CPU's core. As we've discussed in the past, Hyper Threading allows the creation of multiple logical CPUs on a single physical CPU for better usage of execution resources. With Hyper Threading still in its infant stages it results in a performance decrease in almost all applications with the exception of server applications This is why Intel only recommends enabling Hyper Threading (it's a BIOS option) when running servers.

Although all previous Pentium 4 and NetBurst based Xeons had the Hyper Threading circuitry on-die, it cannot be enabled through the same BIOS option. Only the new 0.13-micron Xeons offer the option. For more information on the Xeon MP architecture, be sure to read our original Xeon review.

The Platforms: AMD 760MPX vs. Intel E7500

You should already be pretty familiar with the AMD 760MPX chipset from our review as well as its predecessor, the 760MP. We ran all of our Athlon MP benchmarks using Tyan's Tiger MPX platform, which is their entry-level Athlon MP workstation/server board.

Two sockets, six DIMM slots, and a large heatsink on the E7500 MCH

The newcomer to this comparison is Intel's E7500 chipset. As you may recall from our preview of the E7500 chipset, it is Intel's first dual channel DDR solution. With support for DDR200 SDRAM and by populating two memory banks at a time, the chipset is able to match the i850/860's dual-channel RDRAM memory bus in terms of memory bandwidth. The 3.2GB/s of main memory bandwidth offered by this solution is perfectly matched with the Xeon's 100MHz quad-pumped FSB, which also provides 3.2GB/s of bandwidth to the CPU.

The chipset currently lacks DDR266 support mainly because there's no need for it, so long as the Xeon remains at a 100MHz FSB. The Xeon platform will inevitably transition to a 133MHz FSB (much like the desktop Pentium 4), at which point the E7500's successor will be introduced with DDR266 support.

DIMMs must be installed in pairs

We'll save our review of the E7500 chipset for a separate article, but the only major shortcoming is that it is truly a server-only chipset as it has no AGP controller. It could indicate that Intel doesn't want the E7500 being used in markets that will be serviced by a forthcoming dual channel DDR solution for desktops/workstations. Do keep in mind that the server world saw Intel's first dual-channel RDRAM chipset (i840) before it ended up being on desktops with the i850. We will address this theory in our standalone E7500 review.

Click to Enlarge

We used Tyan's Thunder i7500 motherboard for all of our Xeon tests. Using the latest shipping BIOS we ran into no issues with the board and it completed all of our tests without faults.

The chipset itself is targeted at the entry-level to middle of the server market, leaving the highest demands for ServerWorks' platforms. The main limitation that keeps the E7500 from competing in those markets as well is, at least according to ServerWorks, a lack of I/O bandwidth to its PCI/PCI-X busses. To get an idea of what ServerWorks considers to be lacking, we're talking about 120-drive RAID arrays capable of eating up multiple gigabytes per second of I/O bandwidth. Even our most strenuous tests cannot even begin to saturate that sort of bandwidth.

The Tests

As we've mentioned before, there are three databases that power AnandTech: the Web DB, the Ad DB and the Forums DB; we used all three databases to test these server CPUs.

The Web DB is where all of our content is stored; everything from news and reviews to our own internal article rankings are stored in this database. By far the majority of the transactions on this database are selects (reads). Remember that the web site only really offers one way interaction, the readers come to the site and read articles which are contained in this database. The articles are selected from the database and fed to one of the 4 web servers for assembly into a page for your browser. Internally, some update queries are also run, but they were not recorded in the test trace we ran. This database is the smallest out of the three; the DB was only 300MB when we ran the test.

The Ad DB is very similar to the web database in that quite a few selects are running. The select queries are used to pull the ads from the database for display in the user's browser. There are also a number of stored procedures that run along with the selects, but to keep things as simple as possible (at least for this comparison), we omitted them from the test trace. The Ad DB is noticeably larger than the web database, at a large 2.1GBs at the time of publication.

The final database is the Forums DB, which is by far the most transaction intensive database in the AnandTech Network. While the vast majority of the requests to the DB are in the form of selects (users reading categories and threads), there are significantly more inserts and updates (posting, thread/post counts, etc…) than in either of the other DBs. This database is also our largest, weighing in at just under 3GB during the testing and close to 8GB today (we used an older version of the DB from a few months ago).

In the past, when we used database server testing, it was done using a single trace run on the AnandTech Forums. While we're using two additional databases, the test methodology remains the same. We recorded a trace of transactions on each one of these databases for a set period of time. These were live recordings while the website and forums were being accessed just like they would on any normal day. The traces were then played back at full speed (as fast as the server test bed could replay them) and their playback times recorded. We divided the number of transactions replayed by the playback time and reported all scores in numbers of database transactions per second: the higher the better.

For our test bed, we setup a simple RAID 0 array (which would never fly in a real-world situation, no redundancy spells certain death) of two Seagate Cheetah X15 drives. While this gives us less I/O power than is in our current Forums DB made of 6 Quantum Atlas 10K II drives, it was enough for this limited test. In the future, we'll most likely move to a 4 - 8 drive test array as the benchmarks become more strenuous.

Two DIMMs, 2GBs - Click to Enlarge

The test beds were configured with 2GB of DDR266 SDRAM using two 1GB sticks. Since the E7500 chipset requires memory to be installed in pairs we had to use multiple DIMMs. Also remember that the dual-channel memory controller of the E7500 chipset only ran the memory bus at DDR200 speeds.

AnandTech Website Database Performance

First up on the test bench was our beloved website. Our fifth year anniversary is coming up at the end of this month and look at how far things have come in just five years. From running off of a free shared server on Geocities to complaining about how "slow" our old dual Athlon MP 1.2GHz setup was, thus requiring us to move to faster MP processors. The AnandTech Website DB is actually still running off of the old Pentium II Xeon processors we had from a few years back and we will hopefully be upgrading it soon, what should we upgrade it to? Let's find out:

Next up on the test list is our Ad Database Server test bench. All of the ads across AnandTech and the AnandTech Forums are served using e-Zone Media's FuseAds advertising management software. The software was custom tailored to fit our needs and is tightly integrated with our internal content management system. The end result is that the ad placement, rotation, statistics and customization is very flexible and easy to use for our sales staff, however its tight integration with the entire network means that this server is just as important as the content feeding servers. Should the Ad Database be slow to respond or not respond at all because of an overwhelming load, the entire site and forums would slow to a crawl.

Our Ad DB is still running off of a dual Pentium III 866 machine and will be upgraded over the next few weeks, what should this box be running instead?

AnandTech Forums Database Performance

Now it's time for the beast, the AnandTech Forums DB. With hundreds of users always logged on and thousands more just browsing the forums, this database gets a full workout. Now with over 88,000 registered users it has become a difficult task making sure our upgrade cycles keep up with the growth of the forums.

Currently our Forums DB server is running on a dual Athlon MP 1800+ platform which gives it enough headroom for the foreseeable future.

AnandTech Forums Database Performance Transactions per Second

Athlon MP 2000+ (1.6GHz) x 2 Xeon 2.2GHz x 2 (HT Enabled) Athlon MP 1.2GHz x 2 Xeon 2.2GHz x 2 Xeon 1.7GHz x 2 AMD Athlon MP 2000+ (1.67GHz) Intel Xeon 2.2GHz (HT Enabled) Intel Xeon 2.2GHz AMD Athlon MP 1.2GHz Intel Xeon 1.7GHz 211.1 205.9 185.5 175.9 156.6 141.4 134.5 123.2 112.8 96.7 | 0 | 42 | 84 | 127 | 169 | 211 | 253

If you'll recall from our original 760MP review, we ran a Forums DB trace on the first Athlon MP platforms and they completely dominated the first Xeons. If we were to knock off the recently released Athlon MP 2000+ platform and just compared the new Xeons to the Athlon MP 1.2GHz setup we ran back in June, the performance would still be impressive.

Do notice that although the dual Athlon MP 1.2GHz setup is 18% faster than the dual 1.7GHz Xeon platform, it is also just over 5% faster than the dual 2.2GHz Xeon platform. In our most strenuous database server test, the Athlon MP continues to shine and perform extremely well. The higher clocked CPUs result in a 13.8% boost over the first MPs that were released in June.

But the real kicker here is what happens when Hyper Threading is enabled on the dual 2.2GHz Xeon platform. The resulting 17% improvement brings the new Xeons back into the game and puts them less than 3% away from the fastest dual Athlon MP 2000+ setup. This is by far the largest tangible performance improvement we've been able to produce in a real-world setting on Intel's 2.2GHz Xeons. For being the first true incarnation of a Hyper Threading enabled processor, we can say that we're impressed with the technology at this point. The performance boost not only resulted in a noticeable improvement in DB processing power but it also made the Xeon competitive. Without Hyper Threading, the Xeon would be in some serious trouble in the server world from a performance standpoint. With Hyper Threading enabled however the 2.2GHz Xeons can keep up with AMD's Athlon MP 2000+ CPUs. Oh, how times have changed; it used to be that AMD would never have been taken seriously in the enterprise market, and now we're talking about Intel striving to be competitive enough with the AMD's enterprise level performance.

Final Words

Before we get into the meat of the conclusion, let's have a quick recap of the performance improvements we saw when going from single to dual CPUs and when toggling Hyper Threading (on the Xeon 2.2 CPUs):

AT Web DB Test	Dual AMD Athlon MP 2000+	Dual Intel Xeon 2.2GHz
% Improvement over Single CPU	49.32%	69.74%
% Improvement with HT Enabled	N/A	3.71%

AT Ad DB Test	Dual AMD Athlon MP 2000+	Dual Intel Xeon 2.2GHz
% Improvement over Single CPU	21.72%	19.61%
% Improvement with HT Enabled	N/A	-5.46%

AT Forums DB Test	Dual AMD Athlon MP 2000+	Dual Intel Xeon 2.2GHz
% Improvement over Single CPU	49.35%	42.77%
% Improvement with HT Enabled	N/A	17.05%

What the above tables show is that in spite of the Athlon MP's point-to-point connection to the North Bridge, MP scalability is still not dramatically improved over the shared bus used by the Xeon. In fact, in the Web DB test the Xeon scales much better than the Athlon MP, possibly because of more overall memory bandwidth. We've said it before but it is worth repeating especially now, the Athlon MP needed a dual channel DDR chipset for the enterprise market. Because NVIDIA would never think of bringing nForce for the validation nightmare known as the enterprise market, AMD probably figured that they would be better off waiting for Hammer to introduce a higher bandwidth memory solution for the enterprise market. It saves them the costs of producing and validating an even more complicated chipset that would be replaced by a simpler Hammer solution in a couple years time.

The other thing to take away from these tables is that Hyper Threading does have a great deal of potential. In our most stressful database server test, it improved performance by 17%. For a feature that takes up such a small percentage of the overall die size, it's not bad at all. These tests also show that the performance improvement of Hyper Threading is not always predictable. In the Web and Ad DB tests, the performance change varied from an increase of 3% to a decrease of 5%. You'll remember from our descriptions that both of these tests were heavy on the selects. The more varied benchmark was the Forums DB test and that's where we saw the largest overall improvement that Hyper Threading provided; it also happened to be the benchmark that the Xeons with Hyper Threading disabled did the worst in.

Over the past few months we've had many OEMs approach us asking about a comparison between the fastest Athlon MPs and Xeon processors in a true server environment; well, it doesn't get much more real-world than this.