It’s been a while since AMD has been in the position of the scrappy underdog. For several years they held the performance crown in desktop processors. Then came the Summer of 2006 and at COMPUTEX 2006, Intel released the Core 2 architecture. Suddenly AMD was back to playing second fiddle in the CPU shindig. AMD is now trying to fight back with it’s Phenom line of processors.
The Phenom line of processors is part of AMD’s new Spider platform. This platform was launched in November 2007, and the purpose behind it is to provide the ultimate HD performance platform. There are three parts to the platform, which AMD calls segments. They are as follows.
- ATI Radeon™ HD 3800 Series Video Cards
- AMD Phenom™ Processors
- AMD 7-Series Chipsets
This cadre of hardware is noted by AMD to provide “THE platform for PC gamers, media enthusiasts & content creators, and PC enthusiasts.” This is done through “platform upgradeability and customization, leading-edge performance and bandwidth to deliver next-generation HD experiences, and industry-leading platform performance-per-watt for flexibility and headroom.” Where the Phenom fits into those goals is providing an energy efficient, true quad core processor.
By now, many of us know what happened at the launch of the Spider platform and the Phenom processors. There we saw some initial performance concerns when compared clock-for-clock with equivalent Intel silicon. That problem was exacerbated by an erratum on the B2 stepping of the Phenom processor (the stepping that is currently available in retail) with the Translation Lookaside Buffer (TLB). This issue could result in data corruption or loss in certain performance scenarios; scenarios that are rarely seen on the desktop. However a fix was issued in the form of a BIOS tweak, but performance took another hit.
The Phenom launch was not without its bright spots. First and foremost was the fact that the Phenom would be the first true quad core processor, with all four cores being on the same die. The other plus side came in the form of one heck of an olive branch from AMD, in the form of a multiplier unlocked “Black Edition” of the Phenom 9600. We have just such a creature for a review. Here’s the run down of specs for the Phenom line of processors. Full specs can of course be found to the product page.
- The industry’s first true Quad core x86 processor
- AMD64 with Direct Connect Architecture
- AMD Balanced Smart Cache
- AMD Wide Floating Point Accelerator
- HyperTransport™ technology
- Integrated DDR2 DRAM Controller with AMD Memory Optimizer Technology
- AMD Virtualization™ (AMD-V™) With Rapid Virtualization Indexing
- AMD Cool’n’Quiet™ 2.0 technology
- AMD CoolCore™ Technology
- Dual Dynamic Power Management™
The Phenom 9600 Black Edition is clocked at 2.3GHz, and sports all those features noted above. Even with its unlocked multiplier, it still comes in at a very reasonable price. At many retailers you can find Black Edition Phenom 9600 units priced only $10 – $20 more than their standard Phenom 9600 counterparts. Now many are still left to wonder how much of a performance hit the TLB erratum fix the Phenom takes. We intend to find out with this processor, and see how far we can push the silicon with some good ol’ fashioned overclocking.
Real-Time Price and Stock Check – Shop Like a PRO!
The Test System
We’re going to get right into it with our testing. First and foremost we highlight what I think is the best software overclocking tool available…when it works. Then we get into some juicy overclocking goodness. But before all that, let’s run down the specs for the test rig for this review. We plugged the AMD Phenon 9600 Black Edition in the following system which includes…
- ASUS M3A32-MVP Deluxe WiFi-AP Motherboard
- eVGA GeForce 8800 GTX 768MB PCIe Video Card
- OCZ Platinum XTC REV.2 PC2-6400 2GB 2X1GB DDR2-800 Memory Kit
- Seagate Barracuda 500GB 7200.10 16MB SATA2 Hard Drive
- Cooler Master Hyper 212 CPU Cooler
- Antec P182 Advanced Super Mid Tower Case
- Thermaltake Toughpower 1000W Power Supply
- Microsoft Windows Vista Ultimate RTM
The only thing that we don’t have is an AMD/ATI 3800 series GPU which finishes off the Spider Platform. Since this is a Black Edition CPU, it comes without a heatsink so we used the recently reviewed Cooler Master Hyper 212 CPU Cooler.
OverDrive
OverDrive is AMD’s software overclocking tool, and is compatible with all modern AMD processors. It allows you to either take the easy way out and let the software decide how far you can push your CPU, or completely hand the controls over to you and give the ability to tweak every possible feature until your system is a well oiled machine. Once you’ve applied the changes, you can use the internal benchmarking tools and system monitors to make sure your system will function without dying in a puff of blue smoke.
Now for testing we used version 2.14 of AMD OverDrive. There is newer version out, numbered as 2.17. However in a long line of snags we encountered when trying to use this software for overclocking, version 2.17 appears to have some compatibilities with Vista. Upon attempting to launch version 2.17 of OverDrive, I was presented with an error saying the software couldn’t detect an AMD CPU. This was rather interesting, considering I had just inserted an AMD CPU into the motherboard.
After attempting to reinstall Vista, making sure my drivers and BIOS were up to date, and scratching my head a little bit, I performed a Google search and discovered this is an extremely common problem. The community’s recommendation? Either switch back to version 2.14 of OverDrive, or downgrade to Windows XP. The official response from AMD? Nothing gleaned from the first couple pages of the Google search. So I decided on the former option, since many of our benchmarks are DX10 and this is supposed to be a Vista compatible platform. Version 2.14 loaded without issue, which then started me down the path to the next problem with the Phenom 9600 Black Edition.
Real-Time Price and Stock Check – Shop Like a PRO!
Overclocking
Before I get to show you any screenshots, I have to come out and say that the Phenom 9600 Black Edition is easily one of the more difficult experiences I have had with overclocking. Pushing a CPU past its imposed limits is by no means a cake walk, and nor should it be. That being said, I had to put in a great deal of time and testing for what amounted to very little gain on a part specifically designed for overclocking. I could revel in hyperbole all day, but I think I need to show you screenshot of my final overclock to really get the point across.
For the dozen or so hours I put into tweaking, poking, and proding this system, I managed to only squeeze an additional 200MHz out of the Phenom 9600 Black Edition. I tried every voltage setting, most combinations of voltage settings, and even tried using the BIOS. Still nothing could appease the processing gods. 2.5GHz was the speed limit I was going to experience. Now some other sites have reported 2.6GHz before hitting a wall, a couple have hit 2.8GHz, and one hit what is a very dubious 3.0GHz. But not me. If I tried clocking the Phenom 9600 Black Edition up just 100MHz more, and even if I added some extra voltage, the second I hit Apply the system hard froze on me and I had to restart.
So onward I marched. I decided to benchmark the processor with the TLB fix disabled and enabled, to compare the performance hit. I also tested the Phenom while overclocked to see if in this day and age of nearly GHz overclocks a simple 200MHz boost made a difference. The overclocked tests also had the TLB erratum fix disabled, so as to get the maximum performance out of the processors. Let’s hope we can an 8% performance increase out of our 8% overclock.
Synthetic Benchmarks
For our first round of tests we turned to the synthetic benchmarking suites available from Futuremark, and the first suite we loaded up was PCMark05. Though somewhat long in the tooth, this suite of tests can still put many systems through their paces. Futuremark has also released a couple of patches to ensure PCMark05 can take full advantage of all available cores.
Like its much better known cousin, 3DMark2006, PCMark05 compiles its score from a slew of system tests designed to emulate standard desktop usage. These tests include processing, hard drive usage, memory, web page rendering, and some limited 3D. For this review we also ran all the CPU specific tests, which generated it’s own CPU specific composite score.
The results pretty well followed the curve I was expecting. Enabling the TLB erratum fix resulted in a performance hit, and the overclock resulted in a performance increase. Using the stock clock with TLB fix disabled as the zero point, as well as some quick maths, we can calculate that overclocking our Phenom resulted in a 5% performance increase overall, and 9% when the CPU is isolated. With the same maths we find that the TLB erratum fix caused a 2% decrease in performance overall, and a statistically insignificant drop with the CPU isolated. Moving along.
3DMark is like PCMark in that it generates a composite score based on several measures of performance. For a really good description, I’m going to copy and paste from the 3DMark Wikipedia article.
“The measurement unit 3DMark is intended to give a normalized mean for comparing different visual processing units, which proponents assert is indicative of end-user performance capabilities. Critics counter by stating that it is a synthetic measure of real-world performance.”
3DMark06 is the latest version of this testing suite, and although synthetic in nature it still provides a fairly consistent scale by which to measure system performance. For our tests we ran the benchmark at default settings to see what numbers we could eke out. In the case of our Phenom 9600 Black Edition we see what is going to be a trend through this review. While overclocked, we only saw a 5% performance increase from our test CPU. As for enabling the TLB erratum fix, when the dust settled that resulted in a 7% perform decrease.
Real-Time Price and Stock Check – Shop Like a PRO!
Video Encoding
3DMark and PCMark are great tools for getting a quick peak at to what your system performance should look like. That being said, you still want to run some real world benchmarks to test actual activities people perform on their computers. That’s exactly what we intent on doing. The first up in our suite of real world benchmarks in video encoding. Our tool of choice for this torture is Nero Recode. This video encoder supports many of the different portable formats the average user may use to re-encode their favourite content, and it’s fully multithreaded. This allows it to push any processor to maximum, showing how well it performs in pure processing.
The first movie to be run through our encoding test was the dystopian Sci-Fi romp known as Equilibrium. We chose to re-encode the DVD to iPod Video AVC based on the hunch that many people who have iPods like to put movies on their iPods. As for the title choice, Equilibrium represents a title of average length, a good mix of light and dark (mostly dark) scenes, and hey it’s just good fun.
With our 200MHz overclock, the Phenom 9600 Black Edition posted a performance gain of 6%. Now that’s coming much closer to matching the 8% overclock in percent for percent performance. When we put the processor back to stock clocks, and enabled the TLB erratum fix, we experienced a drop in performance of less than 1%. Now these performance deltas weren’t quite wide enough for my tastes, so I searched through my library for another title to test. Keeping the performance numbers just posted in mind, I looked for a longer movie with a lot more brightly lit scenes. I arrived at a title you may or may not have heard of.
In 1994 a movie came out that launched a franchise. That movie was Stargate, and it’s my second title selected for testing the Phenom 9600 Black Edition in video encoding. As noted above, I selected this movie for its longer….length. I also picked it for it’s brightly lit scenes (it’s shot in a desert), and some of it’s catching details.
This movie took nearly a full ten minutes more to encode in all instances than Equilibrium, and nearly rang in at an hour overall per run. As for performance, we see a 7.3% increase with our 8% overclock. This is even better than the previous title, and much closer to the percent-for-percent target I was hoping for. However with that excellent high comes a brutal low. The TLB erratum fix resulted in nearly a 4% drop in performance, which means this movie took an additional 2 minutes to encode with the fix enabled.
At this point it should be noted that even with all the processing involved in video encoding, and the fact that each movie had been encoded three times for a total of six runs, I had yet to see any data corruption or problems resulting from the TLB erratum. Needless to say a performance hit has been experienced in every benchmark so far due to the TLB erratum fix being enabled. Let’s take a look at some popular game titles now, and see if the same effect is present.
Real-Time Price and Stock Check – Shop Like a PRO!
First Person Shooter Benchmarks
Oh god, I can just see the onslaught of angry geek emails coming my way from the following statement. Anyhow, it goes without saying that 3D First Person Shooter (FPS) games are the de facto standard for system benchmarking. They tend to push your system to its limits in both processing and graphics rendering. For our FPS testing we picked two titles that have the most realistic benchmarks built in. One is a fairly punishing DX9 title, and one is the DX10 title know to make even the most burly (an expensive) computers cry for mercy.
F.E.A.R. was released in 2005, and still to this day it can test a system to see what it’s made of. For our testing purposes we maxed out all the settings, dialled the anti-aliasing up to 4x and anisotropic filtering up to 8x, and set the resolution to 1680×1050. We then ran the internal benchmark, which consists of a timedemo rendered in game. The benchmark was run three times, and the average was taken from all three scores.
Overclocking the Phenom 9600 Black Edition only resulted in an average performance increase of 3 frames per second (FPS), which with the applications of some more maths is only a 3% increase. The TLB erratum fix had a much more drastic effect in the opposite direction. Performance dropped 13 FPS on average when the fix was applied, which works out to a 12% decrease.
At the frame rates produced by F.E.A.R. in our test rig, these performance differences would not be noticeable. As long as the game doesn’t dip below 30 FPS your playing experience shouldn’t be affected. With that in mind, we move on to our next FPS title.
Crysis is the newest DX10 First Person eye candy fest. This game is thought of by many as the spiritual successor to Far Cry; the game that made performance computers and their owners cry throughout 2004. Like it’s predecessor, Crysis is known to bring even the most heartily equipped modern performance rigs to their knees. The game is a visual and auditory extravaganza that truly needs to be experienced. But enough of my fanboy gushing; there are benchmarks to be performed. For testing, we used the Crysis Benchmark Tool and ran the CPU test run. Unlike our previous benchmarks, we didn’t dare try to crank the Crysis benchmark settings to maximum. We used the high settings at a resolution of 1280×1024 with no anti-aliasing or anisotropic filtering. All tests were of course performed in DX10 mode.
At first glance, the numbers posted by our CPU time demo may seem like a small performance delta. There was only positive 2 FPS difference between overclocked and stock clocked, and a negative 3 FPS difference between TLB disabled and enabled. Even at those low numbers the results still work out to a 6% increase and a 9% decrease, respectively. As I had outlined before, when you are only averaging 30 FPS, every frame given or taken away really counts. This can be the difference between a minor lag where you only take a couple bullets before returning fire, and being dead.
Real-Time Price and Stock Check – Shop Like a PRO!
Real Time Strategy Benchmarks
First Person Shooters may be what everyone uses for benchmarks, but it’s not the be all and end all for gaming genres. You tend to see a little more CPU usage from strong RTS titles due to the AI and unit management required. At least that’s the theory. We going to test it with two titles; one was patched DX10 a while back, and one is a DX9 title that can bring a DX10-style system crushing to your rig. Let’s go.
And here’s where my theory falls apart. Company of Heroes was one of the first titles released under the “Games for Windows” umbrella, and was originally a DX9 title. A DX10 patch was later release, and it provided some graphical enhancements. The game also comes with a built in benchmark where you can compare your DX9 and DX10 performance among other things. For our tests, we focused exclusively on DX10. For this benchmark we once again maxed out all the settings, set anti-aliasing to 4x, and cranked the resolution to 1680×1050.
Our results didn’t amount to much. We overclocked the processor and poked and prodded at the TLB erratum fix, but there was very little performance difference from normal. At best we were able to muster only 1.5% plus or minus, which in the area of benchmarks is once again statistically insignificant. Hopefully our next title will prove more of a challenge.
Supreme Commander is a very large, overarching RTS game. Though it may be a DX9 title, many of the effects within the game are DX10 inspired if not driven. Due to the nature of this game, it brutalizes your CPU, GPU, memory, and hard drive. This make it’s a great testing platform for any upstart CPU. Since we’re trying to squeeze out some good numbers for our charts, we decided to once again crank the settings to maximum, apply some 4x anti-aliasing loving, and ramp up the resolution to 1680×1050. As for the game itself, Supreme Commander has an internal benchmark that plays out a full battle scenario for you in real time. This benchmark measures all the aspects of your system perform, and produces a composite score not unlike 3DMark06 and PCMark05.
The numbers for our final benchmark definitely followed the trend started by our first benchmark. There was a performance increase when the processor was overclocked as far as it could, and at 7% it was definitely close to that afore mentioned percent-for-percent goal of 8% I’m looking for. As for the TLB erratum fix, when enabled it resulted in a whopping 28% drop in overall performance in game. This is just staggering, and it caused me to rerun the benchmark several times to be sure. Unfortunately it was true, as I could see the lag in the benchmark as it was happening.
Real-Time Price and Stock Check – Shop Like a PRO!
Final Thoughts and Conclusion
At the beginning of this review I said something about AMD being the scrappy underdog in the CPU market. Well the effort put into the Phenom 9600 Black Edition has definitely shown they aren’t going away, and are continuing the fight. Those characteristics are rather redeeming, and to many in the enthusiast market they have a certain charm. But as charming as they may be they can’t mask the cold hard truth, and that is that this part, in its current iteration, still can’t quite compete with Intel’s equivalent.
It’s really too bad to, as AMD has been dogged by problems since the release of the Phenom line. First was the delays and parts shortages. Then the TLB erratum came out, and the fix bringing down performance didn’t help. I’m glad to see AMD release an off switch of sorts in OverDrive to disable the TLB fix. They have also gone to great lengths to make sure people are aware of the potential issues from disabling the fix, though I have yet to experience any problems from running with the fix disabled full time. They are definitely soldiering on.
I could give AMD props for releasing an unlocked processor at such a tantalizing price point. The only problem with that is the feature is rendered unimportant by the sheer refusal of the Phenom 9600 Black Edition to be overclocked more than a token amount. This is even more un-nerving when Intel has a quad core processor at the same price point that performs better at stock and overclocks better.
There’s no way to sugar coat this, so I’ll be blunt in saying that it’s hard to recommend the Phenom 9600 Black Edition for an overclocker system. The processor doesn’t perform as well as its contemporaries and doesn’t offer any additional value over them. AMD’s efforts are to be commended, as the Phenom and the Spider platform are part of are good starting point. I just think the processor side of things needs to mature a little more before they can really think about taking down Chipzilla.
As of this review is hitting the web, AMD has trickled out the new B3 revision CPU’s. Hopefully we’ll be able to get our hands on one of these new CPU’s to see if some of our grumblings about overclocking headroom and the performance hit on the TLB fix have been rectified.
The Good
- True single die quad-core
- Unlocked multiplier
- Good price point for unlocked part
The Bad
- Stock performance is rather ho-hum
- Performance takes a significant hit when TLB fix is enabled
- Overclocking ability is mediocre at best
- Doesn’t stack up well against competitors
Overall Rating: 7.0/10.0
Discuss This Review in the Futurelooks Community Forums
Real-Time Price and Stock Check – Shop Like a PRO!