Date: August 17, 2007 - Author: Rob Williams
One of the most popular CPUs on the market right now is the Q6600, thanks to the fact that it offers four cores at 2.4GHz. But what about the Xeon's? Their prices are also more affordable now, with their X3210 2.13GHz retailing for $260. Read on as we pit this Quad-Core against the rest of our fleet.
With last months launch of Intel's 6x50 series of processors also came price drops that everyone had been anticipating for months. Rumors were that you would be able to purchase a Quad-Core for close to $300, and to everyone's relief, it proved absolutely true. However, while most will run towards the Q6600, many will overlook the Xeon alternatives. While not normally clocked the same, they had price drops as well, and should not be ignored.
One of the more popular Xeon's that were noticed was the X3210, a Quad-Core clocked at 2.13GHz. While the clock speed leaves a bit to be desired for desktop users, it was hard to ignore the $250 price tag. This was at a time when Q6600s retailed for $280 - $300, so a small frequency drop could save you upwards of $30.
But, to say 'less expensive' almost seems silly. Just last year, an E6600 cost close to $400, while a Q6600 today retails for $300 or lower. It's a great time to be building a new computer, there's no doubt about that.
After last months price drops though, prices have fluctuated constantly. One day, the Q6600 might retail for $280 from your favorite e-tailer and then the next, you'll find it for $350. So needless to say, if you spot the CPU on the cheap, it's probably not the best idea to sit back and wait, if you want to secure it for the lowest price possible.
At the time of writing, US e-tailers are selling the Q6600 2.4GHz Quad-Core for an average price of $280.00, while the X3210 sits at closer to $260. At that point, if big overclocking is in the cards, then the Q6600 is well worth the extra $20, unless you happen to find a X3210 G0 revision. Ahh, revision hunting.
When Intel released G0 revision processors, it was a good day for enthusiasts. Every model seemed to overclock far beyond what was possible with previous revisions, so it's important to keep an eye out if overclocking is important to you. This is where things get tricky, however. If you are purchasing a processor from a random e-tailer, chances are good that it's a luck of the draw. However, you might be lucky enough to visit an e-tailer that will list the entire spec number.
If the X3210 spec ends with SLACU, you know that it's a G0 revision. If it's SL9UQ, it's a B3, which is what we have on the test-bench today. On the Q6600 side of things, G0 revisions will end with SLACR, while B3 ends with SL9UM. So, if you can manage to snag a G0 revision, you will not be disappointed. If you don't have a choice and end up with a B3, hope should not be lost, as they are still amazing performers.
For those wondering if Xeon processors are better in other ways over standard desktop chips, the answer is yes, but it might not mean much to you in the long run. Xeon processors are the cream of Intel's crop, so they go through a more stringent testing process to make sure they are perfect for server use. Aside from that, you could overclock a Xeon and Core 2 Quad to the same clock speeds and performance would be equaled.
That all said, the processor we are taking a look at today, once again, is the Xeon X3210, clocked at 2.13GHz with a TDP of 105W and will run stable at 1.100v - 1.372v, depending on your motherboard and luck. Like it's desktop counter-parts, this CPU also features a 1066FSB with 8MB total L2 cache.
This review will not be as full-featured as our usual offerings, since this product has been out for quite a while, and there aren't any surprises since it's mainly a speed change from the normal Core 2 Quads. Also, a lot of text is borrowed from our previous CPU reviews since it applies here as well. With that said, let's jump straight into our testing methodology.
Regardless of the OS we are running or product being reviewed, there are a few conditions that need to be met:
Here is the machine used for testing, followed by our operating systems configurations.
For our CPU reviews, we use two different versions of Windows and one version of Linux. Even though Vista has been out for half a year now, we focus on XP because it has a much wider user base, and is preferred for the best performance and compatibility. Vista is used only for our SYSmark 2007 Preview suite.
Game benchmarking is an important part of testing the capabilities of a CPU, and for this review we've included four popular titles: Half-Life 2, Prey, STALKER and Supreme Commander. Average FPS is captured using FRAPS 2.8.2, except for Prey, which is our only non-manual game.
Half-Life, Prey and Stalker are played at 1600x1200 with 4xAA, Half-Life being the only one with bumped AF.
Supreme Commander is the only game run at 1920x1200 with all options maxed alongside an 8xAA. This is because the game is multi-core compatible, and we wanted to see if the game would benefit from a quad-core in a realistic scenario.
Each game play through lasts between 3 and 5 minutes, except Supreme Commander which lasts closer to 8. All run-throughs are manually played, except for Prey which uses a time demo, in order to break through the 60FPS hard-limit. Results are captured using FRAPS 2.82.
All other non-game benchmarks will be explained along the way. Without further ado, let's proceed.
SYSmark is an industry leading system benchmarking tool, which is completely automated but utilizes real-world tests. It installs common applications such as Microsoft Word and Excel, Photoshop CS2, 3DS Max, SketchUp! among others.
SYSmark grades the performance of the system by how well it could handle different operations. Systems with more than one core will benefit in the tests, since there is a lot of multi-tasking throughout. Once the test is completed, it will provide you with an overall score, in addition to showing areas where the computer excelled.
Running the entire suite shows us that this benchmark is totally capable of handing out appropriate scores for very fast CPUs. There are a few interesting points here. First is the fact that the extra CPU power helped the QX6850 storm past the others in the 3D tests, scoring a full 40 points higher than the 2.4GHz Q6600 and 53 points higher than our X3210. This is one benchmark where it seems faster Dual-Cores could outperform slower Quad-Cores.
According to SYSmark, an E6750 is essentially on par with our X3210. If the suite put a greater push on using all of the available cores, we would have seen different results. However, it's at that point when the scenario would no longer be realistic. This does show us that Quad-Cores are not necessarily required for multi-tasking in an everyday environment though.
When thinking about faster processors or processors with more cores, multi-media projects immediately come to mind as being the prime suspects for having the greatest benefit. However, anyone who regularly uses Linux knows that a faster processor can greatly improve application compiling, with the GCC compiler. Programmers themselves would see the greatest benefit, especially those who find themselves recompiling their application every few hours.
Even if you don't use Linux, the results found here can benefit programmers in general, or those who simply wish to know what faster frequencies and additional cores are capable of. GCC is completely multithreaded friendly, so the results found here should represent the average increase you would see with similar multithreaded applications.
For testing, we use Fedora 7 x86, as it's an easy distro to install and maintain. It's not bloated, which is important as well. Our target is a copy of Wine 0.9.30. The distro is based on the 2.6.21 Linux kernel and we are using GCC 4.1.2 as our compiler. For single core testing, "time make" was used while dual and quad core compilations used "time make -j 3" and "time make -j 5", respectively.
As a single-core application, the X3210 was outpaced by everything, given the lower clockspeed, but when put into full gear (all four cores) it surpassed the Dual-Core processors on hand. If you have the need to compile, a Quad-Core of any speed is going to be amazing.
Any type of multi-media encoding/re-encoding/conversion will greatly benefit from multi-core processors and those with faster frequencies. Nero Recode is a tool we enjoy using because it's easy to use, and very effective. For testing, we first ripped our copy of Lamb of God's concert DVD, "Killadelphia" and re-encoded it using Recode to fit on a normal sized 4.5GB DVD. The original DVD rip weighs in at 7.7GB.
For whatever reason, Nero Recode doesn't utilize all cores 100%, but rather it hovers around 75%. I believe this to be so that the user will not have a laggy computer while the operation is in progress. Moving up from a dual-core to quad-core exhibits obvious benefits. Between our Quad-Cores, differences are there, but not as large as when compared to a Dual-Core.
If you chose "High-quality mode" for the recode process, it normally will use more than 75% of the overall CPU, but many will normally run the same settings that we had. However, because of the faster processors being released, many may move on over to the high-quality mode, simply because it's faster than 'normal' mode was just a year ago, on any available CPU.
DVD ripping/re-encoding is a popular pass time, but so is video editing and conversion in general. For this test, we used VirtualDub 1.6.19 and DiVX 6.0. Using FRAPS, we originally recorded a three minute gameplay run of Half-Life 2, through the d1_canals_07 level recorded using a 1280x1024 resolution. The outputted file weighed in at 3.7GB, obviously far too large of a file to upload to share with friends. So, the file was re-encoded using the DiVX codec and also resized to a 640px wide resolution.
As you can see, the DiVX 6 encoder doesn't take advantage of the quad-core to the fullest extent, but it usually hovered around 1.2 cores, on all five of our CPUs. For future reviews, we will be testing different video conversion tools to find a more appropriate multithreaded option.
Audio conversion is another popular scenario some people will run into, especially if they have FLAC files on their PC but want lossy versions for their iPod or other music player. Sadly, like VirtualDub, LAME is also a single-threaded application, so the results will scale with frequencies, not because of additional cores. There -is- a multithreaded version of the application available, but it's constantly argued by audiophiles that the resulting quality is not as high as the singlethreaded version. So if you really care about the quality of your music, the single-threaded option is the smarter move.
As was expected, results scaled perfectly with frequency and not the amount of cores. That said, let's move on to our workstation benchmarks, where Quad-Cores do make a big difference.
One crowd that benefits from faster processors and additional cores is the workstation crowd, those who render huge models that can take hours to complete. 3D Studio Max has been multithreaded for quite a while, even back in the day when some workstations had two physical CPUs installed, with one core each.
This is one real-world scenario where a Quad-Core processor can actually increase the rendering time upwards of 3.8x, so the benefits of owning such a chip is undeniable. For our test, we render an object with 406 objects and 106,000 vertices to a 3200x2400 resolution.
As we mentioned earlier, sometimes the lower frequency won't matter, when extra cores are put to good use. Compared to our faster Dual-Core E6750, the X3210 still proved 40 seconds faster overall.
Like 3D Studio Max, Cinebench emulates a similar scenario, where you render a high-resolution image. The nice thing about Cinebench though, is that it spits out a nice, clean number and also shows you your gain when moving to multi-threaded rendering.
The results here reflect those of our 3D Studio Max 9 tests... extra cores are a blessing.
POV-Ray is another program with the sole purpose of creating amazing scenes and models. Similar to Cinebench, POV-Ray includes a multithreaded benchmark that will render a large model and output a score of PPS, or pixels per second.
The results with POV-Ray are again as we expected. The lower clock speed means nothing when extra cores are put to good use.
When people think about faster processors, not many will think about the benefits of photo editing. But it's true, multi-tasking in general is far more efficient on a multi-core processor, and so when running numerous photo-related applications at the same time, thing should prove quicker.
In an age when DSLR cameras are becoming more and more affordable, lots of people are now dealing with RAW file formats. Adobe Lightroom is a fantastic application for altering your RAW photos, which is why we chose it to benchmark with. The actual test consists of taking 100 RAW (Nikon .NEF) 10 Megapixel photos, and exporting them to JPG format, while at the same time, resizing all of them to 1000x667 resolution.
For some reason, version 1.1 of Lightroom -hurts- performance with conversions like this, although exhibiting absolutely zero quality differences. I've e-mailed Adobe regarding the issue multiple times, without a response. So if such tasks are something you plan on performing often, it might be wise to stick to 1.0.
Even though multi-core processors are not necessarily new, it's tricky finding a photo application that handles them properly. Lightroom was one, Photoshop is another. However, writing scripts for Photoshop is ridiculous. Instead, we are testing the single core benefit of ImageMagick, a popular image command line image editing application for Linux, Windows and Mac OS. It's mostly used on servers, but serves the purpose here as well.
The benchmark consists of taking the 100 outputted JPGs from our last test and watermarking them. Then, it creates 500x335 thumbnails for each of those.
This test would be far more impressive if more than one core would be used, but this can accurately show you how performance for single-threaded applications could be increased simply because of the better frequency.
For the last of our non-gaming related benchmarks, we will spin each CPU through both 7-Zip and Futuremark's 3D Mark and PC Mark. 7-Zip is an interesting benchmark, because results can depend entirely on the settings you choose. By default, the LZMA algorithm is used to compress your files, but it's very slow overall and will use no more than two cores in your system. Bzip2, however, utilizes all four cores and is faster regardless.
As a comparison, LZMA took 1,385 seconds on our E6600, while Bzip2 took 801. The differences in outputted file size was that the Bzip2 method was 10MB larger. For a 3.85GB file, I will take the much faster speed over 10MB. So, Bzip2 is the algorithm used here.
All three quad-cores performed exceptionally here, though the extra frequency can also make quite a large difference.
Though Futuremarks products usefulness are often disputed, we like to include them because they somewhat accurately show how one CPUs power will compare to another by giving a single clean number. 3D Mark 06 stresses the CPU in such a way an actual game would, while PC Mark puts it through tests based on real-world scenarios.
Finally... gaming performance up next!
As mentioned on our testing methodology page, all FPS are tracked using FRAPS 2.82 except for Prey, which is the only title here to use a timedemo. There are two reasons for this. First, the game limits the framerate to 60FPS, while the timedemo does not. The downside is that at that point, it's more CPU dependant than GPU dependant. In the case of this review though, that's a good thing.
All games are run at 1600x1200 with mid-level AA and AF. The exception is Supreme Commander which is run at 1920x1200 and completely maxed out settings. An 8800GTX card is used to remove as much of a bottleneck as possible, so that any increase of Avg FPS will be solely because of the faster processor. Note that Half-Life 2, Stalker and SupCom are completely manual play-throughs, so variances in each run could cause a CPU achieve a lower score.
Half-Life 2 is a heavily CPU-bound game, although the game doesn't take advantage of dual core or higher processors. With 160FPS as is though, do we really need it to?
As evidenced by comparing the E6600 to Q6600, STALKER is single-threaded, but improved by faster CPU frequencies.
Being a timedemo, the highest FPS will automatically go to the fastest processor. It gives us a general idea of what benefits to expect from moving up to a faster frequency, however. As mentioned on our testing methodology page, SupCom is run with maxed out graphics (8xAA) at 1920x1200 to push our GPU as much as possible and also to see if more than one CPU core will be used.
The level chosen is Finn's Revenge, run off a saved-game that we previously made which begins us late into the mission. For this run, we create a few extra killing machines in addition to what we already have, and command them to go destroy the enemy. The entire process takes about eight minutes, and you can see an example of how much action and bots are on the screen on our testing methodology page.
At any given time, not more than 1.5 cores were used, and that was only random peaks. I am not sure what it would take to have even more usage, but the fact is, if you have at least a dual core processor, you can expect good results. By comparison, I disabled one of the cores on the E6600 processor and received an average of 37FPS.
Let's wrap the review up with a focus on overclocking.
If you've owned a Quad-Core processor in the past, chances are, you are well aware that overclocking is not always so peachy. On Dual-Cores, there is a lot of overclocking headroom, but cram four cores together under the same IHS and it's hit or miss. Another issue to take into consideration is heat... that's been my biggest issue with overclocking all three Quad-Cores on hand.
As mentioned in our recent review of Intel's QX6850, there is no reliable way to record temperatures from Quad-Cores, however from our tests, Everest 4.0 seemed to be pretty reasonable. During overclocking, the biggest instability hits would occur whenever Everest reported a core at 100°C or higher.
If you have a Quad-Core and have intentions to overclock, it's important that you have a very nice air cooler or water-cooling setup and do your best to keep the voltage no higher than 1.45v.
The first goal we set when finding a max overclock, is to first find a max overclock with stock voltages. We chose to stick with our usual 1.30v and managed to hit 2.48GHz. Nothing extreme, but a nice modest overclock regardless.
With a slight voltage boost (1.35v), 2.67GHz became completely stable, while never reaching the temperature threshold. In the end, this was our top overclock, while retaining stability. Allow me to quote a previous review:
For an overclock to be considered stable, 3D Mark 06 is looped three times, followed by at least eight straight hours of SP2004's Small FFT test, one instance per core. The computer must also be turned on and off multiple times, to make sure that the motherboard is going to cooperate. As a final test, the computer is left turned off for at least a half-hour, then turned back on to make sure it performs normally.
Overclocks that we call stable, have actually proven stable. Major overclocks are nice, but when they are unstable, we could care less. That said, I believe our processor has further pushing room, but we hit a temperature block. At 3.0GHz for example, the computer would reboot shortly after Everest 4.0 reported 100°C on the CPU (which also seems to be the highest that Everest will report).
I believe if you have a solid water-cooling setup, and even possibly a huge air cooler in a low-temp room, then you should be able to accomplish a similar overclock easily. Our testing environment hovers around 80°F at any given time, with >80°F being a rarity. Coupled with the fact that our water-cooling block has been switched around at least 30 times, our testing might not be representative of most systems.
As a 'single-core' though, 3.0GHz was absolutely stable, however not when all four cores were stressed at once. It's with that, that I believe it could be entirely stable with proper cooling. For overclockers looking for pure frequency without the care of stability, there's no doubt this CPU can be pushed beyond the 3GHz mark.
With the recent Intel price drops, many options opened up for consumers who were looking to build a new PC. It's a great situation to be in. No matter what CPU you pick up, you will be walking away with one that offers killer performance, for not much money.
How the X3210 compares to others out there is a tough one. Dollar for dollar, I would easily pick up the Q6600 for the fact it has a faster clock speed and a 9x multiplier, while the X3210 has only 6x - 8x multipliers. This normally won't mean that much anyway, since most motherboards will hit 400FSB with ease, which would give you 3.2GHz. As we found out in our case though, that clock speed will not likely be possible for most people, without heat getting in the way.
If you are simply looking for a Quad-Core at the best price, you cannot go wrong with this one. It retails for $260 - $270 and offers four cores at a very respectable clock speed. It's rare to see a value like this... and just last year, it would have not crossed our minds. That said, if you -are- interested in overclocking a Quad-Core, do what you can to find a G0 revision, as they have slightly lower TDPs and tend to overclock better.
With Penryn right around the corner, the decision to pick up a processor now is made more difficult. However, Penryn is not going to launch at rock bottom prices like the Core 2 Quads are now, because they are revamped architectures, not simply clock bumps. With prices this low, you are not going to go wrong, with whatever you choose.
If you have a comment you wish to make on this review, feel free to head on into our forums! There is no need to register in order to reply to such threads.
Copyright © 2005-2008 Techgage Networks Inc. - All Rights Reserved.