Radeon HD 4890, GeForce GTX 275

Radeon HD 4890

The Radeon HD 4890 borrows a lot from the Radeon HD 4870. There are however some small differences worth noting. The first concerns the PCB that has been updated with a stronger power stage and designed for an 8 pin connector option instead of 6 pins for one of the two connectors, something that may well be adopted by other partners.


The Radeon HD 4870 PCB is on the left and the Radeon HD 4890 PCB on the right.
Although the GPU has truly been updated, the memory is identical. GDDR5 Qimonda 40x certified at 1 GHz (or 2 GHz for data). But while AMD was rather conservative in clocking it at only 900 MHz on the Radeon HD 4870, it is at 975 MHz on the new card.

The cooling system has also been slightly improved and now has three copper heatpipes instead of two:


Radeon HD 4890 cooler is on the left the and the Radeon HD 4870 on the right.

GeForce GTX 275

The GeForce GTX 275 borrows from the design of the recent GeForce GTX 260+. At the beginning of the year NVIDIA introduced a new revision designed to reduce PCB and cooling system costs, notably in removing the back part of the casing. The cooling system here is similar to that used for the GeForce GTX 285 but with a less complex radiator.

The new card is based on the PCB P897 used for the latest GeForce GTX 260+, but with a GT200 with a configuration similar to that on the GeForce GTX 295, namely with the 240 scalar units activated. It is however clocked higher, 633 MHz for the core and 1404 MHz for the shaders as against 576 MHz and 1242 MHz for the bi-GPU card. The memory bus is the same at 448 bits, but the GDDR3 memory is a little faster than that on the GeForce GTX 260+, going from 1000 to 1133 MHz.

Overclocking

We have tried to push these two cards to their limit although you have to keep in mind that this is just one sample, supplied by AMD and NVIDIA themselves (and therefore perhaps the best available). We used Furmark (executable file renamed so as to stop the Catalyst drivers from lowering the Radeon clock) to validate the clocks.

The Radeon HD 4890 was stable up to 950 MHz for the GPU and 1000 MHz for the memory, which is a 12% overclocking of the GPU. Beyond 950 MHz, small rendering errors are visible and at 1000 MHz the system goes down. We didn’t push the memory over 1000 MHz, which is the maximum validated clock, as it is GDDR5 and data retrieval and signal resynchronisation mechanisms in case of large overclocking can make you think that it has been successful when in fact this isn’t the case. Beyond a certain clock then, the memory loses more time in compensating for errors than it gains in functioning faster, which complicates overclocking tests and nullifies them completely if they aren’t carried out at the same time with performance tests in several situations.

For the GeForce GTX 275 we got up to 730 MHz for the core, 1650 MHz for the shaders and 1300 MHz for GDDR3 memory, ie respective gains of 15%, 18% and 15%.

Specifications, pilots, the test

Specifications

The Radeon HD 4890 has a processing power and memory bandwidth that are respectively 13% and 8% up on the Radeon HD 4870.

The GeForce GTX 275 sees an increase in processing power of 25% on the GeForce GTX 260+ and a bandwidth increase of 13%. NVIDIA’s proposition therefore represents a bigger jump forward than AMDs in terms of sheer figures.

Compared to the GeForce GTX 280, it is interesting to note that the GeForce GTX 275 has 8% more processing power but its bandwidth is 9% lower. It will be interesting to see if one of these is more efficient than the other, keeping in mind that the GTX 275 is limited to 896 MB by its memory bus compared to 1024 MB for the GTX 280.

Software

NVIDIA is taking advantage of this launch to inaugurate the 185 driver series that bring some overall optimisation in terms of performances:

Improved texture management in DirectX 10

Improved Z-culling efficiency

Improved AA compression performance

Improved SLI scaling, especially in Half-Life 2 based engines

We noted some important gains, mainly in situations that bring high demands on memory, which confirms what NVIDIA says.

Moreover, these new drivers officialise the appearance of a new option called “ambiant occlusion” that gives improved graphics by creating a simplified version of global illumination. This technique enables representation of the “ambiant” element of lighting, or environment lighting. Instead of being applied in a homogenous manner on the whole scene it is calculated for each point. Broadly speaking this allows the darkening of some parts of the scene, corners of walls for example, as should be the case as less indirect light hits them from nearby surfaces.

This is a very demanding option and is activated on a case by case basis in some games that can benefit from it and do not integrate it directly themselves, such as Crysis or H.A.W.X. for example. It’s an interesting development for high-end cards that will benefit from improved quality in some games.

AMD has put nothing new on the table in terms or pilots for some time, except unification between Windows Vista and Windows 7. This is probably the consequence of a series of redundancies in the company that with no doubt had an impact on development teams. They must now concentrate on what’s really important, ie. Windows 7 and DirectX 11, leaving less space than usual for the little extras.

The test

Our current test protocol is made up of one OpenGL game (Enemy Territory: Quake Wars), 3 DirectX 9 games (Race Driver : GRID, Fallout 3 and Brothers in Arms Hell’s Highway), 2 DirectX 10 games (World in Conflict and Crysis Warhead) and 3 DirectX 10.1 games (Far Cry 2 and S.T.A.L.K.E.R. Clear Sky).

The tests were carried out at 1920x1200, the resolutions targetted for top end cards, without antialiasing and in 4x and 8x mode (called 8xQ in the NVIDIA pilots). We opted for very high quality. For Crysis Warhead we tested both in enthusiast (ultra) and gamer (high) modes because the first mode has very high energy consumption, even with the highest end solutions.

We also decided to stop showing decimals in game performance results so as to make the graph more readable. We have nevertheless noted these values and used them when calculating the index. If you’re observant you’ll notice that the size of the bars also reflects this.

All available Windows Vista updates were installed as well as SP1. The same goes for the patches of the games tested. We did of course use the most recent pilots available, namely Nvidia’s 185.63 beta and the Catalyst betas supplied by AMD.

Test configuration

Intel Core i7 965 (HT disactived)
Gigabyte GA-EX58-Extreme
6 Go DDR3 1333 Corsair
Windows Vista 64 bits SP1
Forceware 185.63
Catalyst beta 8.592 RC1

Energy consumption, noise, temperature

Power consumption

Measurements were taken at the wall socket: This is therefore the total power consumption of the power supply, in this case a Cooler Master Real Power PRO 1250 watts.


In idle the Radeon HD 4890 consumes the same amount of energy as the Radeon HD 4870. In load it increases by several watts but remains less than the consumption of all the GeForce GTX 200s.

The GeForce GTX 200s are however more economical in idle and the GeForce GTX 275 is no exception. In load it is however much more demanding than the GeForce GTX 285, which is odd as the GeForce GTX 285 is clocked slightly higher and has more memory. These are nevertheless the results that we got. It might be down to a higher voltage for the GT2000 on the GeForce GTX 275s or simply a less efficient power supply.

Noise and temperature

We measured sound levels at rest and sound levels after 30 minutes in load.


The Radeon HD 4890 is noisier than the Radeon HD 4870, only slightly in idle but much more so in load, similar in fact to the GeForce GTX 285.

The GeForce GTX 275 is noisier still, again slightly more so in idle and significantly more in load.

We took temperature readings for these 4 GPUs to set these against noise level readings:


Although the Radeon HD 4890 is noisier than the Radeon HD 4870, it is also better cooled, which shows that AMD has reviewed its approach and gone for a better cooled system at the price of increased noise levels.

We got very slightly higher temperature readings for the GeForce GTX 275 than the GeForce GTX 285. Given that we noted that the first has higher energy consumption, it is logical that to maintain similar temperature levels the fan, which is identical, must spin faster, especially as the radiator is less sophisticated.

RV790 vs RV770

RV790 vs RV770

Out of conscientiousness and although there shouldn’t be any difference, we did compare the performance of the 2 GPUs at, and when their respective memory was at, identical clocks. We therefore clocked the Radeon HD 4890 at 750 MHz (core) and 900 MHz (memory) and caried out all our tests in this configuration at 1920x1200 with 4x FSAA:


As you can see there really is no difference in performance, the Radeon HD 4890 consistently a hairs breadth behind the Radeon HD 4870. ie. Identical performance for these 2 GPUs.

Enemy Territory: Quake Wars

Enemy Territory: Quake Wars

While Quake Wars is based on the Doom 3 engine, it has undergone some evolution such as mega texturing which facilitates the work of artists; however, there is the additional cost in terms of decoding and access to mega textures. In the end, Quake Wars is a little more resource heavy than Doom 3 or Quake 4.

We saved a demo in a sequence versus 8 bots. Given that artificial intelligence was not calculated in the time demo, results were less affected by the CPU than in actual gameplay or at least in this case versus our bot adversaries.

All parameters were set to a maximum in the game including 16x anisotropic filtering. Patch 1.4 was used.


In this first game tested, the GeForces do better than the Radeons although the Radeons do much better as soon as you increase the level of ansiotropic filtering. The Radeon HD 4890 is a little bit more than 10% up on the Radeon HD 4870 and the GeForce GTX 275 climbs to the level of the GeForce GTX 280.

Race Driver GRID

Race Driver GRID

To test Codemaster’s latest opus, we carried out a well-defined movement and pushed all settings to a max. It is based on an evolved form of the motor in Colin McRae DIRT without the unnecessarily complicated aspect and therfore giving a better level of performance We used patch 1.2 and activated 16x anisotropic filtering in the pilot control panel.


In this game, the Radeons are more at ease, even if NVIDIA’s 180 drivers are an advantage. The Radeon HD 4890 shows a gain of 10% which puts it on a par with the GeForce GTX 285.

Brothers in Arms Hell’s Highway

Brothers in Arms Hell’s Highway

The most recent Brothers in Arms is one of many games to use Unreal Engine 3.0. What is quite rare is that no patch has been needed and the game is therefore tested in version 1.0. All settings are at max and 16x anisotropic filtering was activated in the pilot control panel.


Here, the advantage is with the GeForces without antialiasing. Once the filter is activated however the Radeons do much better, the HD 4890 finishing in front of the GeForce GTX 295 with 8x FSAA.

Note that, as with many games based on this graphics engine, antialiasing is not supported directly. With NVIDIA, it works via the driver control panel, but slight bugs are visible. The same goes for AMD who, better late than never, have ended up implementing antialiasing in a similar way.

Fallout 3

Fallout 3

The most recent version of Fallout was much anticipated and uses an updated version of Oblivion. Although this version was supposed to give DirectX 10 support, Fallout 3 makes do with DirectX 9, which doesn’t stop it from being a great success. We use the “ultra high” setting with all options at max in the game. Patch 1.4.0.6 was installed. We had to modify the .ini configuration file manually to disactivate the 60 fps limit.


The very high end solutions are limited by the CPU to 67/69 fps. For the rest, the Radeons have the advantage because of better performances with antialiasing.

World in Conflict

World in Conflict

Very resource heavy and with nice graphics, it's only natural World in Conflict should remain part of our test suite, especially as it has DirectX 10 support.We carry out the internal test with patch 1.0.0.9. We use the “very high” mode which includes DirectX 10, 16x anistropic filtering and activated reflections of the clouds on the water.


In this game, the advantage is clearly with the GeForces. Note that it doesn’t support 8x antialiasing.

Crysis Warhead

Crysis Warhead

Crysis Warhead replaces Crysis and has the same resource heavy graphics engine. We test it in version 1.1 and 64 bit mode as this is the main innovation. Crytek has renamed the different graphics quality modes, probably so as not to dismay gamers who are disappointed at not being able to activate the highest quality mode because of excessive demands on system resources. The high quality mode is renamed “Gamer” and the very high quality mode “Enthusiast”. We went for Gamer mode but also Enthousiast, ultra resource heavy, even for these high-end cards.


In Gamer mode, the Radeons make far less use of the 4 cores of the Core i7, in spite of the use of the most recent pilot which introduces some improvements here. This means the GeForce dominates easily without antialiasing. Once this filter is activated the Radeons are back on course and the Radeon HD 4890 equals the GeForce GTX 275 in 4x mode before outdoing it in 8x mode. Note that the odd results for the GeForce GTX 295 without antialiasing is not an error on our part and reappears systematically.

The GeForces show significant gains in Crysis with antialiasing in high resolution with version 185 of the pilots.


Enthousiast mode being more demanding in terms of the graphics card, it limits the advantage of the GeForce pilots a bit in terms of how it uses the multicore CPUs in this game. While the Radeons have a clear advantage here with antialiasing, the new NVIDIA pilots allow the GeForces to approach a similar level of performance that is nevertheless still too weak to really be able to enjoy the game in these conditions.

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

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

This S.T.A.L.K.E.R. suite relies on an old engine reworked to give DirectX 10 support but still based on the principle of differed rendering. A recent patch gives API DirectX 10.1 support which gives a boost to Radeon performances by fusing two render passes. We use patch 1.5.0.8 and DirectX 10 rendering mode which automatically activates DirectX 10.1 when a compatible card is detected. We used “maximum” quality mode.


While the Radeon HD 4000s were previously dominant here, NVIDIA’s recent pilots have brought it back on equal terms. AMD’s multi-GPU is however more effective in this game. Note that it doesn’t support 8x antialiasing.

Far Cry 2

Far Cry 2

This version of Far Cry isn’t really a great development as Crytek was behind the first episode in any case. As the owner of the licence, Ubisoft took responsibility for its development, with Crytek going with Crysis. No easy thing to inherit the graphics revolution that accompanied Far Cry, but the Ubisoft teams have done pretty well, even if the graphics do not go as far as those in Crysis. The game is also less resource heavy which is no bad thing. It has DirectX 10.1 support and Radeon performances benefit. We installed patch 1.02 and used the ultra high graphics quality mode.


In Far Cry 2 (a swallow doesn’t make a summer) the GeForces have the advantage with antialiasing applied.

H.A.W.X.

H.A.W.X.

The most recent Tom Clancy, H.A.W.X. is an action game that takes place in the air. It uses a graphics motor that supports DirectX 10.1 to optimise results. Among the graphic effects it supports, we note the presence of ambient occlusion that is pushed to a max as with all other options. We used the built-in benchmark and installed patch 1.1.


Although the GeForces do well without antialiasing, activation of this filter causes their performances to freefall. The Radeons, however, benefit from DirectX 10.1 and this allows them to carry out antialiasing more efficiently.

Performance recap

Récapitulatif

Although individual game results are interesting, we calculated a performance index based on all tests with the same weight for each game. An index of 100 was given to the Radeon HD 4870 at 1920x1200.


On average then we have a Radeon HD 4890 that is 10% faster than the Radeon HD 4870, whatever the graphics mode. Without antialiasing it gives identical performances than the GeForce GTX 260+ and betters it with the filter activated.

The positioning of the Radeon HD 4890 opposite the NVIDIA offer is more complex. Although it is 10% up on the GeForce GTX 260+ without antialiasing, it is bettered by the GeForce GTX 275 and 280 on several points. 4x antialiasing allows it to inverse these results and 8x mode gives it a real advantage that even places it in front of the GeForce GTX 285.

The GeForce GTX 275 gives similar results to the GeForce GTX 280. A little better without antialiasing, identical in 4x mode and not as good in 8x mode, which is logical if you look at the spec and main demands of these different graphics modes.

Conclusion

Conclusion

With the Radeon HD 4890, AMD has once again made an impact in the top end segment. Although the card is only a small development of its Radeon HD 4870 designed to overclock more easily, which gives a 10% gain in terms of performance, it has obliged NVIDIA to react so as not to lose out in this segment.

NVIDIA’s reaction has been hurried and unfortunately it has had to sacrifice its GeForce GTX 285 to some extent as the new GeForce GTX 275 now positions itself very close to the GTX 285 in terms of performances. This was necessary to counter the Radeon HD 4890 but the problem is that the 350 euro tag for the GeForce GTX 285 is no longer easily justified, even in terms of its status at the highest-performing mono-GPU card. The GeForce GTX 275 is only 5 or 10% down on it in terms of performance and is on sale at 240 euros (33% cheaper).

This should not however mask the fact that what you get for your money does not change much with these new cards that are proportionately more expensive in terms of the gain in performance on the Radeon HD 4870 and GeForce GTX 260+, whose prices have come down recently. What’s more, noise levels are up, which is a step backwards at this level although with the Radeon this does come with lower temperatures.


Deciding between the Radeon HD 4890 and the GeForce GTX 275 is not easy as the gap between them has been reduced to AMD’s advantage in comparison with that between the GeForce GTX 260+ and the Radeon HD 4870. The GeForce GTX 275 gives slightly better performance levels without antialiasing while the Radeon HD 4890 is ahead when the filter is in 4x mode and almost 20% up in 8x mode. AMD is benefitting increasingly from DirectX 10.1 support to gain in performances. NVIDIA however does well with Mirror’s Edge and Sacred 2, the first good games to really use GPU PhysX and which mean an increase in quality if you have a GeForce. The same goes for the new ambiant occlusion option offered by the most recent drivers.

There is therefore an overall performance gain with antialiasing for AMD and access to superior visual quality for NVIDIA on a handful of games. There will also be an advance in terms of GPU computing with CUDA that is now used by more applications, mainly in terms of video processing, than is its Stream equivalent for the Radeons.

We could then call it a drawn match but this is without taking account of the difference in price that puts the Radeon HD 4890 in the driving seat at around 20 euros cheaper than the GeForce GTX 275. Generally speaking and above and beyond any particular preference you may have for one or other of the solutions, it is AMD that wins the round.