HIGH-END GRAPHICS CARDS like NVIDIA’s GeForce FX 5800 Ultra and ATI’s Radeon 9800 Pro attract all manner of adoring fans and groupies, but only a select few can really afford to drop a fistful of Benjamins on these pricier upgrades. Hard-core gaming enthusiasts may always want the fastest available graphics cards, but for the rest of us, more reasonably-priced mid-range graphics cards can still offer buttery smooth frame rates in today’s games. Heck, many of today’s mid-range graphics cards can dish out more than 60 frames per second at high resolutions with antialiasing and anisotropic filtering enabled.
For the last several months, ATI’s Radeon 9500 Pro has dominated the graphics card scene for gaming enthusiasts looking for the best bang for their graphics card dollar. We were thoroughly impressed with the Radeon 9500 Pro when we reviewed it back in November, and the card has been our mid-range graphics card pick ever since. Along the way, NVIDIA has tried to mount an offensive with its GeForce FX 5600, but those cards have yet to materialize in retail. In fact, only a small number of GeForce FX 5600 reviews have slipped onto the web, and thus far it doesn’t look like the card poses a serious threat to the Radeon 9500 Pro.
Still, without a readily available competitor for the Radeon 9500 Pro, ATI has decided to refresh its mid-range offering with a new card, the Radeon 9600 Pro, based on the new RV350 GPU. The RV350 features all the swanky DirectX 9 features you’d expect from one of ATI’s latest Radeons, but it has only half as many pixel pipelines, pixel shaders, and vertex engines as the R300 chip in the 9500 Pro. Of course, the Radeon 9600 Pro isn’t totally neutered. The 0.13-micron RV350 GPU runs at a cool 400MHz, which is faster than even ATI’s flagship Radeon 9800 Pro.
Has ATI juggled pixel pipelines and clock speeds well enough for the Radeon 9600 Pro to inherit the mid-range graphics leader mantle from its sibling? Keep reading to find out.
What’s different?
The biggest news behind ATI’s RV350 graphics chip is, perhaops, the 0.13-micron process technology being used to fabricate the chip. The move to a 0.13-micron process allows ATI to fit more chips on a silicon wafer, which should make the RV350 a more profitable chip than its predecessors if yields are good. Moving to a 0.13-micron process also decreases the RV350’s power requirements, so cards built using the chip won’t need auxiliary power connectors and should run cooler than equivalent chips built using 0.15-micron technology.
Of course, a 0.13-micron manufacturing process doesn’t guarantee a cool-running graphics chip—just look at the GeForce FX 5800 Ultra’s Dustbuster. Fortunately, the RV350 does run cool. The Radeon 9600 Pro has one of the smallest active GPU heat sink/fan combos I’ve ever seen. In fact, I’d bet the chip will run at its full clock speed with only a slightly beefier passive heat sink.
Apart from the process technology used in its fabrication, the RV350 is very similar to ATI’s R300 chip, which powers the Radeon 9500 and 9700 lines. The RV350 has full support for DirectX 9, pixel and vertex shader versions 2.0, and floating point data types for pixels and texture. However, the 9600 incorporates some performance tweaks from ATI’s new R350 GPU (used in the Radeon 9800 Pro). There are a few key differences between the old R300 and the new RV350 worth noting.
- Fewer pipes – Ever-cooperative when it comes to divulging the internal structure of its graphics chips, ATI has revealed that the RV350 has four pixel pipelines, each of which is capable of laying down a single texture per rendering pass. Because it has only four pipelines, the RV350 also has only four pixel shaders. Clock for clock, the RV350 should have half the fill rate of ATI’s 8-pipe R3x0-series graphics chips and roughly half the pixel shader performance. To compensate for this pipeline disadvantage, ATI is ramping up the RV350’s clock speeds, which should be relatively easy to do given the chip’s 0.13-micron die.
- Memory controller optimizations – Like the Radeon 9500 Pro, the Radeon 9600 Pro’s memory bus is 128 bits wide, but the memory controllers used with each card are quite different. Because the R300 was originally designed for the Radeon 9700 Pro’s 256-bit memory bus, the chip didn’t make very efficient use of the Radeon 9500 Pro’s 128-bit memory bus. For the Radeon 9600 Pro, ATI has specifically optimized the RV350’s memory controller for a 128-bit memory interface. As a result, the Radeon 9600 Pro should make more efficient use of its available memory bandwidth than the Radeon 9500 Pro.
- 8:1 Z-compression – The RV350 is ATI’s first graphics chip to use 8:1 Z-compression, which should make the Radeon 9600 Pro even more efficient when it comes to memory bandwidth usage. By comparison, ATI’s R300 GPU peaked at a 6:1 Z-compression ratio.
- No F-buffer – Unlike ATI’s high-end R350 chip, the RV350 doesn’t have an F-buffer. Without an F-buffer, the Radeon 9600 Pro will be unable to handle shader programs more complex than what’s called for by Microsoft’s DirectX 9 spec, but it’s still unclear how much of a limitation that will be in the real world. It will probably be some time before games exploit all of DirectX 9’s shader capabilities, much less move beyond them. You can read more about what exactly an F-buffer is and what it does in our Radeon 9800 Pro review.
Now that we’ve gone over the important differences between ATI’s RV350 GPU and its precursor, let’s check out pictures of the Radeon 9600 Pro itself.
ATI’s Radeon 9600 Pro reference board. Notice the lack of a auxiliary power connector.
The Radeon 9600 Pro from the rear, and oddly not in red.
The Radeon 9600 Pro’s teensy weensy GPU cooler.
BGA memory chips from Samsung
Now, on to the benchmarks!
Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at three times, and the results were averaged.
As ever, we did our best to deliver clean benchmark numbers. Tests were run at three times, and the results were averaged.
Our test system was configured like so:
| System | ||
| Processor | Athlon XP ‘Thoroughbred’ 2600+ 2.083GHz | |
| Front-side bus | 333MHz (166MHz DDR) | |
| Motherboard | Asus A7N8X | |
| Chipset | NVIDIA nForce2 | |
| North bridge | nForce2 SPP | |
| South bridge | nForce2 MCP-T | |
| Chipset drivers | NVIDIA 2.03 | |
| Memory size | 512MB (2 DIMMs) | |
| Memory type | Corsair XMS3200 PC2700 DDR SDRAM (333MHz) | |
| Sound | nForce2 APU | |
| Graphics card | GeForce4 Ti 4200 8X | Radeon 9500 Pro Radeon 9600 Pro Radeon 9700 Pro |
| Graphics driver | Detonator 43.45 | CATALYST 3.2 |
| Storage | Maxtor DiamondMax Plus D740X 7200RPM ATA/100 hard drive | |
| OS | Microsoft Windows XP Professional | |
| OS updates | Service Pack 1, DirectX 9.0 | |
Unfortunately, the Radeon 9600 Pro reference card ATI sent wasn’t stable on my pre-production Asus A7N8X motherboard with AGP 8X enabled. The very same board works fine with AGP 8X enabled on the Radeon 9500 Pro, Radeon 9700 Pro, and GeForce4 Ti 4200 8X, but was unstable with the reference Radeon 9600 Pro. The fact that my A7N8X is a pre-production board made it seem like the most likely source of the problem. However, I found the Radeon 9600 Pro to be unstable (though much less so) with AGP 8X enabled on a production nForce2 board from Albatron, as well.
In the end, I tested the Radeon 9600 Pro on the A7N8X with AGP 4X. I also performed some additional testing with the Radeon 9600 Pro in Albatron’s nForce2 board at AGP 4X and AGP 8X speeds to see if there was any performance difference between the two. In short, there wasn’t. Not even 3DMark03 could tax the AGP bus enough to make use of the extra bandwidth offered by AGP 8X, so the Radeon 9600 Pro at AGP 4X probably won’t be at any significant disadvantage in our testing.
In the end, it’s important to note that I am dealing with a pre-production evaluation sample from ATI, too. The card isn’t even red, so it’s not exactly a finished retail product. I’m hopeful ATI will have any compatibility issues with production boards ironed out before they reach store shelves.
In order to keep a level playing field, image quality-wise, I tested the GeForce4 Ti 4200 8X with its “Application” image quality setting. The Radeon cards were tested using ATI’s “Quality” image quality option, which produces visuals roughly equivalent to NVIDIA’s “Application” setting.
The test system’s Windows desktop was set at 1024×768 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:
- FutureMark 3DMark 2001 SE Build 330
- FutureMark 3DMark03
- ChameleonMark
- Codecreatures Benchmark Pro
- Comanche 4 demo benchmark
- Quake III Arena v1.31
- Serious Sam SE v1.07
- VillageMark v1.17
- Unreal Tournament 2003 with 2199 patch
All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
Fill rate
By cutting the number of pixel pipelines down to four, ATI has potentially castrated the RV350. Let’s see how the combination of fewer pipe and higher clock speeds affects the RV350’s theoretical pixel-pushing power.
By cutting the number of pixel pipelines down to four, ATI has potentially castrated the RV350. Let’s see how the combination of fewer pipe and higher clock speeds affects the RV350’s theoretical pixel-pushing power.
| Core clock (MHz) | Pixel pipelines | Peak fill rate (Mpixels/s) | Texture units per pixel pipeline | Peak fill rate (Mtexels/s) | Memory clock (MHz) | Memory bus width (bits) | Peak memory bandwidth (GB/s) | |
| GeForce4 Ti 4200 8X | 250 | 4 | 1000 | 2 | 2000 | 512 | 128 | 8.2 |
| Radeon 9500 Pro | 275 | 8 | 2200 | 1 | 2200 | 540 | 128 | 8.6 |
| Radeon 9600 Pro | 400 | 4 | 1600 | 1 | 1600 | 600 | 128 | 9.6 |
| Radeon 9700 Pro | 325 | 8 | 2600 | 1 | 2600 | 620 | 256 | 19.8 |
As good as the Radeon 9600 Pro’s 400MHz core clock speed is for the card’s peak single and multi-texturing fill rates, it’s hard to compete with an eight-pipe design. The Radeon 9600 Pro does have the highest core clock speed of any of the cards we’ll be looking at today, but its 4×1-pipe architecture’s peak fill rates are well behind the Radeon 9500 Pro’s. Heck, even the GeForce4 Ti 4200 8X has a more impressive multi-textured fill rate than the Radeon 9600 Pro.
Although its peak theoretical fill rates are a little pokey compared to the competition, the Radeon 9600 Pro’s 600MHz memory bus (300MHz DDR) yields a healthy 9.6GB/sec of memory bandwidth. Let’s also remember that the Radeon 9600 Pro’s memory controller should make more efficient use of its available memory bandwidth than the Radeon 9500 Pro, thanks to ATI’s optimizations for a 128-bit memory interface.
Of course, the above chart only refers to theoretical peaks, which aren’t necessarily indicative of real world performance. Here’s how the cards stack up in fill rate benchmark tests.
It may only have four pixel pipelines, but the Radeon 9600 Pro has a slightly higher single-textured fill rate than the eight-pipe Radeon 9500 Pro. However, when we get into multi-texturing, the Radeon 9600 Pro’s fill rate is far behind even the GeForce4 Ti 4200 8X.
How does each card’s real world fill rate stack up against its theoretical peak?
ATI’s Radeon cards all appear to be quite inefficient when it comes to single-texturing, but they realize nearly all of their peak theoretical fill rates in multi-texturing situations. In 3DMark2001 SE’s multi-texturing fill rate test, the Radeon 9600 Pro realizes over 98% of its peak theoretical fill rate.
Occlusion detection
All of ATI’s R3x0-series graphics chips use a combination of Z-compression, fast Z clear, and Early Z occlusion detection to make efficient use of memory bandwidth and all but eliminate overdraw. The Radeon 9600 Pro shares all of these overdraw-reducing and bandwidth-saving features, but since its Z-compression ratio is 8:1 rather than 6:1, it should be slightly more efficient than ATI’s R300-derived graphics cards.
All of ATI’s R3x0-series graphics chips use a combination of Z-compression, fast Z clear, and Early Z occlusion detection to make efficient use of memory bandwidth and all but eliminate overdraw. The Radeon 9600 Pro shares all of these overdraw-reducing and bandwidth-saving features, but since its Z-compression ratio is 8:1 rather than 6:1, it should be slightly more efficient than ATI’s R300-derived graphics cards.
Unfortunately, 8:1 Z-compression doesn’t help the Radeon 9600 Pro much in VillageMark. Without antialiasing and anisotropic filtering enabled, the card is only as fast as NVIDIA’s GeForce4 Ti 4200 8X. With 4X antialiasing and 8X anisotropic filtering cranked up, the Radeon 9600 Pro easily beats out the GeForce4 Ti 4200 8X, but is still a step behind the Radeon 9500 Pro.
Pixel shaders
The Radeon 9600 Pro supports DirectX 9’s full pixel and vertex shader 2.0 specifications. However, because the RV350 has only four pixel pipelines, it has half as many pixel shader units as either the Radeon 9500 Pro or Radeon 9700 Pro. At 400MHz, the Radeon 9600 Pro’s pixel shaders run faster than those of the Radeon 9500 Pro or 9700 Pro, but can clock speeds alone make up the difference?
The Radeon 9600 Pro supports DirectX 9’s full pixel and vertex shader 2.0 specifications. However, because the RV350 has only four pixel pipelines, it has half as many pixel shader units as either the Radeon 9500 Pro or Radeon 9700 Pro. At 400MHz, the Radeon 9600 Pro’s pixel shaders run faster than those of the Radeon 9500 Pro or 9700 Pro, but can clock speeds alone make up the difference?
Not in 3DMark2001 SE. Here the Radeon 9600 Pro comes in just behind the Radeon 9500 Pro in the first pixel shader test and slightly further behind in the advanced pixel shader test. The Radeon 9600 Pro does, however, maintain a constant lead over NVIDIA’s Gefoce4 Ti 4200 8X throughout.
In NVIDIA’s own ChameleonMark pixel shader benchmark, the Radeon 9600 Pro is again wedged between the faster Radeon 9500 Pro and a distant GeForce4 Ti 4200 8X.
NVIDIA’s GeForce4 Ti 4200 8X doesn’t support 3DMark03’s pixel shader 2.0 test, so we couldn’t include it in this test. The Radeon 9600 Pro lags behind the Radeon 9500 Pro in this DX9-class test. Our test applications seem to prefer the eight slower pixel shaders on the Radeon 9500 Pro to the four faster shaders found on the Radeon 9600 Pro.
Vertex shaders
“Half as many as R300” seems to be ATI’s mantra for the RV350. So far, we’ve seen that the chip has half as many pixel pipelines and half as many pixel shaders as ATI’s R3x0 series. I bet you can’t guess how many vertex shaders it has. Ok, so maybe you can.
“Half as many as R300” seems to be ATI’s mantra for the RV350. So far, we’ve seen that the chip has half as many pixel pipelines and half as many pixel shaders as ATI’s R3x0 series. I bet you can’t guess how many vertex shaders it has. Ok, so maybe you can.
As you might expect, the Radeon 9600 Pro has only two vertex shader units—half as many as the Radeon 9500 Pro and Radeon 9700 Pro. Like everything else, the Radeon 9600 Pro’s two vertex shader units are running at a higher clock speed than the vertex engines found on the Radeon 9500 Pro and Radeon 9700 Pro, but they’re not running twice as fast.
The Radeon 9600 Pro’s twin vertex engines are fast enough nip at the heels of the Radeon 9500 Pro in 3DMark2001 SE’s vertex shader test, but 3DMark03 is much less forgiving. Still, the Radeon 9600 Pro is much faster than NVIDIA’s GeForce4 Ti 4200 8X.
Next up are some legacy transform and lighting tests, which should be handled by vertex shader programs on the graphics cards we’re looking at today.
Despite the fact that its vertex engines are clocked at faster speeds than any other graphics cards we’re testing today, the Radeon 9600 Pro takes a beating in 3DMark2001 SE’s transform and lighting tests. In both the one and eight light tests, the card is beaten by even NVIDIA’s GeForce4 Ti 4200 8X.
Quake III Arena
All right, enough with the synthetic stuff. Quake III Arena kicks off our real-world gaming tests. We’re using id’s own four.dm_66 timedemo.
All right, enough with the synthetic stuff. Quake III Arena kicks off our real-world gaming tests. We’re using id’s own four.dm_66 timedemo.
Surprisingly, the Radeon 9600 Pro hangs right with the Radeon 9500 Pro in Quake III Arena, despite its pixel pipeline disadvantage. Even with 4X antialiasing and 8X anisotropic filtering enabled, the Radeon 9600 Pro is just a hair behind the Radeon 9500 Pro, proving that ATI may have known what it was doing when it chopped the pipelines in favor of higher clock speeds.
Jedi Knight II
The pack really doesn’t spread out in Jedi Knight II until 4X antialiasing and 8X anisotropic filtering are enabled. With these image quality options turned on, the Radeon 9600 Pro trails the Radeon 9500 Pro, but it does manage an impressive 55 frames per second at 1600×1200.
Comanche 4
The Radeon 9500 and 9600 Pro are closely matched in Comanche 4 with and without antialiasing and anisotropic filtering enabled, but the eight-pipe Radeon 9500 Pro still manages to lead its apparent successor. With the Radeon 9500 Pro and 9600 Pro’s scores so close, I have to wonder how things might play out if the Radeon 9600 Pro had an additional 50 or 100MHz worth of core clock speed on its side.
Codecreatures Benchmark Pro
The Radeon 9600 Pro flirts with the bottom of the barrel in Codecreatures. Without antialiasing and anisotropic filtering enabled, the Radeon 9600 Pro just edges out NVIDIA’s GeForce4 Ti 4200 8X. When we turn on 4X antialiasing and 8X anisotropic filtering, the Radeon 9600 Pro slips behind the GeForce4 Ti 4200 8X for a disappointing last-place finish.
Curiously, the Radeon 9500 Pro produced better scores at a resolution of 1600×1200 than it did at 1280×1024 with antialiasing and anisotropic filtering enabled. That score was suspect at best, so we omitted it from the results.
Unreal Tournament 2003
Unreal Tournament 2003 is our most advanced real world benchmark, and it’s likely to be the king of 3D eye candy and game engine of choice until Doom 3 is released.
Unreal Tournament 2003 is our most advanced real world benchmark, and it’s likely to be the king of 3D eye candy and game engine of choice until Doom 3 is released.
Without antialiasing or anisotropic filtering, the Radeon 9600 Pro manages to stay close to the Radeon 9500 Pro. What happens when we ratchet up the image quality?
The GeForce4 Ti 4200 8X falls off the pace, but the Radeon 9600 Pro keeps going strong, relatively speaking. Only at our highest resolution of 1600×1200 does the Radeon 9600 Pro really fall behind the Radeon 9500 Pro; at such a high resolution, it simply doesn’t have the fill rate to keep up.
Serious Sam SE
We used Serious Sam SE’s “Extreme Quality” add-on to ensure a mostly level playing field between the different graphics cards. In these tests, the Radeons are actually doing 16X anisotropic filtering, while the GeForce4 Ti 4200 8X is only doing 8X anisotropic filtering. However, the difference between the images produced by the two is extrememly slight.
We used Serious Sam SE’s “Extreme Quality” add-on to ensure a mostly level playing field between the different graphics cards. In these tests, the Radeons are actually doing 16X anisotropic filtering, while the GeForce4 Ti 4200 8X is only doing 8X anisotropic filtering. However, the difference between the images produced by the two is extrememly slight.
The Radeon 9600 Pro is left in last place without any extra eye candy enabled, but it’s not too far behind the Radeon 9500 Pro.
Of course, no graphics chip review would be complete without a peek at Serious Sam SE’s frame rates over the course of the demo we’re using for testing. Rather than relying on simply average frame rates to gauge performance, Serious Sam SE lets us look for extreme frame rate dips that could disrupt smooth gameplay.
The Radeon 9600 Pro doesn’t show any alarming frame rate dips, but its performance is consistently below that of the Radeon 9500 Pro throughout the test.
With 4X antialiasing and 8X anisotropic filtering, the Radeon 9600 Pro rises up above the GeForce4 Ti 4200 8X, but it still can’t catch the Radeon 9500 Pro.
As we look at the Radeon 9600 Pro’s performance over time, it’s clear that the card has some serious problems keeping frame rates up at 1600×1200. At lower resolutions, the Radeon 9600 Pro’s performance consistently follows that of the Radeon 9500 Pro, but it can’t handle higher image quality settings at high resolutions.
3DMark2001 SE
The Radeon 9600 Pro falls to last place overall in 3DMark2001 SE. Let’s look at home the indvidual game tests came out.
In all but Game 3’s Lobby scene, the Radeon 9600 Pro trails even the GeForce4 Ti 4200 8X.
3DMark03
The Radeon 9600 Pro’s overall score in 3DMark03 closely matches that of the Radeon 9500 Pro. How does the Radeon 9600 Pro handle 3DMark03’s array of simulated game tests?
Pretty well, at least when compared with the Radeon 9500 Pro. The two mid-range Radeons are closely matched until we get to the pixel shader-packed Mother Nature test, where the Radeon 9600 Pro falls a little further behind.
SPECviewperf
SPECviewperf is our lone workstation-class graphics test. I threw it in just for kicks.
SPECviewperf is our lone workstation-class graphics test. I threw it in just for kicks.
The Radeon 9600 Pro screams through the ugs test and ends up at the top of the pile, but sticks close to the Radeon 9500 Pro in the rest of the viewperf suite.
Antialiasing
Next, we’ll isolate the Radeon 9600 Pro’s performance in different antialiasing and anisotropic filtering modes. We’ve already had a glimpse of the card’s performance with 4X antialiasing and 8X anisotropic filtering in our game tests, but this section will give us a more thorough look at the card’s performance with these image quality features.
Next, we’ll isolate the Radeon 9600 Pro’s performance in different antialiasing and anisotropic filtering modes. We’ve already had a glimpse of the card’s performance with 4X antialiasing and 8X anisotropic filtering in our game tests, but this section will give us a more thorough look at the card’s performance with these image quality features.
Edge antialiasing
The Radeon 9600 Pro keeps up with the Radeon 9500 Pro until we get into 6X antialiasing or a display resolution of 1600×1200. At lower resolutions and antialiasing modes, the cards are very closely matched.
Texture antialiasing
To measure texture antialiasing, I used Serious Sam SE with various texture filtering settings.
To measure texture antialiasing, I used Serious Sam SE with various texture filtering settings.
Overall, the Radeon 9600 Pro trails the Radeon 9500 Pro, but not by much. Even at a relatively high resolution of 1280×1024, the Radeon 9600 Pro has enough power to hang with its eight-pipe sibling.
Conclusions
According to ATI, the Radeon 9600 Pro will ship by the end of the month and be available for between $169 and $199, which makes it price-competitive with the Radeon 9500 Pro cards currently on the market. ATI will also be shipping a vanilla Radeon 9600 with core and memory clock speeds of 325 and 400MHz, respectively. Given that the non-Pro Radeon 9600 will share the same 0.13-micron RV350 graphics chip, it could be an impressive overclocking candidate. Unfortunately, the Radeon 9600 Pro just doesn’t have what it takes to go toe-to-toe with the Radeon 9500 Pro. I can’t say I’m surprised, though. The Radeon 9600 Pro may have slightly better Z-compression and faster core and memory clock speeds than the Radeon 9500 Pro, but it’s just not enough to overcome having half as many pixel pipelines and shaders. That said, the performance ATI has wrung from a four-pipe rendering architecture is certainly impressive, especially since the 400MHz graphics chip requires only a whisper of a heat sink and fan to keep cool.
According to ATI, the Radeon 9600 Pro will ship by the end of the month and be available for between $169 and $199, which makes it price-competitive with the Radeon 9500 Pro cards currently on the market. ATI will also be shipping a vanilla Radeon 9600 with core and memory clock speeds of 325 and 400MHz, respectively. Given that the non-Pro Radeon 9600 will share the same 0.13-micron RV350 graphics chip, it could be an impressive overclocking candidate. Unfortunately, the Radeon 9600 Pro just doesn’t have what it takes to go toe-to-toe with the Radeon 9500 Pro. I can’t say I’m surprised, though. The Radeon 9600 Pro may have slightly better Z-compression and faster core and memory clock speeds than the Radeon 9500 Pro, but it’s just not enough to overcome having half as many pixel pipelines and shaders. That said, the performance ATI has wrung from a four-pipe rendering architecture is certainly impressive, especially since the 400MHz graphics chip requires only a whisper of a heat sink and fan to keep cool.
So why would ATI bother releasing a slower replacement for the Radeon 9500 Pro? Margins, baby. The RV350 weighs in at only around 60 million transistors—roughly half as many as the R300. Combine that with a 0.13-micron fabrication process, and ATI is looking at a very small chip. If margins are good, and by all accounts they are, ATI will probably make a pretty penny on the Radeon 9600 Pro.
But there’s more to the Radeon 9600 Pro than the bottom line. The RV350 is ATI’s first desktop graphics processor to use a 0.13-micron fabrication process. It’s a test case, if you will, and one that has apparently gone much smoother than anticipated. Things have gone so well with the RV350 that ATI actually moved up the Radeon 9600 Pro’s release date. How often do you actually hear of a hardware company releasing a product ahead of schedule?
It’s also important to note that ATI is no stranger to releasing products which are slower than their predecessors. When ATI initially launched the Radeon 9000 Pro, it was slower than the Radeon 8500LE cards that it replaced. The newer Radeon 9000 Pro actually sent Radeon 8500LE prices spiraling downward, yielding fabulous deals for those lucky enough to snap up Radeon 8500LE cards before stocks ran out. That could happen again this time around as retailers cut prices on the older Radeon 9500 Pro to clear inventory for the new Radeon 9600 Pro.
Try as it might, ATI hasn’t topped the Radeon 9500 Pro’s performance with the Radeon 9600 Pro, but maybe they don’t need to. The Radeon 9600 Pro is impressively efficient and a poster child for 0.13-micron graphics chips done right. What’s more, it’s consistently faster than the GeForce4 Ti 4200 8X, and should stack up well against NVIDIA’s 4×1-pipe GeForce FX 5600s when they eventually materialize.
In the end, there’s really no need to go out and buy a Radeon 9600 Pro today. The Radeon 9500 Pro is faster and available now, though stocks will dwindle over the next couple of months as ATI phases out the card. If you do miss the Radeon 9500 Pro boat, don’t worry. The Radeon 9600 Pro is a pretty snazzy substitute.
The lack of the f-buffer in the 9600 is somewhat concerning, in the purely theoretical sense. Having different chips in the same “family” with different DirectX capabilities, that is. However, since no one will start using an f-buffer for 3 years, probably not a big deal.
The difficulty in running AGP 8X is more worrisome.
Performance-wise, it seems about right, considering it’s missing four pipes.
ER
Wow, that FX just looks better and better. Having a software-controlled fan just to keep it quiet for 2-D apps is icing on the cake. nVidia should scrap the GPU and just start over….
Damn, I would have edited this into that last post if I could have–
The IDEAL place for the new ATI .13 process GPU will be in boxes such as my SN41G2, which has very little clearence between the case cover and the AGP slot; I did a little temperature testing with my 9500 Pro and submitted it to Overclockers here:
§[<http://www.overclockers.com/articles727/index04.asp<]§ I'm pretty sure that a GFFX card would end up as a pile of slag in this SFF box...
GeForceFX 5800 Ultra will be slag in any box if you run a 3D screen saver:
§[<http://www.pcekspert.com/articles/127-1.html<]§
…when I refer to “Joe Sixpack”, I mean to say as a father buying a computer for a son who knows what gpu power (through advertising/web buzz) that new games need to run (as in “I need a 9600, Dad, all of the websites like it”). Joe Sixpack usually could care less. There I go again, butchering the “King’s English”.
Corrado: Yes, I caught that; it’s just that with fewer pipelines, the stock clock speed leaves the card at a slight /[
I’m confused what to buy, in extreme tech, they report that the 9600 pro is actually faster than the 9500 pro. If that was true, then there would be no reason to buy the 9500 because the 9600 is just as good and has less noise/heat
[H]OCP overclocked their sample to 567/716 (DDR)! Sheez! With all of the trouble that ATI went to to lock the 9500/Pro, one wonders what’s up their sleeve… hmm. Well, I think that as the enthusiasts start snapping up 9600’s (because of the possibility of OC’ing), Joe Sixpack will follow. Makes me wonder if these “review samples” might be selected expressly for this purpose.
In the OCP review, it says Kyle asked ATi specifically about clock locking and they said they would NOT clock lock the 9600s.
§[<http://www.beyond3d.com/reviews/ati/rv350/index.php?p=22<]§ Why the hell is ATI doing? Why not set the clock to 450 or higher?
Passive cooling is an acceptable design option according to ATI. With a higher core speed, that might be more difficult to do.
It’s also possible that ATI targeted 400MHz when the silicon was pulling lower than that and 400 then seemed a reasonable target freq, but the latest core revision actually ended up clocking higher than expected. Perhaps if their early silicon had acheived a higher initial clock, they’d have set their target clock higher.
Beyond3D also suggests that maybe ATI simply isn’t speed binning the RV350 chips — if it works at 400 it goes on a board, regardless of what freq it can actually pull. This way they can sell all working chips, rather than risk selling out of a higher bin and being stuck with chips from a lower bin. If so, perhaps we can look forward to 3rd party manufacturers binning their chips and differentiating their boards by clock speed. 🙂
Pete, your right. They are over-clocking the 9600pro to insane levels with stock cooling. And the cooling on the 9600pro is minimal too. Imagine what ATI’s vendors will do with this card and better cooling systems. Damn. Well over 530 i assume…..
Diss, the RV350 is also lacking Hier-Z and some internal cache, which helps explain the slimmer transistor count. I still think it’s a great card if you factor in overclocking. Reviewers are hitting 530MHz with stock cooling!
And all the while GF FXs are cookin’. Nice Job Nvidia.
§[<http://www.pcekspert.com/articles/127-1.html<]§
I bought a 9500 Pro… thanks, ATI for not making me regret it 3 months later!
Great article Diss, nice reading. I can see the point behind this card, makes a lot of sense for ATI. The performance difference is next to zero, even tho there’s no increase overall. It’s ATI’s Tualatin, 0.13 in volume before moving other things over to the new process.
q[<4) Profit!<]q §[<http://www.dmwright.com/html/ferengi.htm<]§
i am glad I got the faster 9500 Pro card. it runs stable at 300/300 so I am not complaining.
Who cares ATI drivers still blow dead goats
Not really. ATI drivers are now easily a match for Nvidia ones, as the number of Nvidia issues and ‘features’ has been rising steadily for the last year or so. I’d imagine there are more documented problems with 43.45 than with Cat 3.2.
And this one §[http://www.xbitlabs.com/news/video/display/news6149.html<]§
#11 Ok, maybe my use of the word cripple was a bad choices.
But…9500 to 9700
§[<http://www.xbitlabs.com/news/video/display/news6142.html<]§ and §[<http://www.xbitlabs.com/news/video/display/news6115.html<]§ #12 Yeah, along the lines of what I was thinking. the 9600 is a complete different chip.
Shouldn’t they have named this the 9400 slower then the 9500.. omg ATI’s marketing department.. trying to throw this over the heads of the consumers.. the shame…
So the Consumer’s lose, us enthusiasts win with the rush to buy 9500s. I’ll actively be talking my neighbors into giving me the 9500 and saying the 9600 is so much “better”g{
Well, this has been done before… remember the Radeon 9000? it is slower than a 8500. Another example (and the one I hate the most), GF4 MX…. much slower than a GF3 and not even with all the capabilities :me roll eyes:
I remember a lot of folks buying GF4 MX at the beginning because they thought it’d be better than the GF3… At least this new Radeon is somehow close to the performance of the 9500
well wouldn’t a GF2MX be slower then a GF256 then? or is it.. can’t remember.. the price difference between a GF4mx and GF3ti500 was huge. This 9600 how much is it selling for? ~ the same as a 9500?
/[< Another example (and the one I hate the most), GF4 MX.... much slower than a GF3 and not even with all the capabilities :me roll eyes: I remember a lot of folks buying GF4 MX at the beginning because they thought it'd be better than the GF3... <]/ Then those people had ridiculous expectations and got what they deserved. Even the briefest amount of research would have told them otherwise. When the GF3 first came out it cost $300. When the GF4TI/MX first were released, the high end Ti cost $300 while the MX cost ~$150. To look at those price points and expect that the MX would compete with the GF3 is foolish. I understand your frustration with the Ti/MX nomenclature, but lets compare apples to apples. Compare last generations card with its replacement. The GF4Ti4600 replaced the GF3Ti500 at the same price point ($300) and clearly had better performance. The 9600Pro replaces the 9500Pro at the same price point ($200) and has equivalent or diminished performance. All in all, the review seems to have gotten it right. Paraphrased 1) Create part on new fab process 2) Ensure performance doesnt lag too far behind previous product 3) Charge the same $$ 4) Profit!
You just haven’t been paying attention for the past six months. (Maybe you were distracted by all of the hype around GeForceFX 5800 Ultra.)
Radeon 9500 y{<*[
Speed isn’t everything and it’s not like it’s a large step backwards. Without benchmarks, you wouldn’t notice.
24: I wouldnt be so drastic. Many people today like to use AA and AF.
True, but I still challenge someone to sit in front of indentically specced boxes bar the video card, and choose the quicker one out of a 9500 Pro and 9600 Pro, even with AA and AF enabled.
Given a decent base spec, that would be a fairly hard test for most people I think.
the good news is that ati transitioned well to .13 process with promising results
so i think that leads me to believe that they wont have dustbuster cooler problems like nVidia did when they switched (albiet they did the swithc much closer to the potential of their best core)
looking forward to the new .13 process chips after the 9600, they should have nicer clock frequencies with all the high end pro feature set, and o/c relatively well *drool*
#5 I dont think tranistor count has anything to do with it. The reason why the 9500/pro to 9700/pro hack works is because the 9500/pro is actually a crippled 9700/pro.
Your point is contradictory. The only reason you can change a 9500 into a 9700 is because half of the pipelines (a significant amount of the transistors) are disabled. The RV350 is not artificially “crippled” in any way.
BTW, the only thing a 9500Pro can be modded into is a 9500 non pro.
Exactly — the 9500 Pro used the same chip as the 9700 pro, but with stuff disabled. If you could re-enable those parts, you’d get a 9700 out of the deal.
The 9600 Pro and the 9800 Pro, however, use different chips; the 9600 Pro has a ‘mere’ 60 million transistors while the 9800 Pro has somewhere over 100 million. Nothing is ‘disabled’ on the 9600; the extra parts just aren’t there. There’s nothing to re-enable to turn a 9600 into a 9800.
As such, transistor count has *everything* to do with it! 🙂
Me, I’d agree with the review. I’d pick up a 9500 Pro if I was in the market right now, but I wouldn’t worry too much if I put it off for a few months and then can only get a 9600 Pro — especially if the 9600 Pro overclocks well thanks to the .13 process…. 🙂
Does anyone know if ATI will be making a .13 version of the R350, or are they waiting for the R400 to release a flagship .13 chip?
I heard something about there being a R3x0 (380 or 390?) that’s supposed to come out this fall because R400 most likely won’t be out until 2004. I think it’s a .13 shrink of the R350 with GDDR-II.
i thought the FX 5800 Ultra wasnt going to make it anyway…
Maybe they will figure out a 9600 to 9800 hack, like they did for the 9500 to 9700.
Yeah! A simple software hack will magically double the tranistor count!
When pigs fly
I only wish you guys had a vanilla 9700 to run it up against.
Regarding the pricing of 9500 Pro, I expect that it will remain right where it is as people start snapping them up now that it’s been confirmed that it can’t outrun the 9500 Pro.
Hmmm, why bother buying the 9500 pro when this card abviously takes a lot less power and likely runs quiter? In my opinion this is worth as much as performance, if not a little more. This will be interesting when they use this process on the 9700 core, then we may see a large puddle of sweat creeping under the engineering doors of nVidia, and secretive resumes being mailed to ATI.
-ls
Hey…that looks like an 8500
lol, that’s exactly what I thought.