Evolution of ATI’s Catalyst Driver for Radeon HD 4870

In this review we will benchmark the Radeon HD 4870 graphics card with different versions of the Catalyst driver and also cover the image quality aspect.

by Sergey Lepilov
01/13/2009 | 04:55 PM

In my previous article I told you about the effect of Nvidia’s GeForce driver on the performance of the GeForce GTX 260 (216SP) graphics card under Windows Vista x86. And I promised then that I would write another such review about ATI’s Catalyst driver and Radeon HD 4870 graphics card. For some reasons Catalyst has been criticized much more often than Nvidia’s driver by the users. The critics have got even louder since the release of the highly popular RV770-based graphics card series, complaining not only at the inexplicably low performance of the cards on specific levels of games but also at the abundance of various glitches. Therefore, besides just checking the speed of the card with different driver versions, I will tell you about any image artifacts I notice during my tests. I will begin this review by describing the graphics card and offering you the list of changes in the drivers from the moment the Radeon HD 48xx series was released.

HIS Radeon HD 4870 IceQ 4+ Turbo 1GB

Package and Accessories

 

The small box of this card is designed in HIS’s traditional style. You can read all the basic information you may want to know about the product from the face and back sides. There is a cutout in the back of the box for you to take a look at the card’s cooling system.

 

Inside the colorful wrapper there is a thick cardboard box containing a plastic tray with the card and accessories. Here they are:

Unfortunately, the kit doesn’t include video cables or games or even the exclusive multifunctional screwdriver from HIS. This is odd because these accessories (save for games) are all supplied together with the cheaper HIS Radeon HD 4870.

PCB Design

Unlike the reference Radeon HD 4870, the HIS Radeon HD 4870 IceQ 4+ Turbo has a sea-green PCB. Its face side is covered by the cooler:

The reverse side is open for viewing:

The card measures 244 x 100 x 35mm, exactly like the reference sample. As a result, the cooling system will block the neighboring PCI slot on the mainboard. This is not a serious problem, however, because all modern top-end graphics cards come with dual-slot coolers.

The HIS Radeon HD 4870 IceQ 4+ Turbo is equipped with a PCI Express x16 version 2.0 interface, two dual-link DVI-I ports with support for high resolutions, and an S-Video output.

Unlike the reference Radeon HD 4870, this card has gold-plated DVI connectors to minimize electromagnetic interference.

Having removed the cooler, I found that this card uses a reference PCB, the color of the textolite being the single difference.

There are two 6-pin connectors for additional power on the PCB:

According to its specs, the Radeon HD 4870 512MB consumes up to 170 watts of power. A 500W power supply is recommended for a computer with it. A 600W or higher power supply is recommended for an appropriate CrossFire configuration.

Underneath the cooler and a thick layer of gray thermal grease there is an RV770 processor manufactured in Taiwan in early September (37th week of 2008).

The die is 256 sq. mm large, incorporating about 956 million transistors. The GPU has 800 unified shader processors, 40 texture-mapping units and 16 raster back-ends. The GPU frequency of the card is 770MHz, which is 20MHz (6.7%) higher than the reference GPU frequency.

The HIS Radeon HD 4870 IceQ 4+ Turbo has two times as much memory as the reference Radeon HD 4870, i.e. 1024 megabytes. Its eight GDDR5 memory chips were manufactured by Qimonda on the 40th week of 2008.

These revision A1 chips are marked as IDGV1G-05A1F1C-40X. They have an access time of 1.0 nanosecond and a rated frequency of 4000MHz. This is indeed the frequency the memory is clocked at by the card (200MHz or 5.3% higher than the memory frequency of the reference HD 4870).

Thus, the HIS Radeon HD 4870 IceQ 4+ Turbo comes with a double amount of graphics memory and slightly pre-overclocked frequencies.

Cooling System

Besides, the card is equipped with an IceQ 4+ cooler that proves to be a copy of the reference cooler with a somewhat different blower.

The GPU heatsink and the metallic plate that cools the memory chips and the power components are the same as those of the reference heatsink:

The card’s temperature was measured in a simple test. I loaded it by running the Firefly Forest test from 3DMark06 at 1920x1200 with 16x anisotropic filtering for 10 times. I didn’t enable FSAA because the GPU load and temperature would have been lower then. The test was performed in a closed ASUS Ascot 6AR2-B system case (its fan configuration is described below in the Testbed and Methods section). The ambient temperature was 23.5°C. The card’s frequencies and temperature were monitored with RivaTuner 2.20. As I had dismantled the card before testing it, I replaced the thermal interface of the GPU with a thin layer of high-efficiency Gelid GC1 thermal grease.

So, here are the results of the test with the card working with automatic fan speed management.

There are a few important things that must be noted. First, the GPU frequency of the card is not reduced to 500MHz in 2D applications as on reference Radeon HD 4870. People at HIS seem to have forgotten about this power-saving capability of the chip when they pre-overclocked it. And second, the blower works at a very high speed and switches from one speed to another in a jump (the specific speed depends on the GPU temperature). As a result, the card is rather noisy while the sudden changes in the fan’s speed are annoying. The imperfect BIOS is a sad and serious fault of the company’s engineers.

So, I decided to finish the card’s BIOS off manually. Using Radeon BIOS Editor version 1.18 and the original BIOS of the HD 4870 IceQ 4+ Turbo (a 44.5KB WinRAR file), I reduced the GPU frequency in 2D mode from 770MHz to 250MHz and adjusted the GPU voltage from 1.263V to 1.083V. I also typed in the frequency of 500MHz for the UVD mode (video playback), just like on the reference Radeon HD 4870. Besides changing the frequencies and voltages, I also corrected the operation of the blower so that its speed varied smoothly depending on the GPU temperature (from 30 to 100°C).

 

As the result of my tweaking the card’s BIOS, the GPU temperature grew up by 2°C under load but the fan’s speed and noise lowered considerably.

The monitoring graph shows that the speed of the cooler’s fan now changes smoothly within a maximum of 2500rpm and a minimum of 1450-1500rpm (for 2D mode). You can download the edited BIOS, too (a 43.3KB WinRAR file). I want to warn you that if your BIOS update fails, neither I nor X-bit Labs won’t be responsible for the consequences. You can take a look at the graph below to see the temperature of the graphics card at the full speed of its fan (about 5300rpm).

Unfortunately, the card’s GPU proved to have low overclocking potential. I could only overclock it from its default 770MHz to 785MHz, i.e. less by 2% (+4.7% to the GPU frequency of the reference HD 4870). The memory chips did much better. I overclocked them from 4000MHz to 4600MHz (15% above the default frequency and 21.1% above the memory frequency of the reference HD 4870) without any instability or image artifacts.

The card’s temperature didn’t change much at the overclocked frequencies although the blower got about 200rpm faster and somewhat louder under load.

The last thing I want to add about the card is that its cooler is covered with ultraviolet-sensitive paint.

The HIS Radeon HD 4870 IceQ 4+ Turbo 1024MB sells at about $310 in retail whereas the recommended price of the Radeon HD 4870 512MB is $269.

Evolution of ATI Catalyst Driver

Catalyst 8.7 (07.21.2008) is the first official drive for the Radeon HD 48xx series. It introduced a lot of corrections for a dozen of then-popular games and was meant to increase performance in the following benchmarks and games:

Besides, the driver’s control panel (Catalyst Control Center) was updated and the operation of the distributed computing client Folding@Home was improved.

The Catalyst 8.8 driver released in the middle of August did not bring about any performance-related improvements. This release was meant to improve Avivo technology (processing of video content) in the first place. At the same time, the driver claimed to correct various issues in as many as 18 games for Windows Vista and XP. The subsequent Catalyst 8.9 driver (09.17.2008) was not meant to make graphics cards faster, either. It introduced minor improvements into Catalyst Control Center and partial support for OpenGL 3.0. This version corrected fewer bugs in games than the previous version. Since the driver developers did not declare any performance-related improvements, I will not test these drivers in this review.

Catalyst 8.10 (10.15.2008) was not declared to provide any improvements in terms of speed as you can learn from the official release notes. However, user reports and my personal experience suggest that this driver ensures performance benefits in popular benchmarks as well as in games. Therefore, Catalyst 8.10 is included into the test list. This version is not declared to have many bug corrections. Most of them are limited to processing video. New in this version is the option of controlling the graphics card’s fan speed from the driver’s Control Panel.

Catalyst 8.11 (11.12.2008) is the fifth official driver since the release of the Radeon HD 48xx series. It had been preceded by at least two beta versions targeted at increasing performance in newer games. The release notes promise a 3-10% performance growth in Far Cry 2 on Radeon HD 3xxx and 4xxx series in CrossFire mode and a 6-10% performance growth in single-card mode. A performance gain of 10-30% is also promised for Radeon 3xxx and 4xxx cards in CrossFireX mode running S.T.A.L.K.E.R. This release also corrected some bugs in modern games.

Catalyst 8.12 (12.10.2008) is the long-anticipated driver with lots of performance-related improvements. Here they are:

The list is long and impressive indeed. Besides, this driver corrects some bugs and optimizes the processing of video content.

That’s how the ATI Catalyst driver has been evolving since the introduction of the Radeon HD 48xx series. The next section describes the testbed I performed my tests on.

Testbed and Methods

The graphics card was benchmarked in a system case with the following configuration:

To minimize the CPU’s influence on the graphics card’s performance I overclocked the quod-core CPU to 4.00GHz at 1.575V voltage before the tests.

The system memory worked at a frequency of 1000MHz with 5-4-4-12 timings (Performance Level = 6) and 2.175V voltage.

The tests were run under Windows Vista Ultimate Edition x86 SP1 (with all the critical updates available for December 19, 2008). I used the latest drivers available at the moment of my tests:

The drivers were tested in the order of their release. Each driver was installed only after the previous one had been uninstalled and the system had been cleaned with Driver Sweeper 1.5.5. The following changes were made in Catalyst Control Center: the graphics quality level was changed from Quality to High Quality and the Adaptive Antialiasing option was set at Quality. Vertical synchronization was Always Off. Other settings were left at their defaults. I turned full-screen antialiasing and anisotropic filtering on from the menu of each game. If the game didn’t provide such options, I enabled FSAA and AF from the driver’s Control Panel.

The graphics cards were tested at two resolutions, 1280x1024 and 1920x1200, and with or without 16x anisotropic filtering and 4 or 8x full-screen antialiasing in the following games and applications:

I tested the cards twice in each application (do not confuse this with a double run of a demo). The final result is the best fps/score value out of the two cycles. It is shown in the diagrams. The bottom speed is also shown whenever possible.

Performance

The drivers are sorted by their release date in the diagrams. The earliest version (8.7) is colored dark blue; the 8.10 and 8.11 drivers are colored red; and the new 8.12 driver is violet. That’s how it looks in the diagrams:

Two synthetic benchmarks come first.

3DMark 2006

Here you can note the advantage of versions 8.7 and 8.12 over the other the others in the AF + FSAA mode. This advantage can be observed in terms of overall score as well as in the HDR/SM3.0 test score. I did not find any difference in image quality between the fastest and slowest driver (3DMark gives you the opportunity to analyze specific frames in detail).

3DMark Vantage

The first official driver for the RV770-based product series is the only one that fails in 3DMark Vantage. The other three versions released in 2008 deliver comparable performance. Catalyst 8.12 has negligible advantage over versions 8.10 and 8.11 in all of the 3DMark Vantage tests. There were image defects (loss of textures and distortions of the water surface) in the first test with Catalyst 8.10.

Unigine Tropics Demo

Like in 3DMark Vantage, Catalyst 8.12 is ahead and the gap is even wider.

World in Conflict

There is no clear winner in World in Conflict but we can note that the frame rate is the highest with Catalyst 8.12 in three out of the four test modes.

Enemy Territory: Quake Wars

Catalyst 8.7 falls behind in this game whereas the three newer drivers deliver identical performance.

Call of Duty 4: Modern Warfare MP

There is nothing to comment in Call of Duty 4: Modern Warfare MP as the drivers do not differ much.

Unreal Tournament 3

We are promised an 18% performance growth in Unreal Tournament 3 with Catalyst 8.12 when full-screen antialiasing is enabled. And the frame rate is indeed higher, even though not as much higher as promised: about 6% at 1280x1024 and 11% at 1920x1200 relative to the Catalyst 8.11 results. That’s good.

Devil May Cry 4

This test indicates that ATI had been unprepared for the Radeon HD 4870 release in terms of software support. The much lower performance of the card with the first official Catalyst is due to the absolute lack of useful optimizations in the driver. I say useful because I did not spot any difference in image quality between Catalyst 8.10-8.12 and Catalyst 8.7 in any of the four tests of the game’s built-in benchmark.

S.T.A.L.K.E.R.: Clear Sky

You see a small but almost linear growth of the card’s performance upon installing a newer version of Catalyst at a resolution of 1280x1024. The developer promised a 10% performance gain with Catalyst 8.12 for CrossFire configurations only, though.

Crysis Warhead

There is a small performance growth in Crysis Warhead, too, but not in every mode.

Far Cry 2

The new Catalyst 8.12 driver delivers an 8 to 11% performance growth in Far Cry 2, and I did not see that the image quality got any worse with it in the benchmark or in the game (I checked this out on three levels of the game). At the same time, there were occasional slowdowns in the benchmark with every version of the driver. Such slowdowns are less frequent, but still uncomfortable, during actual play.

X3: Terran Conflict

Left 4 Dead

There is nothing to comment in X3: Terran Conflict and Left 4 Dead.

Lost Planet: Colonies

Catalyst 8.12 is faster than the older driver versions in Lost Planet: Colonies. I don’t publish the test results without FSAA and AF because they are almost the same as in the diagram above.

Grand Theft Auto IV

I used the following graphics quality settings in this game:

I ran the game’s integrated benchmark twice at each resolution, writing down the best frame rate out of the two runs.

Here are the results:

This is similar to the Far Cry 2 results. Catalyst 8.12 ensures a small performance growth. Versions 8.10 and 8.11 have identical results, and Catalyst 8.7 is slower then both of them. Talking about image defects, textures would disappear suddenly with every version of the driver. That’s an annoying bug.

Image Quality and Speed in 3DMark 2006

Like in my GeForce driver review, I want to check out the correlation between the graphics card’s performance and image quality using 3DMark 2006. I ran the last two tests (that support Shader Model 3.0), changing the image quality setting from High Performance to High Quality in the Catalyst driver’s Control Panel. Catalyst IA was left at the Standard position. Then, I enabled anisotropic filtering and three levels of full-screen antialiasing (2x, 4x, 8x) using 3DMark06’s settings. The tests were performed with Catalyst 8.12 at a resolution of 1920x1200.

Here is how the quality of the Canyon Flight scene changes (frame 1350):

High performance

Performance

Quality

High quality

HQ + AF16x

HQ + AF16x+AA2x

HQ + AF16x+AA4x

HQ + AF16x+AA8x

And here is how the graphics card’s speed changes depending on the image quality mode:

And now the screenshots for the second part of the diagram above, captured in the Deep Freeze test (frame 1150):

High performance

Performance 

Quality

High quality

HQ + AF16x

HQ + AF16x+AA2x

HQ + AF16x+AA4x

HQ + AF16x+AA8x

The two sets of screenshots suggest that the card’s performance does not change in the two 3DMark06 tests when the quality mode is adjusted from High Performance to High Quality in the driver but the image quality varies greatly. The game looks even better and more detailed when you turn on 16x anisotropic filtering but the frame rate lowers then: the performance hit is less dramatic in Deep Freeze than in Canyon Flight. Full-screen antialiasing, on the contrary, lowers the frame rate in the Deep Freeze more than in the other test. The image quality grows up steadily with 2x through 8x FSAA.

Conclusion

Summing up this test session, I want to note that Catalyst 8.7, the first official driver for the RV770-based graphics card series, did not allow the new cards to show their full potential. Particularly, the performance of my card is low in 3DMark Vantage and Unigine Tropics Demo, in Devil May Cry 4, S.T.A.L.K.E.R.: Clear Sky and Grand Theft Auto IV. Catalyst 8.10 and 8.11 are faster than version 8.7 and equal between each other. And the last driver of the last year, Catalyst 8.12, indeed increases the card’s performance in such games as Grand Theft Auto IV, Lost Planet: Colonies, Far Cry 2, Unreal Tournament 3 (when FSAA is enabled) and World in Conflict. Unfortunately, like in the previous versions, the new version does not have the ATI Avivo Video Converter working properly (according to monitoring tools, the GPU load was less than 10% during encoding whereas the CPU load was over 80%). That’s not a big problem considering the quality of the video content delivered by that codec, though.

A few words must be said about the HIS Radeon HD 4870 IceQ 4+ Turbo 1GB card I tested the drivers with. On the downside, this card has a faulty BIOS which makes its cooler too noisy. Then, the programmers neglected the option of lowering the card’s frequencies in 2D mode. As a result, the card consumes more power and has higher temperatures than it would have otherwise. The accessories are scanty, too. I guess a modern game or the exclusive screwdriver from HIS would come in handy here. But if you don’t care about free bonuses and think yourself capable of perfecting the card’s BIOS (perhaps using my edited version), the HIS Radeon HD 4870 IceQ 4+ Turbo will be an appealing product with a stylish shining cooler. Overclockability varies from sample to sample, but I want to note the superb overclockability of my sample’s memory and the low overclocking potential of its core.