ATI Radeon X1900XTX – Performance and overclocking

We’ve taken a closer look at ATI’s new flagship. Radeon X1900 XTX has spent some time in our test lab and except for our regular tests we did some overclocking with both air and water.

Here we go again. The video card market is bubbling and the manufacturers are playing a multinational chess game with pieces in the form of video cards, paper launches and media hypes. Half a year ago, it appeared that everything wasn’t right with ATI’s R520 video circuit. Because of an irritable transistor failure, ATI was trapped on the playing field and nVidia got a 4 month performance advantage. When the R520 finally hit the market in November, nVidia had saved its best circuits for what would become the 7800GTX512, and also the fastest video card on the market. But nVidia’s shrewd moves were over and now it was ATI’s turn. ATI had in parallel with the R520 also developed a circuit codenamed R580. With a more mature 90nm-manufacturing process and multiple enhancements to the core, it was now time for ATI to perform a castling and get back on the playing field.

On the 24th of January, ATI officially announced their X1900 video card series with the obvious goal to retake the performance throne. The cards included in this series are as of today: X1900XTX, X1900XT, X1900AIW (All-In-Wonder) and we have yet to see if ATI will launch a card corresponding to a X1900XL. In this review we will take a closer look at the fastest card, X1900XTX, see how it performs in some common games and in the end perform an overclocking-evaluation.

We will begin by looking at its specifications.

Because the card is very similar to the X1800-series, we direct those of you who are interested about knowing more about these circuits, to two of our previous reviews that are more thorough. You can find them here:

:: ATI Radeon X1000 series ::

:: ATi Radeon X1000 – Performance and Image quality ::

As previously stated, there’s no big difference between the R580 and the R520 because they are based on the same architecture. The biggest difference is that the number of “Pixel Shader”-units is increased from 16 to 48. Above is a schematic sketch of the 12 shader-bunches that each contains 4 calculation-units. Unfortunately, this doesn’t mean that the performance has been tripled compared to the X1800-cards, because they still share functions with each other. Also, it takes more than shader-performance to make a video card efficient. Below some of the main characteristics of the X1900-series compared to the X1800-series are listed.

By way of introduction, there are all in all 4 different versions in the X1900-series. At the top we have the XTX-card that, not unexpectedly, has the highest core and memory clock frequencies. Then we have two XT-cards with slightly lower frequencies, one of them equipped so that it can be used in a dual setup, what ATI calls CrossFire. These three cards all have a dual slot-cooler, which means that the cooler occupies an extra physical slot, and thus you can’t use the slot directly below the card. The X1900 All-In-Wonder-card is however smaller and doesn’t use an additional slot. Those of you who are familiar with ATi’s AIW-cards know that this card is a fully equipped multimedia card with, among other, a TV-tuner and video-in. Because of its smaller cooling device, this card has a considerably lower core and memory clock frequencies.

Let’s have a look at the card itself.

At first sight, it’s not easy to distinguish an X1900-card from an X1800-card. They both share the same cooler design and the reference cards colors and stickers are also very similar. If you look closer, you see that ATI has beefed up the voltage-part of the card so it will be able to deliver enough current and stable voltages to the core. The five-phase voltage regulator has been increased to seven, which tells us that the power consumption has increased.

We continue with our dissection of the card and when the cooler is removed the heart of the card is unveiled. Because the core was manufactured week 50 last year, it must be considered brand new. The memories are manufactured by Samsung and their label BJ11, tells us that they are specified for 1800 MHz DDR-speed, which means that we should have a good overclocking margin.

Let’s take a closer look at the cooler.

The cooler is considered big when compared to other stock coolers. It’s of course made of copper and two heat-pipes are used to increase its heat dissipation ability. The bottom of the cooler is not completely even and you can feel traces from the machine work when you move your thumb over it.

Here you can see what keeps the core at a reasonable temperature. The fan is specified to 5W at 12v and is completely software-controlled to keep the sound at a reasonable level. And finally, the card has (as all other performance-cards) a 6-pin PCI Express-power outlet to make sure the card is supplied with enough current.

Next up is the system we will use in our performance tests.

Test system
Hardware
Motherabord	Abit Fatal1ty AN8 SLI, BIOS 19
Processor	AMD FX-57 (San Diego, 0512MPMW)
Memory	Corsair XMS3200 2x512Mb (BH-6)
Video cards	ATI Radeon X1900XTX nVidia 7800GTX512
Power supply	OCZ PowerStream 520W
Software
Operating system	Windows XP (SP2)
Drivers	nForce 6.70 Catalyst 8.203 ForceWare 81.89
Monitoring program	Fraps 2.5.0
Test program	3DMark2001 (3.3.0) 3DMark03 (3.6.0) 3DMark05 (1.2.0) Battlefield 2 Colin McRae 2005 Call of Duty 2 – Demo Doom 3 (1.0.1262) Far Cry (1.32) Half-Life 2 Need For Speed Underground 2 Unreal Tournament 2004 (v3399)

We’ve chosen an, to our other reference tests, identical system to make the comparison as exact as possible. One essential difference is, however, that the video card drivers are newer for both ATI’s X1900 and nVidia’s 7800GTX512. In order to simulate the results of an X1900XT card, we underclocked the XTX card to the same specifications as the XT version.

Because of heavily modified conditions in some games since the other X1000 series tests, we’ve chosen to exclude some games. These will, in later comparisons, be replaced by more recent games, but unfortunately not in this test because we no longer have got the GTX512 card.

On the next page, we’ll look into the sound level and measure the power consumption.

Sound level

We weren’t completely happy with the sound level of the X1800 series cards when we reviewed them last fall. That’s why it became extra interesting to see whether ATI had made some modifications on the cooling unit since then. Even though the cooler looks identical to the X1800’s, we had hoped the fan would be silenced down a bit. However, our hopes were crushed during the first system startup, when the characteristic sounds of the X1800 cards were back. The card regulates the fan’s RPM according to the temperature – just like the X1800 card, which makes the card’s sound level at idle acceptable. When in load, the sound level is further risen, and very dependent on the temperature in the chassis. Considering the sound level, this card I unfortunately far behind nVidia’s 7800GTX512 – which even under full load is considerable more quiet than X1900, no matter what load.

Power Consumption

When measuring the power consumption for the whole system, we use an ampere meter connected just before the PSU. Ergo, we’re not measuring only the video card’s power consumption. The X1900 card’s got a higher clock frequency than the X1800 series. The amount of transistors has also been increased – making the power consumption rise. We can also see that ATI’s used a more powerful power circuit compared to the X1800 series, which also indicate of an increased power consumption. Let’s see what the numbers say.

As you can see, the card uses less power than nVidia’s best card. We also took the chance to measure the card’s temperatures after some time intensive testing. On the picture, “T1” is the temperature at the heatsink and “T2” the air temperature just behind the card’s air exhaust. Temperatures of around 70C in an open system are really remarkable. The positive side with the cooler is that all the heat is dumped outside the case.

Now it’s time to bench the card. First up are some synthetic tests.

Even though 3DMark2001 is heavily system limited with this configuration, the X1900 cards do their task very well and are faster than any other card. In 3DMark03, however, it isn’t that easy because GTX512 shows off an enormous strength and leaves a gap of over 1500 points to the next card. In 3DMark05, where X1900XTX really gets go stretch its legs, it’s once again the winner. It’s actually the winner with over 2000 points over the GTX512 – a result that would have been impossible to achieve a year ago, even with two cards in SLI. This really shows how fast the video card performance is growing.

Next up is a setup of game tests, and Battlefield 2’s first.

Setting:	Value:
Overall Quality	High
Terrain	High
Effects	High
Geometry	High
Textures	High
Lightning	Medium
Dynamic Shadows	High
Dynamic Light	High
Viewing Distance	100%

The X1900 cards have no problem what so ever winning the Battlefield 2 performance throne. Similar to the X1800 cards, X1900 manages FSAA and AF without any problem and doesn’t suffer from such a big hit when these are enabled.

Next up is Call of Duty 2.

Setting:	Value:
Number of Dynamic Lights	High
Model Detail	Normal
Z Feather Smoke	Everything
Number of Corpses	Insane

Call of Duty is clearly one of our heaviest game test but even though the FPS’s are relatively low, the game is playable down to 30 FPS. X1900 handles the challenge best but loses, for some reason, ground with FSAA and AF activated, which is the complete opposite of what we’ve seen so far. The card is still fresh and will get more better that’ll get out more performance out of this game.

Now, we’ll look into Far Cry.

Setting:	Value:
Texture Quality	Very High
Texture Filter Quality	Trilinear
Particle Count	Very High
Special Effects Quality	Very High
Environment Quality	Very High
Shadow Quality	Very High
Water Quality	Very High
Lightning Quality	Very High

nVidia’s 7800GTX512 puts on a tough front and gets away with a victory in the first round. At a little higher resolution, X1900XTX gets back its lead and with FSAA and AF activated, X1900 wins with a safe lead.

Time for Half-Life 2.

Setting:	Value:
Model Details	High
Texture Details	High
Shader Details	High
Water Details	Reflect World
Shadow Details	High
HDR	None

ATI shows off some nice performance in Half-Life 2 and even the X1800XT card gets an opportunity to pass the GTX512 card. The X1900 cards are unthreatened in the lead – no matter which setting’s used.

Next up – Doom 3.

Setting:	Value:
Special Effects	Yes
Enable Shadows	Yes
Enable Specular	Yes
Enable Bumpmaps	Yes

nVidia has always done well in Doom 3 and that’s not an exception this time either. The GTX512 got away with a green victory. Even the regular GTX card manages to squeeze in between the XT and XTX cards. At a higher resolution and with FSAA,  XT takes back the third place thanks to more effective anti-aliasing and filtering.

Up next are two racing games, and first up is Colin McRae 2005.

Setting:	Value:
Texture Quality	High
Drawing Distance	10 (max)
Post Processing Effects	Yes

Here we can see further proof of ATi’s efficient anti-aliasing and filtering algorithms. Without FSAA the GTX512 card wins, but with FSAA on we can see that even ATi’s X1800 card beats the GTX512.

The next racing game is Need For Speed Underground 2.

Setting:	Value:
Level of Detail	Max
Car Reflection Update Rate	Max
Car Reflection Detail	High
Car Shadow	Max
Car Headlight	On
Car Geometry Detail	High
Crowd	On
World Detail	Max
Road Reflection Detail	Max
Light Trails	On
Light Glow	On
Particle System	On
Motion Blur	On
Fog	On
Depth of Field	On
Tinting	On
Horizon Fog	On

You can pretty much take a carbon paper and copy the results from Colin McRae to this test. nVidia’s GTX512 performs better as long as FSAA is deactivated, but once it’s turned on we see a switch in the charts and both X1900 cards take the lead.

Now it’s time for some overclocking, and we start with the original cooler.

Overclocking is dear to us, and when ATi even supplied us with a program for that purpose we weren’t going to miss the opportunity. We start off by overclocking with what will probably become the most common cooling solution, namely the original heat sink. The fan is adjustable from within the application and the first thing we did was to set it to maximum speed. We also set up some additional fans to improve cooling capabilities.

First off we tested the memory, which according to specifications should reach roughly 900MHz as we’d established earlier. 909MHz gave occasional artifacts, 918MHz more consistent “snow” and 927MHz caused freeze-ups during 3DMark. We stopped at 900MHz and moved on to the core. The core went all the way up to 735MHz, but to manage a run through 3DMark05 we had to step down to 729MHz. Finally we set the CPU to 3GHz (12×250) and ran 3dMark05. The result was 12983, but 13000 would have been more fun so we turned up the processor to 3050MHz. 13030 points using a video card with original cooler and an air-cooled CPU must be seen as a good result. The voltages are untouched and we get the feeling that the overclocking software didn’t manage to change them, as our overclocking results weren’t affected at all.

We install water cooling and open a window.

We set up a water cooling system, consisting of a Danger Den water block, 3x120mm radiator and an Eheim 1250 pump. The radiator was placed at an open window, and then we waited for the temperature to drop before we began overclocking. +3.3 degrees centigrade was the lowest value we received, which made us begin.

The memory was cooled with a large fan using cool air, but wouldn’t go further than before, 900 MHz. The core clocked nicely with lots of chill, and we succeeded in completing occasional tests at 796 MHz, but for a whole round of tests, we were forced to go down to 790 MHz. Thanks to the colder air, we also managed to clock the processor a bit more. 13497 points was the final result and in spite of numerous tries, 13500 points wouldn’t work with us on this occasion.

The review is coming to an end and we’ll take a moment to summarize it all on the next page.

It’s been a hectic hunt for both CrossFire cards and double sets of nVidia cards, and unfortunately these configurations were absent in this test, though we have already decided to get double sets of cards and present performance values for these in a separate article. For most of the performance tests, the figures speak for themselves when it comes to this card. ATI has learned from its manufacturing mistakes concerning the X1800 cards, and now possess full control over the situation. It looks like there are cards available in retail stores. We will now list some summarizing comments about the card.

What the figures don’t tell is how the card behaves material-wise and user-wise, and here are some things that we would like for ATI to attend to.

We’ve pointed out these things before and we think that ATI should introduce an updated driver when introducing a new video card. We’ve also had different viewpoints on the sound level before, and we would be glad if ATI had developed a larger and more effective fan with a lower sound level.

ATI Radeon X1900XTX

Pros: + Market-leading performance + Anti-aliasing and anisotropic filtering + Availability Cons: – Beta drivers – Heat dissipation and sound levels

We thank ATI for providing us with a X1900XTX card for testing.