Author: Terren Tong
Editor: Howard Ha
Publish Date: Tuesday, December 23rd, 2003
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/9800xt/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.
Late in the holiday season, the jolly man from Fedex filled our stockings not with coal but with packing foam, today's equivalent of coal. As we were about to lay it into him (and permanently cement ourselves on the other fat man's naughty list) we noticed that the package had something else, a blistering 9800 XT courtesy of ATI. As shown in non-scientific studies from major non-accredited universities, Christmas shopping at the last minute makes you more of a man. We figure the same principle applies to reviews that come out late in the holiday season.
Since the release of the 9700 Pro, ATI has executed perfectly at every step in both the mid and high range segments. The 9700 Pro was succeeded by the 9800 Pro during the spring and come fall, we have the 9800 XT. The mid range segment has been filled by the 9500 Pro, 9600 Pro and the 9600 XT which we have recently reviewed here. Drivers on the ATI side have been improving steadily and although they still have their share of issues, ATI's driver development team has made huge strides with the improvement of the Catalyst series of drivers. Although recent parts such as the 5950 and 5900 have been solid cards, Nvidia has largely been kept at bay as they are playing the role of the underdog, a position they have not been in since the early 3dfx days. The performance of the 5950 and the 5900 have been good in most cases, but they have faltered in Pixel Shader 2 performance which has raised some eyebrows as the next generation of games that come out will be relying on these features more.
The 9800 XT is the fall/winter refresh for the R3xx core and is an improvement over the 9800 Pro released earlier this year, the clock rate was bumped up slightly; the core goes from 380 mhz to 412 mhz while the memory is now at 730 instead of 680 on the Pro. The XT also comes with 256 MB of memory over the 128 found on the regular 9800 Pro. There are no new major architectural changes on the 9800 XT, just some minor improvements and tweaking. The VPU is still being built on a .15 micron core unlike the 9600 series which debuted at .13. ATI's RV400 series, their next high-end part, is rumored to be built on the .13. ATI's strategy seems to revolve around testing out next-generation processes with their mainstream parts before moving the high-end down to it. This strategy has worked well for them as the R3xx series did not run into the manufacturing debacle that Nvidia experienced while trying to release the NV30.
Now that the obligatory history lesson is done, we take a look at the subject at hand, ATI's 9800 XT.
He's a badass. He needs to leer less and scowl more
The 9800 XT is built on the ATI trademark red PCB and is a single card solution unlike Nvidia's high end reference offerings. The large heatsink is make of copper and there is acrylic piece on top with a fancy graphic design that kind of look like a bunch of turbine engines laid out in a circle. A large fan (probably Â¼ of the size of the card) sits over the VPU. There is also a heatplate on the back that draws heat away from the memory chips as well.
The 9800 XT does get very warm during operation so for those with poor air circulation inside their cases, it is advisable to keep an eye out on case temperatures with the 9800 XT in there. Unlike a few of the other cards we have recently reviewed, there are no fancy leds, either red or blue. The 9800 XT seems noticeably louder than both the FIC 9800 Pro and the 9700 solutions that we have although it does not seem as loud as the Gainward 5950 that we recently reviewed and that was verified through the use of our handy sound meter which pegs the 9800 Pro at ~52 DB at a distance of approximately 1 meter away from an open case consisting of the 9800 XT and a stock P4 cooler. (For reference, the 9800 pro came in at about 46 DB and the Gainward at 53 DB in the same case. A quick Discovery Channel moment reveals that decibels are logarithmic meaning that for every 10 DB, there is a doubling in noise. The 6 DB between the 9800 Pro and the 9800 XT will be quite pronounced while the Gainward and 9800 XT will be very close. Also note that our lab had a relatively loud 46 DB of ambient noise as measured by the same meter. The 9800 XT is clearly audible above the ambient noise though. Check out this guide for relative noise measurements with the same meter if you want a point of reference outside of videocards.)
The two cards are the same size but notice all the extra caps on the XT
David, meet Goliath.
Our box included a manual, a driver disc, both a RCA and S-Video cable (and an svideo -> rca converter. did not understand that since there is already a RCA cable), a power dongle and a DVI->VGA converter. The retail version should also include a coupon for Half-Life 2 as shown explicitly on the front but alas our 9800 did not include one.
One of the new features in the XT line of cards from ATI is the presence of an onboard thermal diode. This allows the drivers to overclock the videocard automagically when it is in a certain temperature range and throttle it back down if it gets too warm. From our friends at ATI we found out that Overdrive will clock the VPU core up to 419 mhz if it is below 56Â° celsius and temperatures above that will result in the default clockspeed being used. The upside is that the clockrate never drops below the default 412, the downside is that the actual overclocked speed is miniscule. By miniscule, we mean that this is probably something you will not notice during gameplay.
While Overdrive is a nice feature especially for those who are leery of overclocking, it would be much preferable for ATI to have included a Coolbits type utility like Nvidia so the more experienced end user can have more control over the clockrates without having to resort to third party utilities. Overdrive the way it is now, is more of a marketing gimmick than a feature that will bring substantial performance gains.
ATI may want to take a look at AOpen's SilentTek feature on their motherboards (which we like very much). One of SilentTek's modes allows the end user to set the fan speed to a certain level depending on the temperature of the CPU. It would be great if ATI could do the same thing with regards to clock rates- have the VPU adjust to an user specified clockrate depending on the temperature. Not only could the end user overclock their card, but they can do so in a safe manner and if the videocard heats up to a point where the end user starts getting leery, the Overdrive software can kick it back down a notch. Overdrive has great potential and hopefully ATI will implement something in later driver revisions that will really take advantage of it.
A couple extreme close-ups of the heatsink
Intel P4 2.8BGhz
AVC Sunflower P4 Cooler
MSI 648Max motherboard
512MB OCZ PC2700 DDR Memory
Seagate 120GB ATA133 Barracuda ST3100
WinXP with SP1
nVidia GeForce FX 5950 Ultra reference card
ATi Radeon 9700 Pro
ATi Radeon 9800 Pro (made by FIC. reference design)
ATi Radeon 9800 XT
We chose not to include any midrange cards in the lineup because the price difference is obscenely large. For the price of a 9800 XT one could obtain two or three 9600 XTs so clearly they are not in the same class.
Unfortunately most of the cards are running on different driver revisions. The 9700 Pro Benchmarks were done with the hotfix CoD drivers that were released between the Catalyst 3.9s and the Catalyst 3.10s. The 9800 Pro benchmarks were done with the 3.9 drivers with the exception of Call of Duty which used the Cat 3.10s. The GFFX 5950 was done on the 52.14 drivers with the exception of the Call of Duty benchmarks which used the 53.03 driver sets. The 9800 XT was benchmarked using the Cat 3.10s throughout all of the benchmarks. Why the big mishmash of drivers and why did we not rebench everything? For the 9800 Pro and the 9700 Pro we did not feel that the Cat 3.10s delivered a huge performance jump over the 3.9s to warrant a full rebenching of both cards. This was the same rationale with the 5950 and the use of the 52.14s throughout most of the numbers. The driver variation is due mostly to Call of Duty, a recent addition to our benchmark suite; newer drivers were available at the time so the latest available drivers were used in each instance and the numbers we got there were just appended to our dataset.
Our one and only synthetic. There are a couple games that are based off this engine which is why this benchmark is here. Aquamark makes use of a lot of DX9 effects and may be a good indication of how Aquanox 2 and Spellforce will run.
The results are a bit surprising. Without AA/AF, the 5950 powers through this benchmark by quite a significant margin. DX9 is not a traditional strongpoint of the GeForce FX series. The victory though is kind of shortlived because as soon as AA/AF gets turned on, the 5950 goes flaccid; flaccid enough that the 9700 Pro beats it out. The superior memory architecture of the R3xx series is really highlighted here.
Commanche 4 is primarily CPU bound in a non AA/AF situation and the results there are quite boring as there is little variance in performance between 1024 and 1600. ATI also claimed that in the Catalyst 3.10 release, Commanche scores could be 7% higher. That may account for the variance seen between the 9800 Pro and the 9700 Pro (remember that the 9700 was using new drivers than the 9800 Pro). The ATI cards tend to drop off quite sharply in constrast with the 5950 and the 5950 pulls off the highest framerates at 1600x1200.
Come to the darkside. Do not underestimate the power of the Force. ATI was supposed to have fixed their driver problems plaguing Jedi Academy with their 3.9 Catalysts and they had to to an extent (the framerates before were quite abysmal) but it does not seem they are all the way there yet. The entire line of ATI cards lag behind the 5950 in both custom demos we have with both AA/AF on and off.
Call of Duty
Our second Q3 based shooter. One of our recent favorites is Call of Duty (conviniently it is also a pack in with NVidia's 5900 series of cards). Great WW2 based shooter. If you have not tried it yet it is well worth picking up a copy. Online battles are fast and furious and the single player portion is great also. Intially ATI gets shelled here and needs to find the nearest bunker to take shelter in. Of particular concern is the close grouping of the ATi cards at 1024 with no AA/AF. The numbers there should not be that close together and it seems that ATI still needs to work on their drivers somewhat as our particular demo is not particularly CPU bound. At 1600x1200 though the 5950 loses steam and finds itself right near the 9800 XT.
CoD with AA/AF on has the 9800 XT powering out a sweep and it is a tad faster than the 5950 at all resolutions. Notice that there are only 3 lines representing the 9700, 9800 XT and the 5950. There is a single datapoint for the 9800 Pro marked in orange and positioned near the 9700. No, we did not get bored and refuse to finish benchmarking the 9800 Pro. The 9800 Pro would lock up almost immediately during demo playback at 1280 and 1600 with AA/AF on. This happened with both the CoD hotfix drivers as well as the Catalyst 3.10s. Strangely enough this particular affliction only affected the 9800 Pro as testing last week of the 9600 XT during our review of the BFG Asylum 5900 turned out okay also.
Another shooter that we have enjoyed recently. Halo is a demanding game that really stresses even high end systems. We run a set of tests with both the Pixel Shader 2 and the Pixel Shader 1.4 pipelines. The 9800 XT leads in both resolutions and with both rendering pipelines. It looks like ATi has improved their Pixel Shader 1.4 efficiency after the 3.9 series of drivers as the 9700 Pro passes the 9800 Pro by a small margin.
Our final FPS is UT2k3. Results here are bit more varied. The 5950 and the 9800 Pro are neck and neck and flip flop back and forth. Again the 9800 XT cleans up. The numbers for all the cards are very good though and is playable at all resolutions with AA/AF on so if UT2k3 is your bag then pretty much any of these cards would suffice.
Tomb Raider has been the source of more contention than Lara Croft's b... uh brown hair. Angel of Darkness makes use of a lot of DX9 shaders and it gets very bogged down in several areas of our demo based on the prague3a level. All the cards slow down significantly in the water room and we can honestly say, the 9800 XT is the only card to pass through our labs so far that actually looks smooth in that particular room. Yes, there is a noticeable decrease in FPS but it was not molasses slow but looked very good. Very impressive. There is not much to say. The 9700/9800/9800XT all show pretty consistent performance; the slopes of the graphs are pretty much the same for all three cards. The 5950 is not as slick and must've fallen into one of those devious traps that Lara always finds herself in as it lags significantly behind.
X2 - The Threat Rolling Demo
Our final benchmark for the day. X2 is a slick looking space sim based from Egosoft. Very close performance between the 9800 Pro and XT. More of a dogfight between the 5950 and the 9700 Pro although the 5950 has the upperhand.
Our particular card hit 456 / 394 up from the defaults of 412/365. The overclock on the memory is a bit on the low side while the core speed is a pretty decent jump at around 10%. We let 3dmark03 loop for about 40 minutes or until we saw visual artifacts (so we lied about Aquamark being the only synthetic benchmark we used).
The 9800 XT is the undisputed speed king placing first in nearly every single benchmark with the exception of Jedi Academy where it is still hampered apparently by driver problems. A couple of the benchmarks like Aquamark and Call of Duty see-sawed a bit but the 9800 XT was right up there. Unfortunately unparalleled performance comes with a large price tag to the tune of 500$ MSRP. A quick search reveals that a 9800 XT can be had for about 430$ which is still quite the sum of money. The 5950 is a tad cheaper at a street price of approximately 410$. Those who already have a 9700 or 9800 should probably hold off on upgrading as we see them both being still extremely competitive with its big brother. The 9800 XT seems to be a better buy than the 5950 considering that it nearly sweeps our benchmark suite and it also includes a coupon for a copy of Half-Life 2. A minor knock against the ATi cards is that they were more finicky in our test systems with more quirks than the Nvidia based counterparts. A couple times after driver installs, XP would not finish booting but a quick reboot managed to fix it. It is hard to say that the 9800 XT is a bargain but as was mentioned in our review of the Gainward 5950, flagship cards are for the really hardcore users and in this particular case there is no compromise being made in the performance department. The 9800 XT should last well into the next generation of games and if you have a large wad of cash this is the card to get. If the budget is a little less forgiving a 9700 or a 9800 Pro would still be an excellent buy.
Please do not redistribute or use this article in whole, or in part, for commercial purposes.