Original Link: http://www.anandtech.com/show/2233
ATI HD 2900XT CrossFire: Intel 975X versus Intel P35by Gary Key on May 16, 2007 12:00 PM EST
- Posted in
A lot of you are just like we when it comes to new product. You cannot wait to get your hands on the latest and greatest technology when it's released. Early previews of the next great technology are shown and shortly afterwards performance numbers are leaked on various forums. These numbers almost always show the product in its best possible light and performance records are consistently being broken. User hysteria kicks in, credit card limits are extended, fastest shipping routes are checked and rechecked, you beg your significant other for purchase approval, and then a wave of anticipation builds until that fateful day the product finally arrives.
In the meantime, you sneak a peek at the latest forum posts about the product throughout the day instead of finishing that analysis, presentation, class assignment, or uATX article that was due by 8 am. The marketing pitches continue in waves with professional pictorial layouts, more earth shattering performance numbers are leaked, rumors spread about the product curing cancer and saving the earth - never mind if any of the information can be substantiated; for goodness sakes it's on the Internet so it must be true!
And then the day comes when previews or reviews of the product start showing up at various websites or user reviews start coming in from those lucky enough to live near the first reseller to have the product in stock. All of a sudden those early results from people already praising the next great product are not starting to look like reality. Doubt sets in if you made the right decision, but more importantly did you just exhaust your last kitchen pass with the significant other or miss an outing of a lifetime with friends to purchase what now appears to be a fairly ordinary product?
You shrug it off; after all, those product reviews cannot be correct as the reviewer missed something or the setup was incorrect (editor's note - believe me we are not always perfect but we do admit when we are wrong). Worse yet, they must be on the payroll of the competitor as there is no way those early forum results could have been so far from the truth. The thought process continues - those actual users with the product reporting less than stellar results are probably just newbies' who have no idea how to setup much less use a computer system.
It is at that point the doorbell rings and you hug your favorite delivery person as the anticipated package has arrived. It's time to set the world straight. The product is installed, everything is setup perfectly, and wham bam... the product fails to deliver on those early promises. You try everything, you desperately ask for technical assistance from those who posted the early glowing reports only to find out they are already singing the praises of the product that will replace the one you just bought.
Sound familiar? We have all been through it at one time or another. Some of us (clears throat) are even guilty of getting caught up in the hype machine at times. In fact, it happens more often than not it seems with major new product releases. While our example is usually the worst case scenario, we are starting to see more and more stories like this. The rapidly growing number of emails in our inboxes asking for additional information or test results is a testament to the current trend of early product previews in the forums. This is not necessarily a bad trend or practice as we at times find out about new products before the manufacturer has even notified us.
Our interest and curiosity is right up there with others when it comes to seeing how well a new product performs before release along with what new features will be available. In fact, we sometimes wish that we could post our first results in the forums to give our readers a first look at the product, only outlining the positive and negative aspects about our early experiences instead of just glowing marketing speak. Alas, the NDA gods will usually not allow it, although we are starting to see some wiggle room by the manufacturers in this area.
You might be wondering what any of this has to do with our article today but we find it to a be an excellent segue into our initial CrossFire test results with the new ATI HD 2900XT GPU and Intel P35 chipset. We are posting early results so you can have some additional information and/or alternate perspective before making a significant purchase decision. We also hope this answers questions we have received about CrossFire performance on the P35.
If you have decided like us to take a leap of faith in ordering a pair of ATI HD 2900XT cards for a CrossFire setup then the information that follows might be of interest to you. If not, it might still provide an example of what happens when a product is released to a different set of expectations. After going through our checklist which includes ensuring we had the proper OPEC friendly power supply and PCI Express 2.0 (8-pin) connectors, additional A/C cooling, noise canceling headphones, and a bevy of alpha and beta drivers, we decided it was time to see how these newly released cards perform in CrossFire for our upcoming P35 chipset article. Let's take a look at the early results and see who gets knocked out in the first round.
Our test configurations today consist of the ASUS P5K-Deluxe sporting the new P35 chipset and the Intel D975XBX2KR based on the venerable 975X chipset. Our retail P5K-Deluxe board was purchased recently even though an embargo is supposedly in place until June 4th for distribution of P35 product. Likewise, P35 boards from other suppliers such as MSI and Gigabyte can also be purchased at this time making this one of the stranger product releases in recent memory.
The P5K-Deluxe features ASUS's C.G.I. technology from their P965 motherboards. ASUS C.G.I. stands for ASUS Cross Graphics Impeller (marketing still reigns) and is a feature that when enabled will automatically optimize system performance if a CrossFire configuration is detected. These optimizations occur within the Direct Media Interface between the P35 MCH and ICH9R that is utilized to enable CrossFire operation on this motherboard.
The 975X chipset utilizes Peer-to-Peer write capability within the MCH to enable 2x8 PCI Express lane capabilities for CrossFire. This feature is not available in the P965 or P35 without a special PCIe controller chip and BIOS support. ATI/AMD enables CrossFire support utilizing the Direct Media Interface (DMI) to link the x16 GPU slot (16 PCI Express Lanes) residing on the MCH and the x4 GPU slot (4 PCI Express Lanes) residing on the ICH. Contrary to rumors and initial reports in certain forums, the P5K-Deluxe does not perform CrossFire operations with a 2x8 PCI Express lane configuration.
|Standard Test Bed
CrossFire Test Configuration
|Processor||Intel Core 2 Duo QX6700
(2.66GHz, 8MB Unified Cache)
|RAM||OCZ Reaper PC2-9200 (4x1GB) 2.32V, 3-3-3-9 975X, 4-4-3-6 P35|
|Hard Drive||Western Digital 150GB 10,000RPM SATA 16MB Buffer|
|System Platform Drivers||Intel - 184.108.40.2063|
|Video Cards||2 x MSI HD2900XT|
|Video Drivers||ATI 220.127.116.11 (HD2900XT Release Drivers)|
|CPU Cooling||Tuniq 120|
|Power Supply||OCZ ProXStream 1000W|
|Optical Drives||Plextor PX-760A, Plextor PX-B900A|
|Case||Cooler Master CM Stacker 830|
|Motherboards||Intel D975XBX2KR (Intel 975X) - BIOS 2692
ASUS P5K Deluxe (Intel P35) - BIOS 0304
|Operating System||Windows Vista Ultimate 64-bit|
Test conditions were maintained the same, as much as possible, over the platforms tested. Our game tests were run at settings of 1280x1024 4xAA, 1600x1200 4xAA, and 1920x1200 4xAA with 8xAF implemented in games that support this feature. These settings were used on both our single card and CrossFire setups. We feel like these settings and resolutions will provide accurate benchmark results for the typical user utilizing a CrossFire setup with a high end CPU.
All results are reported in our charts and color-coded for easier identification of results. We utilize new drive images on each board in order to minimize any potential driver conflicts. Our 3DMark results are generated utilizing the standard benchmark resolution for each program. We run each benchmark five times, throw out the two low and high scores, and report the remaining score. All results are run at stock speeds for this article although we will provide overclocked results in the next article. For those wondering, our cards generally had no issues running at 853/1000 provided we had notified the electric company of a pending power surge.
This preview is not a graphics card review and as such we are not including results with products from the Big Green Machine yet. Those comparisons will come in our P35 chipset article. We are simply providing results on how each chipset handles CrossFire operations at this time. We will provide P965 and RD600 results in our follow-up to this article so you can have a clear picture of which Intel chipset performs the best with a CrossFire configuration. We might even throw an RD580 into the mix to see how well the R600 performs on it.
We also booked several sessions with a psychologist so we can understand why there was a lapse in our thought process for choosing Windows Vista Ultimate 64-bit as our operating system. The R600 already has enough early driver issues to make one think twice about using it, but to throw a new operating system and chipset into the mix and then to push matters further by going 64-bit was clearly not the actions of a sane person. It sounded good at the time, it really did, but after several days of constant frustration, hair pulling, dog kicking (relax PETA, just a joke), finger nail chewing, and general panic attacks about missing the article deadline... well, we would have not have done it any differently as it turns out.
Why? Whether we like it or not, Vista is the future of Windows for the time being and is required for DirectX 10. Honestly, it was time to see how far the various vendors had come since release in providing decent driver or game support. 64-bit OSes are also the future - after all, AMD released x86-64 on the world over three years ago. We collected enough information to generate a weekend short story on the subject but as we feared, progress has been slow.
NVIDIA released their first decent set of Vista drivers this last week and we are busy redoing all of our 8800GTS/GTX numbers for the P35 launch article. In the meantime, we chewed through four different driver releases from AMD and decided to stick with the publicly released 18.104.22.168 drivers for this article. We generated some really impressive 3DMark numbers with the alpha 22.214.171.124 drivers but let's just say when it came time to using actual applications those drivers were not always stable or feature capable. We did receive a new set of beta 8.38 drivers a couple of days ago and those are in testing but we do not have enough experience with them yet to publish meaningful numbers.
You might notice in our game testing that several of the more popular games are not benchmarked. We had screen corruption issues in Oblivion, S.T.A.L.K.E.R., Half-Life 2: Episode One, and even Sims 2 when utilizing CrossFire. These same issues are not evident under Windows XP so we contribute most of the issues to driver maturity, though several games we tried are also having some minor issues with XP as well. Also, our Battlefield 2142, Flight Simulator X, and Half-Life 2: Lost Coast benchmarks would not run consistently under Vista so we are back to the drawing board on those and a couple of other games.
As for providing current DX10 benchmarks from the upcoming Lost Planet and Call of Juarez games we decided it was best to wait on the next driver release before providing results as any scores generated now are basically useless. When running CrossFire with the R600 each demo has problems with rendering, tearing, jitters, and several other issues that are likely to be fixed shortly. Needless to say, our first experiences with DX10 and the R600 were not pleasant.
General Graphics Performance
The 3DMark series of benchmarks by Futuremark are among the most widely used tools for benchmark reporting and comparisons. Although the benchmarks are very useful for providing apples-to-apples comparisons across a broad array of GPU and CPU configurations, they are not a substitute for actual application and gaming benchmarks. In this sense we consider the 3DMark benchmarks to be purely synthetic in nature but still valuable for providing consistent measurements of performance.
Well, the results of this test are confusing on the surface although driver maturity and memory sensitivity across the DMI interface has a great deal to do with the P35 results. The P35 chipset scores about 2% better in single card operation than in CrossFire mode and also leads the 975X CrossFire setup. We found in testing that the P35 CrossFire scores in each scene were slightly higher until the Nature test where the single card scored about 12% better. The 975X CrossFire setup just barely ekes by its single card performance results. This benchmark is currently a better indicator for CPU, chipset, and memory performance. In this regard, we can see that the P35 single card performance leads the 975X slightly in platform performance as our game benchmark testing will indicate shortly. The fact is, in unbuffered memory testing the P35 was generally about 5% faster than the 975X across the board with CPU throughput testing being higher with a quad core processor.
The DirectX 8 centric tests in 3DMark05 benefited greatly from the improved chipset throughput performance of the P35 chipset at stock settings with our quad core processor. The P35 CrossFire results are up to 7% faster than the 975X results with the single card P35 setup once again finishing ahead of the 975X CrossFire setup. Although we have found the P35 chipset to be a fierce competitor to the 975X in initial testing we think some additional BIOS and driver tuning would allow the 975X performance to improve by a few percent in these tests.
In our more strenuous graphics test utilizing 3DMark06 we find the P35 results once again leads the 975X chipset but the margin of difference is a negligible 1~2%. We decided to see why the results were so close in this particular test. We looked over the results and found in the SM2.0 tests the P35 solution was about 2% behind the 975X scores, the P35 CPU score was slightly better, and the HDR/SM3.0 tests showed a 4% advantage for the P35. Since the HDR/SM3.0 tests heavily stress both the CPU and graphics bus we figured the x4 PCI Express lane limitation would cause a bottleneck in this test.
Our initial assumptions turned out to be incorrect. After working with ASUS we discovered in their internal testing they noticed the same issue, and they decided to see what would happen on the 975X if the MCH was programmed at x16/x4 operation between the two GPU slots instead of x8/x8. Their test results revealed a surprise as the difference in throughout in all areas of testing was less than 1%. The issue lies in the limited bandwidth and speed of the Direct Media Interface between the P35 MCH and ICH9R. The time required to simultaneously move the data between the two chipsets imposes a significant overhead and bandwidth issue in memory sensitive applications, hence our issues in the memory sensitive 3DMark01 benchmark. Of course, 3DMark performance doesn't necessarily have anything to do with actual game performance anyway, so these results are only mildly interesting.
While ASUS has optimized this link and will continue to do so, it appears we are now near the maximum efficiency of this interface. This simply means that as games become increasingly complex and data bandwidth increases then the differences between the P35 and 975X in CrossFire operation will widen. Let's see how this potential issue and driver maturity affects our initial gaming benchmarks. We would like to stress once again that synthetic benchmark results do not necessarily correlate into real application results.
As usual, gaming performance was tested with a variety of current games. We ran benchmarks with our standard 1280x1024 resolution with 4x antialiasing and with 8x anisotropic filtering (if the game has support) enabled. Given the number of users that run 19" LCDs these days, 1280x1024 represents one of the most commonly used resolutions. We decided to also stress the graphics subsystem since we are benchmark testing the core logic chipset to compare the performance of ATI CrossFire on the P35 and 975X platforms.
In order to do this we were particularly interested in increasing the resolutions and graphic settings of the graphics cards so we are including 1600x1200 4xAA/8xAF and 1920x1200 4xAA/8XAF resolutions. We feel these resolutions will be the best indicator of performance for users with a high-end CrossFire and CPU setup. Our review is based on the capability of the Intel P35 chipset and how well R600 CrossFire works on the ASUS P5K-Deluxe with the current 126.96.36.199 drivers. We will examine other GPU solutions with this chipset in our P35 article.
This benchmark is performed using DICE's built-in demo playback functionality with additional capture capabilities designed in house. When using the built-in demo playback features of BF2, frames rendered during the loading screen are counted in the benchmark. In order to get a real idea of performance, we use the instantaneous frame time and frames per second data generated from our benchmark run. We discard the data collected during the loading screen and calculate a result that represents actual game play. While DICE maintains that results over 100fps aren't always reliable, our methods have allowed us to get useful data from high performing systems.
During the benchmark, the camera switches between players and vehicles in order to capture the most action possible. There is a significant amount of smoke, explosions, and vehicle usage as this a very GPU intensive Battlefield 2 benchmark. We run Battlefield 2 using the highest quality graphics settings available in the video settings. The game itself is best experienced with average in-game frame rates of 40 and up.
The P35 CrossFire solution performs well at 1280x1024 but as we crank up the resolution our scores start to drop and end up being about 5% worse at the 1920x1200 resolution. We still found the P35 setup to be perfectly playable and did not witness any stutters or graphic anomalies during testing. We noticed the same results in our initial XP testing and since the single card results are actually better than the 975X we believe this issue is a limitation of the x16/x4 design. However, we firmly believe that driver optimizations could regain a couple of percent based on P965 testing with an X1950 XTX setup.
Serious Sam 2
This benchmark is performed using Croteam's built-in demo capability in the Serious Sam II engine. We utilize the included Branchester Demo and capture the playback results using the Ctrl-~ function. The benchmark features a large number of combatants, explosions, and general mayhem. The benchmark is primarily GPU sensitive with the actual percentage of GPU/CPU/Audio activity being displayed during the benchmark run. We typically find this game is very playable at average in-game rates of 60 and above. We maximize all settings except antialiasing and anisotropic filtering within the general and advanced video settings.
We see our P35 CrossFire setup performing 5% better at 1280x1024 and then following the same pattern in BF2 with it trailing the 975X platform by 10% at 1920x1200 as the game becomes GPU limited. As in the Battlefield 2 testing the game playback was perfect and generated all of the correct visuals. Considering the single card scores favor the P35 setup we once again believe the differences at the higher resolutions are due to a combination of driver maturity and the x16/x4 limitation in high data traffic games. The initial XP testing reveals the same pattern although the difference at 1920x1200 is 5% which leads us to Vista driver improvements still being required for CrossFire operation on the P35.
F.E.A.R. uses a built-in performance test that generates graphical test scenes based upon the actual game engine. This test consists of a couple of different action sequences, a stressful water flyby, and heavy use of shadows while traveling through hallways. F.E.A.R. is a very graphics intensive game and we switch all settings to maximum for both the system and GPU. During our testing of F.E.A.R., we noted that the "soft shadows" don't really look soft and the performance hit is drastic, so we disable this setting. An average frame rate for F.E.A.R. can dip drastically during game play and that is not good for a first person shooter but the game is still playable around 35fps although we prefer a solid 45fps.
This game is still a GPU crusher when the settings are dialed up and our P35 CrossFire setup falls flat on its face in this game when compared to the 975X. The single GPU scores are indicating another issue with both driver optimizations and throughput issues on the DMI link. The minimum and maximum frame rates followed the same pattern and will be presented in the P35 chipset article. Our initial testing on the P965 platform revealed a minor difference in frame rates between the P965 and 975X chipsets so we think driver optimizations can close the gap in this game.
Our benchmark utilizes the IdNetDemo with the playnettimedemo option. This includes mainly outdoor areas with numerous players trying to kill each other. We tested the game with Ultra Quality settings (uncompressed normal maps), and we enabled all the advanced graphics options except for VSync. Id does a pretty good job of keeping frame rates consistent so in-game frame rates above 30 are acceptable for single player and 60 for multiplayer. The important thing to remember is this test will directly translate to an actual Quake 4 experience.
After seeing a consistent pattern in our other game tests we now see a complete flip in the CrossFire results. We see the P35 setup being around 8% faster on average across the CrossFire results although it loses in the single card results. We ran this test several times with different settings but the results were always the same. We attribute the differences to driver optimizations and game code for the most part as the single card scores are better than the CrossFire scores.
During actual game play the perception was the CrossFire setups were more fluid and "seemed" faster than the single card setups although the net timedemos said otherwise. We also ran our custom timedemo and the scores were just the opposite with the 975X scoring up to 12% better. However, this was one of our benchmarks that had issues consistently finishing under Vista and showed signs of rendering corruption during playback. At present, we would have to say that our Q4 test is almost completely CPU limited, and this is often not the case during actual gaming. We are working on a new benchmark at this time.
Gaming Performance, Continued
Supreme Commander is one of the better RTS games to be released in recent memory although we are still huge fans of Command and Conquer 3 along with Company of Heroes. We chose Supreme Commander as it is both a GPU and CPU hog when it comes to systems resources. We utilize the built-in performance test to benchmark the game. We set all of the settings to high and only change the resolutions between benchmark runs. This benchmark provides a cornucopia of results but for our tests we will report the average frame rates during the benchmark. We generally find this game to be playable with frame rates at or above 35fps.
We witnessed stutters and graphic anomalies during CrossFire testing with the P35 setup along with the fact that the scores were generally lower than either of our single card setups. We attribute this problem to driver optimizations and/or BIOS issues with the R600 and P35 as our X1950XTX testing did not reveal the same pattern with this setup. This is yet another game where CrossFire support seems to be broken, at least with the P35 platform and the R600.
Company of Heroes
Company of Heroes was released last year and is still proving to be a very addictive RTS game around the office. The game is extremely GPU intensive and also requires a hefty CPU at times. If this is beginning to sound a lot like Oblivion, well CoH is very similar to Oblivion in system requirements. The visuals and audio experience within the game will at times have you believing the game is based more on a First Person Shooter than a traditional Real Time Strategy game. We set all options to High and turn on all additional video options.
The game contains a built-in performance test that utilizes the game engine to generate several different action scenes along with a coffee argument as a sideline distraction before the war starts. We found the performance test gives a good indication of how well your system will perform throughout the game on average. We have found some of the in-game action sequences to be more demanding than the performance test and are working on game play benchmark that is repeatable. We generally found the game to be enjoyable with an average frame rate above 35fps.
We see the P35 CrossFire solution stinking up the office in this benchmark although the single GPU scores favor it. The P35 setup is running about 15% to 20% behind the 975X platform. We did not notice any stuttering during the performance test or actual game play but we did notice banding on the P35 setup that was not present in the same scenes on the 975X setup leading us to believe there is a definite driver issue. However, the spread between the 975X and P35 in backup testing with the X1950XTX CrossFire setup still showed a 7% to 9% difference in performance under both Vista and XP. This indicates the x16/x4 link is probably saturated by this game and leads to a noticeable performance difference. We will see how well the next set of drivers address these problems.
Prey offers some superb action sequences, unique weapons and characters, and is a visually stunning game at times. It still requires a very good GPU to run it with all of the eye candy turned on. We set all graphic settings to their maximum except for AA/AF and utilize a custom timedemo that takes place during one of the more action oriented sequences. We generally found the game to be enjoyable with an average frame rate above 35fps.
Once again the P35 CrossFire solution just cannot compete with the 975X solution. Although the frame rates are still very good the P35 setup is about 20% to 30% behind the 975X platform. Game play and visuals were excellent throughout testing but we did notice a few slowdowns during level changes that were not present on the 975X platform. The spread between the 975X and P35 in backup testing with the X1950XTX CrossFire setup still showed an 11% difference in performance under both Vista. This indicates the x16/x4 link is again saturated to some degree although driver optimizations should improve performance notably.
And there you have it: the good, the bad, and the ugly when it comes to CrossFire performance on the P35 chipset with the HD 2900XT. There is nothing really revolutionary or surprising to us in the initial results after spending some significant time testing both products. We will say that the P35 launch should go smoothly when compared to the P965 launch last year if the first wave of motherboards is any indication of what is to come. Our ASUS P5K-Deluxe motherboard has been a joy to work with over the past few weeks and the retail unit has not changed our initial opinion of the board. While not perfect, it is definitely a step forward for ASUS in several areas when compared to the P965 family. Expect to see our first review on this board and others on Monday.
P35 may be good so far, but the same can't be said of R600. We fully expect to see performance and quality improvements over the coming weeks as the R600 drivers continue to mature and AMD pushes the performance envelope. How much actual applications like games will benefit is anyone's guess at this time but early results with the latest beta drivers show impressive gains in the 3DMark applications if that's important to you. However, what is not mentioned in most of these early results is continuing issues with CrossFire and OpenGL operation under Vista along with IQ issues in games, not to mention certain video decoding and playback functions not being up to full speed. Don't get us wrong, though: for as much grief as AMD has endured or deserved for this product launch we still like the card and expect to see more of it in upcoming product reviews. (Yes, some of us are eternal optimists that figure performance and quality will come up to speed and prices will drop.)
We are not here to single-handedly knock AMD for these issues as it has taken NVIDIA close to six months to get decent Vista drivers out for the 8800 series of cards and we are still waiting on new platform drivers. We could continue the listing of suppliers that have not stepped up to the plate yet in providing stable drivers or applications for Vista 64, even though everyone has known for the last three years that the industry would be progressing to 64-bit operating systems. At present, it's hard for us to determine what caused more problems in our testing: the new products that still need optimizations and tuning or an operating system that just does not seem quite finished yet.
It seems getting new hardware products to work together seamlessly is something that no longer is possible to do upon product launch. Maybe we propagate that problem by not being more direct in our conversations with the suppliers during the test phase of new products, but we do typically tell it like it is and those suppliers who listen take it to heart as you will see with the P35 products. It could also be that a few of the suppliers are fooled into believing their own marketing spins when providing engineering test results or products for those early previews that read like a bad infomercial. In the end, we need less hype, better products, and clear and concise information about the products' strengths and weaknesses on a timely basis.
Overall, our game testing indicates that driver optimizations will be paramount to improving CrossFire performance on the P35 chipset when utilizing Vista. Due to the x16/x4 limitations, the expected performance capability of the R600, and a whole new generation of games coming out over the next few months that promises to pound your current system into submission at the highest settings, we cannot recommend the P35 at this time for a CrossFire capable system. That honor still belongs to the chipset that refuses to die, the 975X which is slated to be replaced this summer with the new X38 chipset. While not always the fastest solution for an R600 CrossFire platform it does offer a level of consistency and stability that is not always present on the P35. We fully expect the P35 to perform significantly better in several areas once AMD has the time to tune their drivers for it. For now we understand and accept this lack of optimization, but that does not mean we like it.