Original Link: http://www.anandtech.com/show/1180
Fall 2003 Video Card Roundup - Part 3: ATI's Radeon 9600 XTby Anand Lal Shimpi & Derek Wilson on October 15, 2003 10:26 AM EST
- Posted in
Fourteen days ago we introduced our brand new GPU test suite composed of a total of 18 games, and as shocking as it may be, we tested with more than first person shooters. Unfortunately we launched the new test suite on quite possibly the least important set of cards for such a suite the ultra high end $500 solutions from ATI and NVIDIA. Ever since the release of the Radeon 9700 Pro we have not had a reason to recommend any $400+ card simply because none of todays games truly need the kind of power offered by those cards. The Radeon 9700 Pro (and the modded Radeon 9500 Pro) was an excellent solution that could all of the games out at the time at extremely high resolutions, with antialiasing and anisotropic filtering enabled. It was the release of the Radeon 9700 Pro that forced us to start testing with 4X AA and 8X anisotropic filtering all over the place in order to truly stress the beast of a card.
Since the release of the Radeon 9700 Pro however, games have not become any more demanding. The titles that successors like the Radeon 9800 Pro and NVIDIAs GeForce FX 5900 Ultra were built for, have still yet to be released. The battle between $500 cards will occur with titles like Doom 3 and Half Life 2, both of which wont see the light until next year. This holiday season should bring a few more stressful DirectX 9 titles to our hard drives, but for the most part, weve found it silly to recommend purchasing any of the ultra expensive cards until a game you want to play comes out that requires $500 worth of GPU. Thus, for the most part, introducing a comparison of todays most popular games did little more than expose driver bugs and show that a lot of games are CPU bound when youre running a $500 card.
The real comparison starts today, but it wont end until both ATI and NVIDIAs cases have been made later this month. The comparison we have in front of us now is amongst much more affordable cards, and most definitely cards that you would buy for their performance in todays games not for their promise of sunny days tomorrow. The cards were talking about are aimed at that magic $200 price point and given that its the fall, its time for a refresh of the cards in this segment.
The Radeon 9600 XT is ATIs $199 successor to the Radeon 9600 Pro and it is their fall refresh product for the mainstream market. Today Radeon 9600 XT will be paired up against NVIDIAs GeForce FX 5600 Ultra, but later this month we will be able to bring you comparison of the 9600 XT and the new 5700 Ultra, which NVIDIA has been quite confident in as of late.
Before we get to the tests, lets talk about whats changed with the 9600 XT
The definitive Fall Refresh
After NVIDIA released the TNT2 Ultra, we saw the first incarnation of the now common 6-month product cycle. The strategy was the exact one used to dethrone 3dfx, and is based on a very simple principle of using parallel design teams. If you have three design teams, one working on the current generation product, one working on the 6-month follow-up and one working on the next-generation solution, assuming all teams work efficiently, you should be able to maintain a stream of GPU releases in 6 month intervals.
To make the job a bit easier, you only work on inventing new architectures every 12 months, giving you a little break in between the hectic lifestyle of a GPU design engineer. But in order to maintain competitiveness you have to have a product every 6 months, so in the time between architectures you simply refresh your current generation architecture. A refresh is generally a higher clocked GPU, potentially with some faster memory, made possible due to more experience with manufacturing that particular GPU (yields improve over time) and the availability of faster memory. Sometimes we get advancements in process technology that allows for a boost in clock speed as well.
When NVIDIA introduced the 6-month product cycle the idea was that new architectures would debut in the Fall, and refresh products would hit in the Spring. The delay of NV20 (GeForce3) changed things around a bit and the GeForce2 Ultra became the first Fall refresh product. Since then, little attention has been paid to when various GPUs hit, as long as we get something new every 6 months were happy. Earlier this year we heard that both ATI and NVIDIA would be releasing their true next-generation hardware next Spring, leaving this Fall as the refresh cycle.
ATIs high-end refresh was the Radeon 9800 XT, and as you can guess their midrange refresh is the new Radeon 9600 XT. Much like the Radeon 9800 XT, the 9600 XT only adds two features: a higher clock speed and support for OverDrive.
The Radeon 9600 XT GPU now runs at 500MHz, a 25% increase in clock speed over the 9600 Pros 400MHz clock. The memory speed of the Radeon 9600 XT remains at 300MHz DDR (effectively 600MHz), so there is no increase in memory bandwidth over its predecessor.
The hefty increase in clock speed is due to improvements in process technology as well as the introduction of a low-k dielectric. As we briefly explained in our 9800 XT review, the benefits of a low-k dielectric are mainly related to shielding from crosstalk in high transistor density chips, which gives us the clock speed boost we see with the 9600 XT. Because were just talking about an increase in core clock speed, the games to receive the biggest performance boost from the XT would be those that are GPU-limited, which unfortunately are few and far in between these days. Games that are largely shader bound such as Half Life 2 will definitely enjoy the 9600 XTs increase in clock speed, but for now well see most of the performance benefits go to waste.
We explained OverDrive technology in our Radeon 9800 XT review and tested it in our Catalyst 3.8 driver update. The Radeon 9600 XT includes an on-die thermal diode that measures the temperature of the core; when the temperature is cool enough the driver will instruct the core to overclock itself by a set margin. The Radeon 9600 XT will run at one of three speeds depending on its temperature: 500MHz, 513MHz or 527MHz. The combination of this driver and hardware support makes up ATIs OverDrive feature.
OverDrive is currently not enabled for the Radeon 9600 XT in the Catalyst 3.8 drivers, we will have to wait for the Catalyst 3.9s before we can test the 9600 XT with OverDrive. If youre curious about the performance implications of enabling OverDrive, have a look at our Catalyst 3.8 review its nothing to get too excited about.
We used the exact same test bed and settings as our Catalyst 3.8 review, the only difference here was that we benchmarked at 1024x768 given the power and target market of the Radeon 9600 XT. We also used the "almost final" version of the 52 series Detonators from NVIDIA (52.16) which have been submitted for WHQL certification.
As a refresher, here are the games we benchmarked with:
C&C Generals: Zero Hour
EVE: The Second Genesis
F1 Challenge '99-'02
Final Fantasy XI
Jedi Knight: Jedi Academy
Neverwinter Nights: Shadow of Undrendtide
Tomb Raider: Angel of Darkness
Unreal Tournament 2003
Warcraft III: Frozen Throne
Wolfenstein: Enemy Territory
X2: The Threat
Our testbed remained the same:
AMD Athlon64 FX51
1GB DDR400 (2x512MB)
ASUS nForce3 motherboard
The only issues we encountered were as follows:
1) Homeworld 2 would not run on either the Radeon 9600 Pro or the Radeon 9600 XT. This is the same issue we ran into the first time we tried to run this benchmark on ATI hardware. Interestingly enough, it works on all of ATIs high end cards just not their midrange hardware.
2) Tomb Raider would not run on the GeForce FX 5600 Ultra with the latest 52.16 drivers at 1024x768. The game kept on returning an out of memory error at any resolution higher than 1024x768. Given that we tested with a 128MB card and none of the other cards had a problem, this seems like more of a driver issue or a game issue than anything else.
3) Since the GeForce4 Ti 4200 only supports PS1.1, we could not include this card in the Tomb Raider tests either. Performance under PS1.1 is much higher than performance under PS2.0, so the comparison would not be fair to ATI if we ran all of their hardware using PS2.0 and ran the Ti 4200 in PS1.1 mode.
We have not had time to go back and figure out a benchmark for BF1942 yet nor work out the issues with Microsoft Flight Simulator 2004. We havent received much feedback in terms of any ideas for benchmarking under these two games, but were still open to suggestions.
For image quality comparisons refer back to our Fall 2003 Video Card Roundup - Part 2.
With that out of the way, lets get to the games.
Despite what some people would like everyone to think, Aquamark3 is really a test of how people developing software now envision DirectX 9 pixel and vertex shaders will be used in the future. The situation is very reminiscent of the first Sony PlayStation: the first games that used the technology were limited by the hardware until developers really learned to work with the hardware rather than on the hardware. As time progressed, we went from what were essentially ports of 16bit console games to amazingly complex and beautiful games like Gran Turismo 2. The same thing will happen with shader technology, and no amount of guessing and throwing functions at a GPU will tell you how its performance will really be in the future. Essentially, my advice is that any piece of software that claims it is a valid predictor of future performance should be taken lightly. We based our decision to include Aquamark3 on its popularity in the community. Aquamark3 is a cool piece of software, with some pretty neat tests, and a high score in any benchmark can still earn bragging rights in the forums. The only Aquamark3 test we ran was the publicly available 1024x768 4X AF no-AA in order to maximize the usefulness of these numbers to the community. Our drivers were set to allow application control of AF and AA.
The 9600 XT makes some good gains over the 9600 Pro in Aquamark. For a 25% increase in clock speed, we get more than a 17% increase in frame rate. We still aren't close to the 9700 Pro, but those kinds of numbers are not bad by any stretch.
We can attribute the solid increase in frame rate under Aquamark to its extensive use of shaders which make performance more defendant on core clock speed than any thing else.
All the other cards in this benchmark fall where we would expect them.
C&C Generals: Zero Hour Performance no AA/AF
The recently released expansion to the very popular Real-Time Strategy (RTS) game Command & Conquer Generals seems to do a good job of living up to the standards set by its prestigious ancestry. RTS games usually get overlooked in GPU roundups and comparisons as they aren't considered graphically intense. However, smoothness is very important to gameplay; goodness knows I've blamed plenty of lost armies on ill timed drops in framerate. For this benchmark, we created a multiplayer game consisting of 6 hard armies on one team with us, and one easy army. We then used the replay feature in conjunction with FRAPS to measure performance. This was done with and without 4X AA/8X AF
At 10x7, with no AA/AF, C&C is CPU limited, and the performance of the ATI cards is all but identical. The NVIDIA cards trail the ATI cards by about 20%, and they seem CPU bound as well.
C&C Generals: Zero Hour Performance 4X AA/8X AF
We can see some separation in the ranks when we turn on up the juice. Basically, we see the RV3x0 based cards fall off in performance to where the NVIDIA cards were without AA/AF.
The NVIDIA cards drop off in performance as well, but the 5600 Ultra is able to maintain its speed a little better than the Ti 4200.
EVE: The Second Genesis Performance no AA/AF
EVE is one of our new additions and was chosen as a representative of the MMORPG group of games. Of course, this game is a bit different than most MMORPGs, but the advantage with this game is that one of the most graphically intensive aspects of the game is just sitting in a space station all alone while having a bunch of transparent windows open to manage your character.
Of course, the first thing Anand said about this game was "Oh look, it's the Linux desktop!" And, to be honest, it really does look more like my visions of Longhorn than any other game I've ever played.
This game is was very slow on my computer at home, so these new cards really do have a high impact on performance.
EVE: The Second Genesis Performance 4X AA/8X AF
F1 Challenge '99-'02 Performance no AA/AF
We've had plenty of requests to benchmark with a racing simulation. When we were designing the new testing suite this was one of the first games that came to mind. The game is faced paced, has lots of graphics options, and could keep someone who is into F1 racing busy for weeks at a time. Combine all of that with a nice replay feature and we have a very useful benchmark. We just ran a lap at Australia and counted the framerate of the replay via FRAPS while following one of the drivers in the middle of the pack.
We are still seeing the same jittery action with the NVIDIA cards that we have noted in Parts 1 and 2, and the AA also falls a little short of what it should be doing. As we have stated previously, these are known issues, and they are being addressed.
We see no real performance gains from the 9600 XT under this game, and we know we aren't CPU bound as both the 9700 Pro and 9800 XT show performance gains over the RV3x0 cards.
F1 Challenge '99-'02 Performance 4X AA/8X AF
We can see that the 9600 cards are able to hold their speed a little better than the NVIDIA cards when AA and AF are enabled.
Final Fantasy XI Performance
The Final Fantasy series has always been popular, and will probably always be popular. I have a personal affinity towards the series, but I'm not so sure about the Massively Multiplayer Online style they are going for this time around. The benchmark is much more straightforward than the success of their future game: we set it to high-res benchmark mode (1024x768) and record the number of frames that get rendered (in the bottom right hand corner). Out of reader requests we are now reporting the total number of frames rendered instead of dividing it to get a fps value:
A 10% to 15% gain is what we see from GunMetal. Again, this is a DX9 game that makes use of shaders, so clock frequency plays a large role in performance.
The packaged benchmark that comes with the Halo is made up of all the cut scenes between levels (which are pretty graphics intensive). We ran the benchmark at 1024x768 @ 75Hz , which provided plenty of work for all our cards. We opted to simply include average framerate for this article, but there are some really cool features of this benchmark (like percentage of time above a certain frame rate) that we may revisit later.
We used the new 1.02 patch for this review which improved the performance of all cards under the benchmark a decent amount thanks to the removal of a memory check that used to occur every frame.
The 9600 XT makes good gains over the Pro in Halo. Of course, this is not surprising as Halo makes heavy use of pixel shaders which really benefit from the clock speed boost of the XT. This is a very nice gain of over 19% and one of the top gains in our lineup. We still aren't coming near 9700 Pro speeds though.
Unfortunately, neither of our RV3x0 cards would run Homeworld 2. We would get hard locks while running the game every time on both cards. This did, however, facilitate our testing of the VPU Recovery functionality introduced with Catalyst 3.8. At some points during other games we would see the VPU recovery kick in and everything would be fine and we'd go about our business. When Homeworld 2 locks the card we get a message that says VPU recovery wasn't able to fully reset the hardware and we will be blessed with the wonders of software rendering until we reboot. This is definitely better than having no recourse but to pull the plug from the wall, and we're glad to see that we couldn't freeze the entire system as we had previously with the 9600 Pro under Catalyst 3.7.
Both ATI and NVIDIA have known issues with Homeworld 2, and we are hoping that they will both be working closely with developers to solve these problems quickly. In the meantime, we are looking into possible workarounds and hope to bring you numbers for RV3x0 cards under Homeworld 2 in future articles.
Jedi Knight: Jedi Academy Performance no AA/AF
Jedi Academy is another game in a series with a distinguished history. This game is a bit of a blend between first person shooter and third person (Tomb Raider style) action depending on your weapon of choice. In order to bench this game we ran FRAPS over an in game cut scene featuring Chewie, gunfire, and a nice explosion.
Jedi Knight: Jedi Academy Performance 4X AA/8X AF
Neverwinter Nights: Shadow of Undrendtide Performance no AA/AF
Shadow of Undrendtide is the expansion to the popular RPG game Neverwinter Nights, and it features some incredible particle effects. The magic is all very well done and this is simply a beautiful game to watch. You may not like playing it unless you are already into RPGs or story driven adventure games, but anyone can enjoy watching a cut scene or two from Neverwinter. We happened to be able to benchmark the very first in game cut scene in the expansion as it had a nice sized battle with magic flying everywhere. FRAPS was again used to determine framerate.
Neverwinter Nights: Shadow of Undrendtide Performance 4X AA/8X AF
SimCity 4 Deluxe Edition Performance no AA/AF
SimCity is another game that was tough to benchmark. We finally decided that most of the time, people are scrolling around, so why not benchmark framerates while we scroll. This was done in the Big City Tutorial map; we essentially just moved side to side over a set path and recorded the framerate with FRAPS.
The NVIDIA cards are running neck and neck in this one, and both trail the ATI cards. The XT gains about 10% over the Pro in this bench.
SimCity 4 Deluxe Edition Performance 4X AA/8X AF
We actually see our largest gain from the XT in SimCity (~22%). We are getting really close to our maximum theoretical performance increase due to core clock speed in this case.
Splinter Cell Performance
Splinter Cell has been used before in our testing suites, and we are using the same procedure used for our Athlon 64 tests.
Tomb Raider: Angel of Darkness Performance no AA
As we mentioned before, we were unable to test the NVIDIA cards under TROAD. The Ti 4200 would only run with PS1.1, and as such we couldn't make a fair comparison. The 5600 Ultra was unable to run at 1024x768 as we received an out of video memory error. We have run into this issue in previous reviews. Unfortunately, even if a future patch fixes this bug, we are still stuck with v49 for benchmarking.
The gains the 9600 XT made over the Pro in this game were less than we had expected to see. This tells us that core clock speed is less of a factor in performance than other DX9 games like Halo, X2, and Aquanox. This is counter intuitive to the trends we have seen in shader intensive games, but at the same time, we can't really draw any solid conclusions from this. All we know is that TRAOD performance is less affected by core clock speed than some other games.
Note also that the 9700 Pro leads the 9600 XT by somewhere near 50%.
Tomb Raider: Angel of Darkness Performance with AA
In benchmarking with more ATI cards, it is very clear that enabling anisotropic filtering degrades image quality in a very unacceptable way. When trilinear filtering is enabled, everything looks fine, and anisotropic filtering on NVIDIA hardware doesn't exhibit this problem. This is the worst of the image quality problems we have seen with either ATI or NVIDIA in any game we have tested.
Since memory bandwidth between the 9600 XT and the 9600 Pro hasn't changed, its not surprising that our performance delta decreased when we turned on AA and AF.
Tron 2.0 Performance no AA/AF
Tron is a DX9 game, but we have been unable to find the extent to which it supports different pixel shaders. The game is actually pretty cool and incorporates some interesting effects. We benchmarked this game in one of the light cycle levels.
Tron 2.0 Performance 4X AA/8X AF
Unreal Tournament 2003 Performance no AA/AF
Unreal Tournament 2003 Performance 4X AA/8X AF
Warcraft III: Frozen Throne Performance no AA/AF
Warcraft III is an incredibly popular game, and while its not the most stressful graphics test out there, the replay mode offers the ability to track a player and speed the game up to 8 times its original speed. This can really push some numbers through the CPU and GPU. We ran this test on a replay of a tournament match we found online.
Here we see a marginal gain from the 9600 XT. Interestingly enough, the Ti 4200 performs better than the 5600 Ultra in this test. Of course, none of these cards can catch the 9700 Pro and 9800 XT.
Warcraft III: Frozen Throne Performance 4X AA/8X AF
Our performance improvements from XT evaporate when we turn on AA and AF. This seems to indicate that we have a memory bandwidth limitation which both cards share in this game. The 5600 Ultra and Ti 4200 appropriately swap positions when the memory bandwidth requirements are increased.
Wolfenstein: Enemy Territory Performance no AA/AF
Enemy Territory is actually entirely downloadable off the Internet (legally). We downloaded the replay we used from 3dcenter (radar.dm_82), and ran it as a timedemo.
Wolfenstein: Enemy Territory Performance 4X AA/AF
X2: The Threat Performance no AA/AF
X2 is a game that is due out in November, but for now, they have a really extensive benchmark/demo out that pushes a lot of different aspects of the graphics engine. This benchmark is a simple download and run. We used the benchmark with everything but antialiasing enabled.
The performance of the Ti 4200 under this benchmark was horrendous. The jerkiness problems we have noted in past articles were intensified. We could honestly watch the frame rates drop to 4 fps and then jump to ~120fps. These frame rates would alternate at somewhere near 0.5Hz (by visual approximation).
We can see a solid gain from the 9600 XT in this benchmark.
X2: The Threat Performance 4X AA/8X AF
The problems with the Ti 4200 were even more amplified here, which prompted a ten minute break from closely watching over the benchmarking process.
The Radeon 9600 XT ended up not being nearly as interesting as ATI would've liked it to be, but it does continue ATI's success in the midrange segment. We crowned the original Radeon 9600 Pro the winner of this segment back in April, and with the Radeon 9600 XT ATI extends the lead (although the improvement is only marginal). The GeForce FX 5600 Ultra is quite disappointing when put up against the Radeon 9600 Pro and 9600 XT; although NVIDIA will tell you to wait for the GeForce FX 5700 Ultra, try telling that to those that did invest in the FX 5600 Ultra.
The GeForce FX 5700 Ultra is architecturally different from the 5600 Ultra, so there is a chance that the gap could shrink but we'll have to wait another week or two before deciding on that.
We continue to see that the Radeon 9700 Pro is performing quite well and unfortunately ATI's assertions that the 9600 XT would perform similarly to the 9700 Pro are simply untrue. Given the very low price of the Radeon 9700 Pro we'd strongly suggest buying a 9700 Pro over a Radeon 9600 XT, or if you don't want to spend that much money we'd suggest a Radeon 9600 Pro as the performance difference isn't all too great.
We're wary of making a final recommendation for this segment right now because of the fact that the 5700 Ultra is right around the corner; we haven't heard anything about its performance relative to the 9600 XT so only time will tell. However if you are going to go the 9700 Pro route we mentioned above then feel free to pull the trigger, as the 5700 Ultra shouldn't be able to outperform the 9700 Pro.
What's also worth mentioning is how competitive the GeForce4 Ti 4200 remains in all non AA/aniso modes. If you are a current GeForce4 owner and don't plan on turning on any of those features then you're better off sticking with your current card until you do need DX9 support.
With another round the saga continues; the picture is almost fully in focus for this holiday season though and it's shaping up to be a very red and white scene...