ATI’s Radeon X1950 XTX graphics cards

Today is a day of dance remixes for ATI. You can hear the thump-thump-thump of the drum track throbbing in the background if you listen closely. Fresh off the announcement of its public engagement to AMD, the red team has cued up five new Radeon video cards, from the low end to the very high end, and they are all remakes of already familiar tunes.

Fortunately, in the world of video cards, remixes actually bring improvements most of the time. They tend to offer more graphics horsepower at lower prices, not just a torrid, syncopated rhythm from a drum sequencer. The new flavors of Radeons range from the X1950 XTX at just under five hundred bucks to the X1300 XT at well under a hundred. In the middle of the pack is a potential gem for PC enthusiasts: a new $279 version of the Radeon X1900 XT that looks to redefine the price-performance equation. Keep reading for the info on ATI’s revamped lineup, including our tests of the most appealing cards for enthusiasts.

The Radeon X1000 remixes
All told, ATI is unveiling five new video cards today. To cover them, we’ll start at the high end and move down.

The two cards you see pictured above are the Radeon X1950 XTX and its CrossFire Edition. The X1950 XTX is based on a chip that ATI has dubbed “R580+” for its status as a tweaked version of the R580 GPU found in all Radeon X1900-series graphics cards. Like its forebear, the R580+ is still manufactured at TSMC on a 90nm fabrication process, and it still tops out at 650MHz on the Radeon X1950 XTX, just as the R580 does on the X1900 XTX. The plus, however, extends ATI’s tradition of pioneering new types of graphics RAM by adding support for GDDR4 memory. ATI’s PR types claim GDDR4 memory uses less power per clock cycle than the current GDDR3-standard memory chips.

So the big gain with the R580+ is memory clock speeds. They’re up from 725MHz on the X1900 XTX to a cool 1GHz on the X1950 XTX—or 2GHz effective, once you take the double data rate memory thing into account. The faster RAM gives the Radeon X1950 XTX a grand total of 64GB/s of peak theoretical memory bandwidth, well above the 49.6GB/s possible on the X1900 XTX.

That, of course, raises an intriguing question: was the Radeon X1900 XTX really so limited by memory bandwidth that the switch to a new RAM type alone can yield real performance benefits? We’ll soon find out.

You may also have noticed the X1950’s fancy new cooler. ATI says it switched providers in order to get this puppy, which looks to be an improvement on the double-wide cooler used in the X1800 and X1900-series cards. This fansink still exhausts hot air out the back of the PC case, but the blower is located further inside of the case, with the aim of reducing the noise that escapes the enclosure. The new cooler is also endowed with a heatpipe that pulls heat away from the GPU into a battalion of copper fins. You won’t doubt that the thing is real copper when you pick it up; it carries more heft than a U.N. resolution.

At the back of the Radeon X1950 XTX is a pair of DVI-out ports and a video-in/video-out connector. The “built by ATI” versions of the X1950 will have support for HDCP via the DVI ports, enabling playback of DRM-encrusted Blu-ray and HD-DVD content. If you’re buying a version of the X1950 XTX from an ATI partner, you’ll need to check the spec sheet to ensure HDCP support is present, should you want it.

Oh, and of course, this is a PCI Express video card; we have no word on plans for an AGP version.

The Radeon X1950 CrossFire Edition is essentially the same thing as the X1950 XTX, save for the fact that it adds a special compositing engine for use with multi-GPU setups. ATI hasn’t yet incorporated this image compositing engine into the GPU, so a CrossFire Edition card is still required. This time around, though, the CrossFire card runs at the same clock speeds as the XTX.

ATI says to expect both the Radeon X1950 XTX and the CrossFire Edition to sell for $449. That puts it directly opposite the current prices of the GeForce 7900 GTX at online vendors.

On the left above is the next stop in our tour through the new Radeons. This is a 256MB version of the already-familiar Radeon X1900 XT. This card is still based on an R580 GPU clocked at 625MHz and mated with 725MHz memory, so it packs nearly as much graphics processing power as the former top-of-the-line Radeon X1900 XTX. The only change here is half the memory of the original X1900 XT and a much nicer price—$279, to be exact, about the price of a GeForce 7900 GT. The X1900 XT 256MB offers a heckuva lot of graphics processing power for the money.

One of the few potential drawbacks to the X1900 XT 256MB is the lack of a CrossFire Edition card that’s well matched to it. The card will operate in CrossFire mode with either the Radeon X1900 CrossFire or the Radeon X1950 CrossFire, but both of those cards are more expensive and will have to disable half of their RAM in order to work with it. ATI claims to be evaluating the possibility of enabling dongle-free CrossFire that operates via PCI Express for the X1900 XT 256MB, but they haven’t committed to a timetable for delivering it. That’s probably just as well, since this beast is probably too fast to work well in a PCI-E-based scheme.

With the introduction of these new cards, the current Radeon X1900 XT 512MB and X1900 XTX will eventually be phased out. Such things take time, though, so both products will probably linger in the market for some time to come.

We have the new X1900 and X1950 cards in our hot little hands for testing, but we haven’t get gotten our mitts on the other two cards ATI is cooking up. The first of those is the Radeon X1650 Pro, which is a dead ringer for the current Radeon X1600 XT. Both are based on the RV530 GPU. While the X1600 XT runs at 590MHz with 690MHz memory, the X1650 Pro runs at 600MHz with a 700MHz RAM frequency. Accompanying this fine-tuning of clock speeds is a price cut to $99 for the X1650 Pro, between 10 and 40 bucks less than current X1600 XT prices. The X1650 Pro is the first installment in the plan for a new Radeon X1650 family to supplant the X1600 line with faster, cheaper parts.

If a 99-dollar video card is beyond your means, there’s now an option at $89 in the form of the Radeon X1300 XT. This is the first and only member of the X1300 lineup to be based on the same RV530 chip found in the X1600/X1650 lines. For this application, the RV530 will be clocked at 500MHz and paired with 400MHz memory, so performance should be quite a bit lower than the X1650 Pro. (Think twice about saving that ten bucks, folks.) However, the XT should easily be the fastest Radeon X1300 thanks to its 12 pixel shader processors, five vertex shader processors, and eight Z-compare units, versus the RV515’s four of each. The remaining RV515-based members Radeon X1300 family will stick around, but come down in price to slot below the X1300 XT.

The scoop on pricing and availability
All of these new cards are scheduled to become available at online retailers on September 14. We’ve seen claims about pricing and availability fail to work out as planned in the past, though, so we talked to ATI board partners to see what they had to say. Fortunately, both Diamond Multimedia and Connect3D confirmed that they’re on track to hit that date.

Diamond is planning a full lineup of new Radeons from top to bottom, and they were willing to divulge suggested retail pricing for those products. Both their Radeon X1950 XTX and its CrossFire Edition are slated to list at $499, and their X1900 XT 256MB should list at $399. Diamond will also be building both PCI-E and AGP versions of the Radeon X1650 Pro with 512MB of memory onboard, and both will list at $229, while their Radeon X1300 XT with a PCI Express interface and 256MB of RAM will list at $149. That may sound pricey, but those are the price tags you can expect to see at big-box retail stores. Online vendors will discount substantially off of list price, as always seems to be the case. Diamond’s Radeon X1900 XT 512MB, for example, is currently selling for under $449 online, well below the $499 suggested retail price.

The folks at Connect3D are even more oriented toward selling through online stores, and they were willing to give us a sense of likely pricing at e-tailers. Their versions of the Radeon X1950 XTX and CrossFire will come in at “under $450” at places like Newegg right out of the gate, and their X1900 XT 256MB should arrive at “well under $300,” likely in the neighborhood of $275—all of which is right in line with ATI’s projections. They also let slip word of another new card coming a few weeks behind the others: the Radeon X1950 Pro, which won’t have all 48 pixel shaders enabled on it. They’re expecting 256MB versions of the X1950 Pro to sell for under $199, with the 512MB version arriving at about $240. They were especially excited about the X1950 Pro and X1900 XT 256MB, which they acknowledged will fill some gaps in ATI’s enthusiast-class product lineup.

Both Diamond and Connect3D told us they were considering producing liquid-cooled versions of the Radeon X1950 XTX, as well, although those products aren’t likely to arrive in the first wave of X1950 cards.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and the results were averaged.

Our test systems were configured like so:

Thanks to Corsair for providing us with memory for our testing. Their quality, service, and support are easily superior to no-name DIMMs.

Our test systems were powered by OCZ GameXStream 700W power supply units. Thanks to OCZ for providing these units for our use in testing.

Unless otherwise specified, image quality settings for the graphics cards were left at the control panel defaults.

The test systems’ Windows desktops were set at 1280×960 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

We used the following versions of our test applications:

• Quake 4 1.3 with trdm1 netdemo
• The Elder Scrolls IV: Oblivion 1.1
• Ghost Recon Advanced Warfighter 1.21
• F.E.A.R. 1.07
• Half-Life 2: Episode One with trdem1 demo
• FutureMark 3DMark06 Build 1.02
• FRAPS 2.7.2

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Pixel-filling power
We’ll begin our testing with a quick look at some theoretical numbers for pixel throughput. These numbers are mainly related to the card’s ability to push pixels through the traditional graphics pipeline, and they are becoming increasingly less important as programmable shading becomes more prevalent. They are still relevant in many of today’s games, though, so we’ll have a quick look at them. I’ve included the Radeon X1300 XT and Radeon X1650 Pro in the table below for reference, even though we don’t yet have examples of those cards to test.

The Radeon X1900 XT 256MB has the same basic specs as the 512MB version, save for memory size. That puts it in a very respectable place. The X1950 XTX, meanwhile, has a tremendous memory bandwidth advantage on everything but the “SLI on a stick” dual-GPU GeForce 7950 GX2.

How do these numbers translate into performance on synthetic fill rate benchmarks?

The X1950 XTX’s faster RAM helps it make big gains over the Radeon X1900 XTX in the single-textured fill rate test, despite having the same GPU clock frequency. Once we get into multitextured pixels, though, that advantage is muted.

Of course, these numbers aren’t destiny in real-world applications. The GeForce cards do especially well here thanks to their ability to use one of the ALUs in each of their pixel shader processors for texturing. The R580-derived cards, meanwhile, have 16 dedicated texture address units and 48 dedicated pixel shader processors at their disposal. In other words, the two architectures allocate their resources quite differently, and the implications are difficult to predict with a simple fill rate test.

Quake 4
In order to make sure we pushed the video cards as hard as possible, we enabled Quake 4’s multiprocessor support before testing.

The Radeon X1950 XTX’s faster memory doesn’t seem to help very much here, since it’s barely any faster than the Radeon X1900 XTX. The Radeon X1900 XT 256MB, though, asserts itself by beating out the GeForce 7900 GT across the board.

Notice one other thing about the results, which is how small the spread is between the Radeon X1900 XT 256MB, the Radeon X1900 XTX, and the Radeon X1950 XTX. In Quake 4, all of them run slower than I’d like at 2048×1536, and all of them hit acceptable frame rates at 1600×1200. That’s a dynamic to consider as we look through the rest of the results.

F.E.A.R.
We’ve used FRAPS to play through a sequence in F.E.A.R. in the past, but this time around, we’re using the game’s built-in “test settings” benchmark for a quick, repeatable comparison.

Not only does the X1900 XT 256MB easily surpass the GeForce 7900 GT here, but it also gives the 7900 GTX a run for its money. The X1950 XT is clearly faster than the GTX, and it gains more of an edge on the X1900 XTX in this game.

The line graphs really illustrate the wickedness of that problem with performance scaling on the Radeon X1950 CrossFire setup at 1280×960. I’m not sure what the story is here. I ran these numbers multiple times, and the behavior was consistent. Must be some kind of driver quirk or something. Notice, also, that the low frame rates on the CrossFire rig aren’t much higher than they are with a single X1950 XTX.

Half-Life 2: Episode One
The Source game engine uses an integer data format for its high-dynamic-range rendering, which allows all of the cards here to combine HDR rendering with 4X antialiasing.

The Radeons have quite the edge in Half-Life 2: Episode One—so much so that the X1900 XT 256MB outdoes the GeForce 7900 GTX.

The Elder Scrolls IV: Oblivion
We tested Oblivion by manually playing through a specific point in the game five times while recording frame rates using the FRAPS utility. Each gameplay sequence lasted 60 seconds. This method has the advantage of simulating real gameplay quite closely, but it comes at the expense of precise repeatability. We believe five sample sessions are sufficient to get reasonably consistent and trustworthy results. In addition to average frame rates, we’ve included the low frames rates, because those tend to reflect the user experience in performance-critical situations. In order to diminish the effect of outliers, we’ve reported the median of the five low frame rates we encountered.

We set Oblivion’s graphical quality settings to “Ultra High.” The screen resolution was set to 1600×1200 resolution, with HDR lighting enabled. 16X anisotropic filtering was forced on via the cards’ driver control panels.

The GeForce 7900 GTX and Radeon X1950 XTX are neck and neck here, while the Radeon X1900 XT 256MB continues to outperform the GeForce 7900 GT. I should mention, though, that there was a clear visual difference in this game between the 16X anisotropic texturing filtering modes offered by the two brands of cards. The Nvidia cards showed lots of high-frequency noise at certain mip levels and on mip-map boundaries, particularly in the cobblestone texture on the ground. The ATI cards did a much better job of presenting a sharp image without pixel shimmer, moire, or crawling. Frame rates also felt subjectively smoother on the ATI cards, especially during the first of our five runs through this sequence.

Ghost Recon Advanced Warfighter
We tested GRAW with FRAPS, as well. We cranked all of the quality settings for this game, with the exception of antialiasing. However, the game doesn’t allow cards with 256MB of memory to run with its highest texture quality setting, so the GeForce 7900 GT and Radeon X1900 XT cards were both running at the game’s “Medium” texture quality.

It’s a toss-up between the Radeon X1950 XTX and the GeForce 7900 GTX. In both single-card and multi-GPU configs, the 7900 GTX produces higher average frame rates, but the X1950 XTX achieves higher minimum frame rates. The X1900 XT 256MB, meanwhile, rolls on.

3DMark06
I’ve included scores for the Radeon X1900 XTX CrossFire rig below, despite the fact that we saw big problems with visual artifacts in graphics tests three and four of 3DMark06. The CrossFire system’s performance also didn’t scale as well as expected from one card to two, and I suspect the visual artifacts played a part in that. I was able to confirm that the problem wasn’t confined to the Radeon X1950 CrossFire setup by trying a Radeon X1900 CrossFire rig, which exhibited the same problems. I suspect some sort of driver problem or a mild incompatibility with our Intel D975XBX motherboard. We’re working with ATI to find a resolution, and we’ll update the scores here once we have one.

Update: We were able to overcome the CrossFire issue by switching to an Asus P5W DH motherboard. On that board, the Radeon X1950 CrossFire system executed 3DMark06 tests three and four without any problems, and the overall 3DMark score is higher as a result. The updated graphs are now in the article below.

The Radeon X1950 XTX’s extra memory bandwidth allows it to open up a slim but decisive lead over the GeForce 7900 GTX here. At the same time, the Radeon X1900 XT 256MB demonstrates that it has the GeForce 7900 GT outclassed.

3DMark’s quick vertex and pixel shader tests don’t reveal much new to us, but we’ve included them for the sake of completeness.

Power consumption
We measured total system power consumption at the wall socket using a watt meter. The monitor was plugged into a separate outlet, so its power draw was not part of our measurement. We tested all of the video cards using the Asus P5N32-SLI SE Deluxe motherboard, save for the CrossFire system, which required a different chipset. For that system, we used an Intel D975XBX motherboard.

The idle measurements were taken at the Windows desktop. The cards were tested under load running Oblivion using the game’s Ultra Quality setting at 1600×1200 resolution with 16X anisotropic filtering.

Power consumption at idle is quite good for all of the single-GPU systems. When running a game, though, the ATI cards simply require more power. The Radeon X1900 XT pulls 75W more under load than the GeForce 7900 GT, and the Radeon X1950 XTX out-pulls the GeForce 7900 GTX by a similar margin. The one bright spot for ATI here is the fact that the Radeon X1950 XTX draws 17W less power than the X1900 XTX. Apparently, the combination of R580+ and GDD4 memory has the potential to reduce power draw somewhat.

One reason that the power draw is relatively low for the X1950 CrossFire system at idle, by the way, is the fact that the Intel motherboard supports the Core 2 Duo processor’s C1E enhanced halt state, which cuts the CPU clock speed at idle. The P5N32-SLI SE Deluxe doesn’t appear to support C1E halt, so the CPU draws more power when sitting idle on that board.

Conclusions
See, like I said, remixes can bring good things. In this case, we have a new top-end Radeon X1950 XTX that furthers ATI’s claim on the title of the fastest single-GPU graphics card around, and does so without costing any more than the product it replaces. I’m not sure I can say definitively that the Radeon X1950 XTX is faster than the GeForce 7900 GTX—that depends on the situation. You will have to live with higher power consumption if you choose the Radeon, but ATI has worked to mitigate the worst effects of that fact with its new cooler. The X1950 XTX also has a few feature advantages in its corner, including the ability to handle floating-point texture filtering for high-dynamic-range rendering in concert with antialiasing and a superior default algorithm for anisotropic filtering. That’s a pretty good combination.

Meanwhile, with the 256MB version of the Radeon X1900 XT, ATI has captured our full attention. The X1900 XT 256MB clearly has the GeForce 7900 GT running at Nvidia’s stock speeds beaten on performance. We’ve seen the now almost-ubiquitous “overclocked in the box” versions of the GeForce 7900 GT deliver somewhat better performance, but they’re not likely to make up the gap entirely. As I said early on, the X1900 XT 256MB’s biggest liability is the lack of a well-matched CrossFire Edition card for it. If you hope to upgrade to multiple GPUs, you may be better off with a 7900 GT.

Still, the X1900 XT 256MB is so good, it raises all sorts of interesting questions. If ATI and its board partners can deliver this thing at the promised $279 price point, we’ll have a new favorite graphics card to recommend. In fact, given the performance we’ve seen out of the X1900 XT 256MB, one wonders why anyone should pay any more for a graphics card. Yes, the X1950 XTX is certainly faster, but it’s not in an entirely different class or capable of offering playable framerates in many resolutions where the X1900 XT 256MB can’t. But then you probably knew that the high-end card wasn’t exactly at the sweet spot for price and performance.

Now the big question is: how will Nvidia respond? These products won’t hit store shelves for three weeks, and that’s a veritable eternity in PC graphics. Remixes don’t take long to paste together, so we may be hearing another new track or two soon in the thump-thump-thump of progress.