Secret materials from ATi. Review of new graphics cards of the Radeon X800 series. Review of the ATI RADEON VE video card Considerations about what was cut and why
The 4800 series graphics cards are outdated by today's standards. At the time of release, they were quite an interesting solution, but today they are only suitable for outdated computers, the main task of which involves surfing the Internet and working with multimedia programs. The characteristics of the ATI Radeon HD 4800 Series do not allow the cards of this series to provide performance in games, especially modern ones.
HD4850
This is the very first video card in the ATI Series line, the characteristics of which were quite good at the time of the announcement. Core operating frequencies - 625/1986 MHz. The card has GDDR3 memory (outdated today) with a capacity of 512 MB and a 256-bit bus.
It also uses a good RV770PRO GPU with 956 million transistors. This is much more than the previous generation processor - RV770PRO (there were 666 million transistors). So it's safe to say that ATI (now it's AMD) has made quite a big leap forward in terms of increasing performance. The chip itself is made according to 55-nm standards. At that time it was not bad, but now modern cards are made on a 16-nm process technology, which allows you to fit many more transistors in the core.
The graphics support DirectX version 10.1, but not DX11. The characteristics of the ATI Radeon HD 4800 Series are close to those of Nvidia's 9800 GTX. It is this video card that is the rival. When testing both models in the game Crysis, they showed approximately the same result. And at the same time, AMD's HD 4850 is a cheaper solution than its competitor, which is why it wins this argument.
Let's not forget that the HD 4850 is not the only one in this series, there are others.
Overview HD 4870
The performance of the ATI Radeon HD 4800 Series, in particular the HD 4870, is better, because this model itself is an accelerated and more expensive version. It has no architectural or processor changes, but its operating frequency is higher.
The only more or less serious change in favor of HD 4870 graphics is the new GDDR5 memory standard. It allowed, without improving the bus parameters (the bus remained with a bit depth of 256 bits), to increase the bandwidth of the video card, and much more. While the bandwidth of the HD 4850 was 63.6 Gb/s, the HD 4870 increased this to 115.2 Gb/s. The price for this is the higher price of the video card itself and the increased power consumption up to 160 W. This is where the characteristics of the ATI Radeon HD 4800 Series (HD 4870 models) end. But there is another video card in the lineup.
ATI Radeon HD 4800 Series Review: HD 4890 Specifications
After the release of two solutions created using the 55nm process, people started talking about the possible release of new graphics by ATI based on the improved Super RV770 processor. Indeed, very soon the manufacturer announced a new card based on the improved RV790 GPU chip.
The new solution is expectedly called HD 4890, and the updated video chip is already used here. As a result, the card has become faster due to the fact that it operates at higher frequencies than the previous model in the lineup. The main competitor of the video card is the Geforce GTX 275, which was released on the same day.
And although the manufacturer himself says that the main task was to create a new GPU that would become a leader in the upper price segment, the improved RV790 chip is only an improved version of the RV770. There are no other changes to the video card. Even the manufacturing technology and capabilities remain the same, there are no architectural changes either. Most likely, AMD tried to show in this way that they are not far behind a competitor that releases new and advanced solutions.
Specifications ATI Radeon HD 4800 Series (HD 4890):
- Yarn frequency 0 850 MHz.
- Number of transistors - 959 million.
- 256-bit bus.
- 10 SIMD cores.
Developer Partner Versions
After the release of the reference sample, many partner manufacturers created their own solutions based on this chip. However, they all turned out to be approximately the same and were reference products created by order of AMD. That is, the partners simply bought ready-made cards from AMD, pasted their own stickers and put them on the market as products own production. In reality, there is practically no difference between cards from Sapphire, HIS or Powercolor, so the choice here is only a matter of price.
Moreover, all these products are equipped with the same cooler and have a similar cooling system.
Conclusion
With the release of this card, AMD managed to strengthen the product line of the 4xxx series and almost come close to the flagship from Nvidia (we are talking about the Geforce GTX 285 card). However, Nvidia also did not lag behind and created an improved version of the Geforce GTX 285 card, which had a 448-bit bus. This reduced the memory from 1 GB to 896 MB, but increased throughput. The new graphics from Nvidia got a little better, but this product was a limited edition, and the graphics card from AMD was shipped to the market in large quantities.
Note that at the time of release, the characteristics of the ATI Radeon HD 4800 Series memory were quite serious. But now these cards are no longer sold, and they can only be used in older computers. By modern standards, it is even outdated graphics that are not able to satisfy the needs of users, not to mention gamers. Nevertheless, at one time the card was popular and even today is actively used in computers for everyday use and surfing the Internet.
Recently, taking into account the trend in the development of the graphics accelerator market, we are all accustomed to the rapid change in generations of video adapters. Although for quite a long time the leading role in AMD was occupied by the ATI HD3870 video card (later ATI HD3870X2), based on the RV670 chip. As soon as the first rumors about the new RV770 chip began to leak out, media interest shifted to the future “master of the throne”.
The appearance of the new chip marked the debut of new AMD Solutions ATI (on RV770 PRO chip) and AMD ATI HD4870 (on RV770 XT chip).
Before the release of graphic solutions based on the RV770 chip, the company's market position was not the best. In the HD card family, there was not a single worthy rival for the top solutions of the age-old California competitor, NVIDIA. The release of a new chip was more of a vital necessity than just the release of a new accelerated solution. The engineers did their best - the chip turned out to be very successful and promising.
In the new chip, it was decided to change traditions and switch to an architecture with a central hub instead of the already familiar ring bus.
According to ATI press releases, this arrangement greatly improves bandwidth efficiency. In addition, the memory controller now supports the new GDDR5 memory chips.
New GPU already contains 800 scalar processors capable of performing 32-bit and 64-bit calculations.
But the architecture of stream processors has not changed much (compared to RV670), although their density has been increased, which made it possible to increase their number without changing the manufacturing process. Now the theoretical peak performance of the RV770 chip has increased to 240 gigaflops.
Technical details of HD4800 series accelerators:
- Chip codename RV770;
- 55 nm technology;
- 956 million transistors;
- Unified architecture with an array of common processors for streaming vertex and pixel processing, as well as other types of data;
- Hardware support for DirectX 10.1, including the new shader model - Shader Model 4.1, geometry generation and intermediate data recording from shaders (stream output);
- 256-bit memory bus: four 64-bit wide controllers with GDDR3/GDDR5 support;
- Core clock 625-750 MHz;
- 10 SIMD cores, including 800 scalar floating point ALUs (integer and floating point formats, support for FP32 and FP64 precision within the IEEE 754 standard);
- 10 enlarged texture units, with support for FP16 and FP32 formats;
- 40 texture address units;
- 160 texture fetch blocks;
- 40 bilinear filtering units with the ability to filter FP16 textures at full speed and support for trilinear and anisotropic filtering for all texture formats;
- Possibility of dynamic branching in pixel and vertex shaders;
- 16 ROPs with support for anti-aliasing modes and the possibility of programmable sampling of more than 16 samples per pixel, including with FP16 or FP32 frame buffer format - peak performance up to 16 samples per clock (including for MSAA 2x / 4x and FP16 format buffers), in colorless mode (Z only) - 64 samples per clock;
- Write results to 8 frame buffers simultaneously (MRT);
- Integrated support for two RAMDAC, two ports Dual Link DVI, HDMI, HDTV, DisplayPort.
Reference Card Specifications:
- Core clock 625 MHz;
- Number of universal processors 800;
- Number of texture blocks - 40, blending blocks - 16;
- Effective memory frequency 2000 MHz (2*1000 MHz);
- Memory type GDDR3;
- Memory capacity 512 MB;
- Memory bandwidth 64 GB/s;
- Theoretical maximum fill rate is 10.0 gigapixels per second;
- Theoretical texture sampling rate of 25.0 gigatexels per second;
- Two CrossFireX connectors;
- PCI Express 2.0 x16 bus;
- Two DVI-I Dual Link connectors, output in resolutions up to 2560x1600 is supported;
- TV-Out, HDTV-Out, support for HDCP, HDMI, DisplayPort;
- Power consumption up to 110 W (one 6-pin connector);
- Single slot cooling system design;
- Suggested price $199.
One of them will be discussed in today's review, namely AMD ATI with 512 MB of memory on board.
|
GeForce 9800 GTX |
GeForce 9800 GTX+ |
|||
| Graphics chip |
RV770PRO |
|||
| Core frequency, MHz | ||||
| Frequency of unified processors, MHz | ||||
| Number of universal processors | ||||
| Number of texture/blend units | ||||
| Memory size, Mb | ||||
| Effective memory frequency, MHz |
2000 (2*1000) |
|||
| Memory type | ||||
| Memory bus width, bit |
The video card is based on AMD's ATI graphics processor based on the RV770 PRO chip using 55nm technology. At the same time, all the recommendations of the GPU manufacturer mentioned above are observed, so the accelerator repeats the capabilities and appearance the vast majority of 512 MB, except, perhaps, a complete set.
Let's move on to a closer acquaintance with the tested video card EAH4850/HTDI/512M.
The video card comes in a large double cardboard box that opens up like a book. Unlike previous packages of top models, this box does not have a plastic handle.
The appearance and design of the box has not changed. As before, black and orange colors symbolize that the adapter belongs to the AMD ATI family. At the bottom of the box, there is usually the name of the accelerator, as well as some of its features. This time the main focus is on the DVI to HDMI adapter, which the buyer gets "free of charge".
On the back of the package, the features of the graphics accelerator, recommended system requirements are described, as well as a brief presentation of proprietary technologies, which can be found in more detail on the official ASUSTeK Computer website.
The delivery set is sufficient for the full use of the video adapter. In addition to the video card itself, it includes:
- Adapter from Molex to 6-pin video card power connector;
- Adapter from S-Video to component output;
- Adapter from DVI to D-Sub;
- DVI to HDMI adapter;
- Bridge CrossFire;
- CD with drivers;
- CD with electronic documentation;
- Brief instructions for installing a video card.
Externally, the test sample is very similar to the AMD ATI HD 3850. The video card itself is made according to the reference design on red textolite and is equipped with a single-slot cooling system that covers most of it. The only external difference of our video card is that the plastic shroud does not completely cover the PCB. The dimensions of the adapter are compact, the length is 233 mm, which will allow it to fit into almost any case.
On the back side there are stickers with the exact name of the graphics accelerator, serial number and batch number.
All connectors are protected by plastic caps, which is not always seen on video adapters from. The interface panel has two DVI outputs and a TV-out. To connect an analog monitor, you will need to use the adapter supplied.
Now let's consider the cooling system of the tested video card. As we have already described above, it occupies one slot and is a massive plate. In the middle is a copper heat sink that is adjacent to the GPU.
Memory chips and power elements are in contact with the wafer substrate through thermal pads.
Under the plastic casing of the cooling system, there is a radiator consisting of thin copper fins interconnected. The air flows from the cooler pass through these fins to the rear wall of the case, so for the normal output of warm air, it is necessary to remove the plug in the rear panel next to the video card.
The printed circuit board is not saturated with a large number of elements, but there is an innovation - the memory chips are located in two lines above and to the right of the graphics chip, and a pair of central chips of each of the lines are grouped.
The power part of the board does not surprise with the complexity of execution. In the upper corner is a 6-pin video card power connector, which is not surprising given the declared power consumption of up to 110 watts. According to the specification for the video accelerator, a power supply unit with a power of 450 W or more is required.
The memory consists of eight GDDR3 chips manufactured by Qimonda (HYB18H512321BF-10) with an access time of 1.0 ns, which allows them to operate at frequencies up to 2000 MHz. The effective memory frequency of the tested video card model is slightly lower and amounts to 1986 MHz, which leaves a narrow frequency corridor in reserve. The total amount of memory is 512 MB, and the width of the exchange bus with it has not changed and is 256 bits.
The frequency of the GPU corresponds to the recommended value of 625 MHz. As already described above, the RV770 chip itself is made according to the 55 nm process technology, which causes its relatively low power consumption, despite the fact that it includes 956 million transistors. The number of unified shader processors has been increased to 800, texture units to 40, and the number of ROPs has remained unchanged at 16. The chip's operating frequency in 2D mode is reduced to 500 MHz.
To evaluate the efficiency of the standard cooling system, we used the FurMark utility, and monitoring was carried out using GPU-Z version 0.2.6. Running at stock frequencies, the GPU warmed up to 92°C, which is not too low, especially considering the appearance of some noise from the cooler.
Testing
The test results show that it is a direct competitor for the GeForce 9800GTX and comes close to the performance of more expensive GeForce GTX260-based accelerators. The exception is gaming applications optimized for the NVIDIA architecture.
The video card was overclocked using standard ATI Catalyst Control Center tools. The video card was able to operate stably at 670 MHz for the graphics core (+45 MHz) and 2180 MHz (1090 MHz DDR) for video memory (+186 MHz).
A rather modest result, we initially expected more, but let's see how much the adapter's performance will increase.
|
Test package |
Standard Frequencies |
Overclocked graphics card |
Productivity increase, % |
|
| Futuremark 3DMark'05 | ||||
| 3D Mark Score | ||||
| SM2.0 Score | ||||
| HDR/SM3.0 Score | ||||
|
Serious Sam 2, Maximum Quality, NO AA/AF, fps |
1024×768 | |||
| 1280×1024 | ||||
| 1600×1200 | ||||
|
Serious Sam 2, Maximum Quality, AA4x/AF16x, fps |
1024×768 | |||
| 1280×1024 | ||||
| 1600×1200 | ||||
|
Call Of Juarez, Maximum Quality, NO AA/AF, fps |
1024×768 | |||
| 1280×1024 | ||||
| 1600×1200 | ||||
IntroductionAs you remember, last summer ATI released two new graphics chips - RADEON 8500 and RADEON 7500.
The RADEON 8500 contains the latest developments from ATI, and I won't be mistaken if I say that lately this graphic chip has become the most rich in new features and capabilities, especially against the backdrop of the emergence of NVIDIA's "kind of new" Titanium series chips. The RADEON 8500 turned out to be very fast, interesting, not without flaws, of course, but it adequately accepted the challenge thrown by the top models from the "sworn friend".
The younger brother from the new family - the RADEON 7500 - which is in the shadow of the RADEON 8500, did not attract such attention, was not treated kindly or scolded on the Web pages with the same force, although it received as much attention as New Product deserves nothing less.
ATI RADEON 7500 is manufactured using 0.15µm technology, its 3D part is completely inherited from ATI RADEON, and its 2D part - from RADEON VE.
Thus, the RADEON 7500 contains well-adjusted architecture details that have passed the test of time. And rewiring the chip and transferring it to a thinner process technology made it possible to achieve operation at very high frequencies - up to 300 MHz and higher. As a result, a new chip appeared, which has the ATI RADEON architecture, but is almost twice as fast, and at the same time has high-quality support for dual-monitor configurations. The purpose of the release of ATI RADEON 7500 is to oust its "sworn friend" in the sector of mid-range video cards - NVIDIA GeForce2 Pro, GeForce2 Ti and, to some extent, GeForce3 Ti200.
Now boards based on the ATI RADEON 7500 chip are already sold everywhere, and this is one more ATI merit: finally, the production of video cards was given to third-party companies, which immediately allowed to reduce the price and increase the number of video cards produced.
Our goal is to figure out what the RADEON 7500 is and try to objectively compare boards based on it with similarly priced video cards based on NVIDIA chips.
Shall we start?
Characteristics and features of ATI RADEON 7500
The main characteristics of the chip and 3D capabilities:
Operating frequency - 270-290 MHz
Video memory interface - 64/128 bit SDRAM or DDR SDRAM
Number of pixel pipelines - 2
Number of texture modules - 3 in each pipeline
Overlay up to three textures in one pass
Bilinear, trilinear and anisotropic texture filtering
Relief simulation using Emboss, Dot3, EMBM methods
S3TC/DXTC texture compression support
Full screen anti-aliasing 2x, 4x supersampling
Hardware block T&L
HyperZ Support
Capabilities in terms of 2D and video playback:
Two built-in CRT controllers
Two built-in 350 MHz RAMDACs
Built-in TMDS transmitter for output to digital monitors
Built-in TV encoder for displaying images on TV.
Support for adaptive deinterlacing
Support for DVD hardware decoding - iDCT
So, it follows from the above that ATI RADEON 7500 supports dual-monitor configurations. The following combinations of connection to the RADEON 7500 are possible:
Analog monitor + analog monitor (with DVI-I to VGA adapter)
Analog Monitor + Digital Monitor
Analog Monitor + TV
Digital monitor + TV
Remarkably, any display device can be "primary" or "secondary" on the RADEON 7500, since the RADEON 7500, like the RADEON VE, has equal rights for both CRT controllers.
In the ATI RADEON VE review, work in dual-monitor configurations is considered in sufficient detail, so there is no point in repeating it now.
ATI RADEON 7500 board
The ATI RADEON 7500 board is equipped with VGA, DVI and S-Video outputs, but does not impress with an abundance of chips - everything you need is integrated into the RADEON 7500 core: 
The "heart" of the board is the ATI RADEON 7500 chip, made using 0.15 micron technology:
The board is equipped with 64 MB DDR SDRAM by Hynix with a cycle time of 4 ns.:
The clock frequencies of the core and video memory on the board that took part in the tests - ATI RADEON 7500 OEM, are by default 270/460 (230 DDR) MHz.
Tellingly, the situation with the core frequencies of the RADEON 7500 and RADEON 8500 is similar: only ATI RADEON 7500 retail boards have a core clock speed of 290 MHz, while all other video cards based on ATI RADEON 7500, including the RADEON 7500 in the OEM version from ATI itself, have core frequency 270 MHz. The frequency of the video memory on all boards based on ATI RADEON 7500 is, fortunately, the same and amounts to 230 (460 DDR) MHz.
During tests, we set the operating frequencies of the ATI RADEON 7500 board to 290/230 MHz, like Retail ATI RADEON 7500
test system
The following system was used to test the boards:
Processor - AMD Athlon XP 1500+;
Motherboard– MSI K7T266 Pro2 v2.0 (VIA KT266A);
Memory - 2*128 MB DDR SDRAM PC 2100 Nanya CL2;
HDD– Fujitsu MPF3153AH.
Software:
Driver version 6.13.10.6011 for Windows XP for ATI RADEON 7500;
Detonator 23.11 driver for Windows XP for boards based on NVIDIA chips;
Max Payne;
Serious Sam v1.05;
3D Mark 2001;
Quake3 Arena v1.27;
Windows XP.
The following boards were tested together with the ATI RADEON 7500 to compare performance:
SUMA Platinum GeForce2 Pro (NVIDIA GeForce2 Pro, 200/400 MHz, 64 MB DDR SDRAM)
VisionTek Xtasy 5864 (NVIDIA GeForce2 Ti, 250/460 MHz, 64 MB DDR SDRAM)
VisionTek Xtasy 6564 (NVIDIA GeForce3 Ti200, 175/400 MHz, 64 MB DDR SDRAM)
Speed in 3D
3D Mark 2001First, synthetic tests 3DMark 2001 for the speed of filling and processing polygons:
ATI RADEON 7500, due to optimization of work with Z-buffer and the ability to overlay three textures in one pass, has the lowest losses in scene filling rate when switching from 16-bit screen modes, Z-buffer and texture quality to 32-bit ones. In addition, due to the higher video memory frequency, the maximum video memory bandwidth of the ATI RADEON 7500 is higher than that of its competitors in this test.
But losses are losses, and the performance of the ATI RADEON 7500 turns out to be lower than that of a motherboard based on NVIDIA GeForce3 Ti200, despite the lower clock frequency of the latter's video memory. Both the twice as many pixel pipelines of the GeForce3 Ti200, which ensured a higher theoretical fill rate, and the GeForce3's LightSpeed Memory Architecture, which made it possible to efficiently use the video memory bandwidth, played a role here. As a result, NVIDIA GeForce3 Ti200 and ATI RADEON 7500 are leaders in this test.
When using the T&L hardware block, the ATI RADEON 7500 turns out to be the leader, however, this does not mean that the transformation and lighting block on the RADEON 7500 is much more powerful than on the GeForce2 Ti or GeForce2 Pro. It is worth remembering that ATI RADEON 7500 has the highest core frequency among all test participants, and if you make simple calculations, imagining what the result of ATI RADEON 7500 will be at a core frequency of 175-200 MHz, it will become clear that the performance of the T&L RADEON 7500 approximately equal to the performance of GeForce2 Pro / Ti - at the same frequencies it will be slightly slower in the test with one light source, and slightly faster in the test with 8 lights.
In the case of software calculation of scene geometry, ATI RADEON 7500 turns out to be an obvious outsider, and only the quality of driver optimization can be blamed for this.
To be honest, we could also consider other tests, show other features of the ATI RADEON 7500 architecture, for example, the operation of three texture modules or HyperZ, but, firstly, this is not so interesting, and secondly, there is nothing new in this regard. you shouldn't expect anything from the RADEON 7500 - the 3D part of the ATI RADEON 7500 has no innovations compared to the good old RADEON.
Therefore, we round off with synthetic tests and move on to gaming ones.
Testing in 3DMark 2001 was carried out only in the Dragothic and Lobby tests - the first of the remaining ones, Car Chase, was found to have too wide a spread of results and a strong dependence on the processor speed, and the second one, Nature, you understand, will work only on the GeForce3 Ti200.
Not bad result for ATI RADEON 7500. Due to the fact that three textures are applied in this test, and the Overdraw indicator is large enough, which gives room for HyperZ, the RADEON 7500 turns out to be slightly slower than the GeForce2 Pro/Ti in the "Low Details" mode, and in "High Details" it is completely confident stays in second place. Of course, the RADEON 7500 was unable to outperform the GeForce3 Ti200, which has a more modern architecture.
ATI RADEON 7500, having only 2 pixel pipelines and a lower scene fill rate, turns out to be slower than GeForce2 Pro/Ti in 16-bit modes. But at 32 bits, thanks to the action of HyperZ and large caches, the uncertain position of the RADEON 7500 strengthens and, as the resolution increases, it turns into a solid lead, with a 20% speed margin, from the GeForce2 Pro/Ti.
The leader is still NVIDIA GeForce3 Ti200.
Max Payne
For testing in Max Payne was used benchmark mod and test scene PCGH's Final Scene No1, the description of which is on the German site 3DCenter.
Testing was carried out in two versions:
1st option - " quality"- the maximum graphics quality settings are set, the color depth of textures and frame buffer is 32 bits, but texture filtering is not anisotropic, but trilinear, and full-screen anti-aliasing is disabled (after all, we are not testing the GeForce3 Ti500 and RADEON 8500 :)... )
2nd option - " speed" - minimum settings for graphics quality, color depth of textures and frame buffer - 16 bits.
With such tests, I hope, both "speed at all costs" lovers and those for whom image quality is more important than the number of frames per second will be satisfied:
As you can see, the difference in speed between the "quality" and "speed" modes is large, however, the results shown by video cards are very close. The results of ATI RADEON 7500, NVIDIA GeForce2 Pro and GeForce2 Ti are approximately on the same level, the leader is again GeForce3 Ti200.
There were no complaints about the quality of work in Max Payne against the ATI RADEON 7500, only in the 1600x1200 mode it refused to work, displaying an error message:
Quake3 Arena
Testing in Quake3 Arena was carried out under traditional conditions: maximum quality settings, trilinear filtering enabled, texture compression disabled:
In 16-bit modes, as expected, the RADEON 7500 is an outsider, however, in 32-bit modes its performance compares to GeForce2 Pro/Ti due to a more balanced architecture, and at high resolutions, thanks to HyperZ, it even turns out to be higher. The result of NVIDIA GeForce3 Ti200 again turned out to be out of reach for the other participants of this review.
Serious Sam
We decided to test in Serious Sam in the same way as Max Payne in two modes:
1st option - graphics quality settings" quality", frame buffer depth - 32 bits;
2nd option - graphics quality settings" speed", frame buffer depth is 16 bits.
For testing, the standard demo record DemoSP03 was used:
So the result is very interesting. Having lost to everyone in the "speed" mode, the RADEON 7500 performed well in the "quality" mode, even outperforming the previously unattainable GeForce3 Ti200 in the 1600x1200 mode!
In the "quality" mode, the Serious Sam engine turns on anisotropic filtering, and this is precisely the reason for the RADEON 7500's success. a lot of performance.
By the way, you can quote fragments of the Serious Sam configuration files, which indicate what level of anisotropic filtering is used on different video cards in the "Quality" mode:
NVIDIA GeForce256 / GeForce2 / GeForce3:
if(sam_iVideoSetup==2) (
gap_iTextureAnisotropy = 4;
gap_fTextureLODBias = -0.5;
}
ATI RADEON, RADEON 7xxx, RADEON 8xxx family:
if(sam_iVideoSetup==2) (
gap_iTextureAnisotropy = 16;
gap_fTextureLODBias = -0.5;
}
As you can see, for the RADEON 7500 the level of anisotropy set by the developers of Serious Sam turned out to be even higher, and the RADEON 7500 is still in the lead.
How the RADEON series chips manage to perform anisotropic filtering so painlessly, I will try to explain in the "3D Quality" section, and now, regarding Serious Sam, about a new feature of its engine.
Serious Sam version 1.05 introduced the ability to use Direct3D, and of course, I did not fail to use it. The performance results for boards based on NVIDIA chips turned out to be close to the results in OpenGL, and, without expecting any tricks, I was about to compare them with the performance of the ATI RADEON 7500... However, when Serious Sam was running via Direct3D on the ATI RADEON 7500, I saw a terrible picture:
Of course, there can be no question of any comparison of performance in Serious Sam with such work of the ATI RADEON 7500.
The question remains: who is to blame for this - the Direct3D driver from ATI or the developers from Croteam, who tested Direct3D only on NVIDIA boards? :)
Quality in 3D
The most interesting feature of the ATI RADEON 7500/8500 is the fast implementation of anisotropic filtering.Let me remind you that anisotropic filtering is the most correct texture filtering method that allows you to get the highest quality image. When using anisotropic filtering, to obtain the color of a pixel, it is not the color of the texture at the point on the surface of the object corresponding to this pixel, nor the interpolated color value of the four adjacent texels surrounding the projection of the pixel, as in the case of bilinear filtering. With anisotropic filtering, a pixel is considered as a small circle or rectangle with an ellipse or quadrilateral projection onto the texture, and the colors of all texels falling into this projection are taken into account to obtain the pixel color.
Accordingly, when the angle between the line of sight and the observed surface decreases, the ellipse - the projection of the pixel - will stretch, which will lead to the need to average the colors of more and more texels. With this construction method, the computational costs turn out to be very high, but the quality of the resulting image will also be high, not without reason, for example, all modern 3D modeling packages use anisotropic filtering for the final construction of scenes.
Of course, video accelerators use simplified methods of anisotropic filtering. For example, to obtain the final color of a pixel, NVIDIA GeForce3 apparently evenly "arranges" on the long axis of the ellipse - the projection of the pixel - several points (1,2,4,6,8, their number depends on the elongation of the ellipse or the level of anisotropy), in which it performs bilinear filtering, and then averages the resulting colors, possibly with different weight coefficients.
Of course, these are all guesses, but they are perfectly consistent with practice. And practice has shown that processing each such point of GeForce3 requires an additional cycle, for example, anisotropic filtering over 32 samples (8 points, 8 bilinear filtering operations, 8x4=32) turns out to be exactly 8 times slower than bilinear.
The ATI RADEON family does not implement anisotropic filtering, apparently.
I'll start from afar :).
It is known that in order to avoid "dance" and "graininess" of textures on distant objects, MIP-Mapping is used, that is, the replacement of the original texture with its less detailed versions as the object moves away from the viewer. In the figure, the original texture is shown at the top left, and its MIP levels go diagonally right-down from the original texture:
The dimensions of the texture at each of the MIP levels are 2 times smaller than its sizes at the previous one, and the color of each of the texels is the average of the colors of the four corresponding texels of the previous MIP level.
However, this is not what is interesting, but two more rows in which the texture is filtered and compressed along only one of the two axes. In the figure, these rows are shown going to the right and down from the image of the original texture.
Let's call it, by analogy with MIP levels, "RIP levels". Why are they remarkable? The fact that the color of each of the texels of any "RIP-level" from this series is the average value of the colors of the two texels of the previous "RIP-level". Why is all this necessary? And let's imagine this situation: we look at a plane with our texture at an acute angle, something like this:
The projection of one of the pixels onto the texture is shown as a red ellipse. In theory, in order to correctly perform anisotropic filtering, we need to average the colors of all texels that fall into the ellipse - they are circled in green.
However, it is worth remembering that we have prepared a number of "RIP-levels", and from them you can choose the one in which the compression ratio is as close as possible to the degree of anisotropy, that is, the degree of "elongation" of the ellipse, make bilinear filtering on it and get as a result, a color that is an average of the colors of the texels of the original texture we need. I hope I was able to show this clearly enough in the figure.
As a result, having a number of pre-prepared variants of the original texture - "RIP-levels", we can perform filtering with any reasonable level of anisotropy, in fact using only bilinear filtering, that is, with virtually no performance loss.
From the nature of this method, called RIP-mapping, it follows that its results will be most correct in the case when the angle of the ellipse is close to one of the texture axes, and at "uncomfortable" angles close to the diagonals, RIP-mapping cannot ensure the quality is higher than bilinear filtering. Accordingly, in order to avoid loss of texture filtering quality at "uncomfortable" angles, you can use some combined RIP levels, compressed along two axes at a different number of times (sorry for the tautology :)), introduce a number of "diagonal" RIP levels , or perform anisotropic filtering by other methods, such as NVIDIA GeForce3.
To all appearances, ATI's RADEON family uses RIP-mapping. When using this method, the switching lines of MIP levels, or RIP levels, are broken.
It's quite easy to check this: by enabling anisotropic filtering in a small test application from NVIDIA that uses standard OpenGL extensions and works on any video cards, I took screenshots in which these lines are very noticeable. Left - picture on ATI RADEON 7500, middle - on NVIDIA GeForce2 Ti, right - on NVIDIA GeForce3 Ti200:
On the ATI RADEON 7500, the switching lines of MIP levels have kinks, intersections, and generally behave in a completely wild way (and how else can you do trilinear filtering?), unlike NVIDIA GeForce2 and GeForce3 MIP levels, where there are no anomalies.
By the way, users sometimes notice artifacts associated with anisotropic filtering on ATI video cards, and I could show a few typical fragments from games, but, firstly, there are actually not so many such remarks, and secondly, the artifacts are stronger everything is noticeable in dynamics, which you can’t show in the screenshots ...
Therefore, the story about the unpleasant aspects of anisotropic filtering should be finished, and the review should be continued with its pleasant aspects. Firstly, it's speed: when anisotropic filtering is enabled, the ATI RADEON family loses a few percent in performance, and secondly, quality: under favorable conditions, anisotropic filtering on RADEON chips is superior in quality to its implementation in NVIDIA chips.
As an example, here are screenshots from Serious Sam, where the quality of anisotropic filtering was set to the maximum for each of the boards. Like last time, on the left - pictures on ATI RADEON 7500, in the middle - on NVIDIA GeForce2 Ti, on the right - on NVIDIA GeForce3 Ti200:
Summarizing the part about the anisotropic filtering of ATI RADEON 7500, I can only say one thing: NVIDIA GeForce2 / GeForce3 and ATI RADEON 7500 have completely different implementation algorithms with their pluses and minuses, and the right to choose what we like is in our hands .
Let's put together the pros and cons:
Anisotropic filtering ATI RADEON 7500/8500:
Plus - high quality;
Plus - high speed;
Minus - it is impossible to work simultaneously with trilinear filtration;
Minus - the occurrence of artifacts in certain situations.
NVIDIA GeForce3 Anisotropic Filtering:
Plus - high quality;
Minus - big losses of productivity.
Overclocking
The ATI RADEON 7500 was overclocked using the PowerStrip 3.12 utility.An interesting situation arose during overclocking: increasing the core frequency, as expected, led to an increase in performance, while increasing the frequency of the video memory did not lead to anything. Interestingly, it was possible to set any frequency of the video memory, even 800 MHz, the board did not react to it at all.
After searching and reviewing messages from ATI RADEON 7500 owners in various conferences, I had no choice but to agree with them - it looks like the RADEON 7500 chip or ATI drivers have a video memory overclocking block.
Therefore, overclocking was carried out only on the core. The maximum core frequency at which the board worked stably was 340 MHz. The performance gain with this overclocking is shown in the graph:
Agree, with a 17% increase in the core frequency (290->340 MHz), a 15% increase in Quake 3 and an 8% increase in Serious Sam is quite good. However, this was to be expected: the architecture of the ATI RADEON 7500, like that of the "old" RADEON, is well balanced, and the board's performance is not always strictly limited by the video memory bandwidth.
Conclusion
ATI RADEON 7500 is a very interesting video card that provides excellent image quality, full support for dual-monitor configurations, output to TV and digital monitors. Together with all this, its performance in 3D applications is also at a good level.If we compare ATI RADEON 7500 with video cards based on NVIDIA GeForce2 Pro / GeForce2 Ti, then in terms of 2D it turns out to be definitely better, both in terms of quality and functionality. In 3D games, the performance of the ATI RADEON 7500 is on average at the same level as the GeForce2 Pro / GeForce2 Ti.
The prices of these boards are slightly lower than those based on ATI RADEON 7500, so if you choose a video card based on ATI RADEON 7500 instead of GeForce2 Pro / Ti, consider yourself paying extra for it. quality work in 2D.
A comparison between ATI RADEON 7500 and NVIDIA GeForce3 Ti200 shows that the latter is faster in almost all 3D games. Lacking full support for DirectX8, the RADEON 7500 is even more unable to compete with the GeForce3 Ti200.
On the other hand, boards based on the GeForce3 Ti200 cannot compete with the ATI RADEON 7500 in terms of 2D functionality. The quality of the output on the screen for video cards based on NVIDIA chips may also turn out to be unimportant - there are different manufacturers. And boards based on ATI RADEON 7500/8500, interestingly, are excellent by any manufacturer. Strict quality control by ATI?
In general, if you need a purely gaming video card, then you can choose something based on the GeForce3 Ti200, the price will be higher than that of the ATI RADEON 7500, the speed will be higher, but what the quality of the board will be is a big question.
Advantages of ATI RADEON 7500:
Excellent quality of installation;
Full support for dual-monitor configurations;
Availability of DVI and high-quality TV-out;
Excellent image quality on the monitor;
Good speed in 3D.
Minuses:
Lack of support for DirectX8 pixel and vertex shaders;
The scarcity of the delivery set.
For more than a year, nVidia has remained consistently the leading company in the computer graphics market, and leadership has been so strong and stable that no one thought of a possible change in leadership.
Remembering the old days, we came to the conclusion that the most competitive model of recent times was the Radeon HD 2900XT, which, nevertheless, could not surpass the GeForce 8800 GTX. But last year, the Radeon HD 3870 also appeared, which, unfortunately, again hardly has a serious chance of getting "gold" compared to its competitors, although it did its job quite well - raising the price target for most of these products.
The year 2008 has begun, and Nvidia still hasn't presented the public with a true next-generation product that could really move forward compared to all of the company's previous products. At the same time, we hoped that AMD/ATI would keep their promises to develop a dual graphics processing unit (dual-GPU) based on the Radeon HD 3800, which, unlike traditional multi-graphic processors (multi-GPUs), would not require platforms for two compatible graphics cards (crossfire technology) and which can become a solid competitor in the high-quality graphics market.
Developers Radeon HD 3870 X2 put two graphics processing units (GPU) on one board
To be honest, we were rather skeptical about the Radeon HD 3870 X2 when we first got acquainted with the then purely theoretical idea of its creation. The introduction of a single card for Crossfire technology - originally a technology for cross-working two video cards - could lead to huge delays in production and to an overall decrease in the attractiveness of devices for consumers. In fact, as practice has shown, AMD managed to confirm all statements about the functionality of its new product, because. the developers quite successfully launched the production of the Radeon HD 3870 X2 and, more importantly, were able to immediately provide it for retail sale.
As you may have guessed, the ATI Radeon HD 3870 X2 is essentially two Radeon HD 3870 GPUs on the same printed circuit board (PCB). The processors use the PCI Express 1.1 bridge to work together, which is very reminiscent of the cross-work of ATI video cards. However, while the X2 is clearly designed as a single card device, you can still use two graphics cards in "crossover" mode. This means that with a properly selected driver, the operation of a two-bit system in cross-functional mode has become a reality.
Long before we actually tested the X2, we thought that dual Radeon 3870 GPUs would be quite efficient, but we also expected that, as with standard dual SLI cards and crossfire technology, the X2 would experience hesitation during gameplay. So it, in principle, turned out to be - see the description of our benchmarks for more details.
A huge advantage of the Radeon design is that it allows the user to be independent of drivers and chipsets (because both GPUs are on the same board). It is worth noting that modern drivers (Catalyst 8.1 and 8.45) seem a little "immature" to us. While overall the system worked stably during testing, there were occasional glitches that weren't typical when playing a particular computer game. Is this a consequence Windows usage Vista Ultimate 64-bit or not is not yet clear.
Pay
Now, more than ever before, it is typical for many manufacturers in the world to create video cards that fully follow the design patterns of ATI and nVidia; and the Radeon HD 3870 X2 is no exception, as most of the cards it uses are identical to those of ATI. And yet, during the short period of time that this product was available, we were able to find completely different cards made by ASUS. Their differences were in an improved cooling system, the presence of four DVI ports and a slightly modified frequency of the memory and microprocessor core. In all other respects, the cards are exactly the same.
In terms of dimensions, the ATI Radeon HD 3870 X2 is huge: 26.5 cm long, which is much larger than the size of an entire ATX motherboard. Nevertheless, the dimensions of the X2 are very close to those of the GeForce 8800 GTX/Ultra, so these numbers won't surprise anyone. As for the design itself, the ASUS version has several obvious differences, and we believe that this is an absolute plus for this product.
Instead of using a standard plastic shell to protect against overheating, ASUS created a black aluminum case for this purpose, on which there are two ventilation holes (each 65 cm in diameter) to cool the two integrated GPUs on the board. If this case is removed, then two rows of small aluminum mesh plates become visible, and in the second row of plates there are much more. Each of these plates is connected to the copper board via two heat pipes. In fact, such a cooling system is more radical than its standard options, and, as it seemed to us, it sometimes simply lacks power.
In addition to the new approach to the cooling system, you will also notice a large steel plate with the inscription "DUAL" located on the front of the video card. On closer examination of the back plate, you will also find that there are not two DVI ports, but four, which is typical for devices using crossfire technology. Those who are very interested in the possibility of connecting LCD monitors will be intrigued by this discovery, since when using two video cards, the number of DVI ports will increase to 8!
The memory capacity of the ATI Radeon HD 3870 X2 is 1 GB (512 MB for each GPU). The default specification for this amount of memory is 1800 MHz, although ASUS has increased the frequency to 1910 MHz on the most advanced versions of their devices. Moreover, each processor is supposedly designed to run at 825 MHz, while ASUS made it possible to run its top versions at 850 MHz. This is a 3% increase in the frequency level for each GPU, but we can imagine how much the attitude of manufacturers towards the standard board design has changed.
Ideally, ASUS devices can use Hynix's 1GB GDDR3 .8ns memory "HY5RS123235B FP-08", which makes the board work quite fast. We were able to reach 2122 MHz before the system started to crash. For microprocessors, the limitation on the above parameter turned out to be 861 MHz, which is only 11 MHz more than before.
Benchmarks: specifications and test for Crysis
Specifications: hardware (hardware)
- Intel Core 2 Duo E8400 (3.00GHz) LGA775
- x2 OCZ DDR3 PC3-12800 FlexXLC Edition Module(s)
- x2 OCZ DDR2 PC2-6400 CL4 FlexXLC Edition Module(s)
- ASUS P5N-T Deluxe (nVidia nForce 780i SLI)
- ASUS P5E3 Deluxe (Intel X38)
- OCZ GameXStream (700 watts)
- Seagate 500GB 7200RPM (Serial ATA II)
- ASUS Radeon HD 3870 X2 (1GB) - ATI Catalyst 8.4
- ASUS GeForce 8800 GTX (768MB)
- ASUS GeForce 8800 GT (512MB) SLI
- ASUS GeForce 8800 GT (512MB)
- ASUS Radeon HD 3870 (512MB) Crossfire - ATI Catalyst 8.1
- ASUS Radeon HD 3870 (512MB) - ATI Catalyst 8.1
Software:
- Microsoft Windows Vista Ultimate (64-bit)
- nForce 9.46
- Intel System Driver 8.4.0.1016
- nVidia Forceware 169.28
-ATI Catalyst 8.4
-ATI Catalyst 8.1
As you can see, the Radeon HD 3870 X2 is only slightly faster than the Radeon HD 3870 Crossfire setup, averaging 2-3 fps more per expansion. However, this makes the Radeon HD 3870 X2 the fastest graphics card outperforming the GeForce 8800 GTX. However, for all the Crysis fans out there, the GeForce 8800 GTX is still the best card for this game.
Benchmark: Company of Heroes
Again we find that the Radeon HD 3870 X2 only shows a few more frames per second than the Radeon HD 3870 Crossfire setup, which we can classify as a small overclocking factory. While the Radeon HD 3870 X2 performs better than the single GeForce 8800 GTX, it still doesn't outperform the SLI GeForce 8800 GT graphics cards.
Benchmark: F.E.A.R
For the first time, we found that the Radeon HD 3870 X2 shows significant performance advantages when running two Crossfire cards. With a 1920 by 1200 resolution and 4 on AA/16xAF, the 3870 X2 was 27% faster than the Crossfire, with a slight 1fps over the GeForce 8800 GT SLI. Stable performance in FEAR, no doubt!
Benchmark: PREY
Prey's benchmark results are very interesting, and despite the fact that the game is a little dated, we wanted to make sense of our research here. Without AA/AF, the Radeon HD 3870 X2 outperforms the Radeon HD 3870 Crossfire setup on all three expansions tested, outperforming the GeForce 8800 GT SLI cards by 1920 by 1200. However, when set to 4xAA/16xAF, it outperforms the 3879 X2 by shifting the Radeon HD The 3870 Crossfire cards are behind the others in two out of three tests.
Benchmark: Supreme Commander
Like most of the games we've tested here, Supreme Commander shows that the Radeon HD 3870 X2 is only marginally superior to the Radeon HD 3870 Crossfire setup.
Since the Radeon HD 3870 smoothes the edges well in Supreme Commander, it was no surprise to us that the X2 outperformed both the single GeForce 8800 GTX and the SLI GeForce 8800 GT.
Benchmark: UT3, World in Conflict
The Radeon HD 3870 X2 was able to outperform the 3870 Crossfire setup by 6-7 fps. however, it was outpaced by GeForce set at 1440 by 900 and 1680 by 1050, keeping the resolution at a maximum of 1920 by 1200.
World conflict shows mixed (different) results, although all maps worked similarly. The Radeon HD 3870 X2 was the fastest solution at the 1440x900 test resolution, however the GeForce 8800 GTX graphics were slightly better.
Absorption of power and temperature
When using a single card, the Radeon HD 3870 X2 consumes slightly less power than the GeForce 8800 GTX at idle, while requiring much more power under load. In fact, when used in the Crysis 3870 X2 it consumes 23% more power than the GeForce 8800 GTX.
When compared to the Crossfire Radeon HD 3870 technology, we find that the 3870 X2 uses 9% less power. This is not like other results we have seen online, but we can assure you that we triple checked our measurements.
We're very happy to see ASUS' improved cooling solution perform so well, as the Extreme AH3870X2 ran slightly "cooler" than a single Radeon HD 3870 card. However, 80 degrees is still too much for our taste, so to speak, with all that hot air circulating around the case. I guess we need to wait for Arctic-Cooling to make Accelero Xtreme for the Radeon HD 3870 X2 before we see better temperatures.
Final Thoughts
Better late than never, AMD/ATI did it! There's no question that the Radeon HD 3870 X2 is the fastest single card to date.
We were impressed with the capabilities of the new Radeon card. Especially because, once installed, it's easy to forget that we were working with two GPUs. The 3870 X2 worked flawlessly with every game we used it for. There was no red tape that usually happens with the Crossfire. We could just sit and relax while playing.
The performance of the Radeon HD 3870 X2 proved to be more stable than we initially expected, as it worked very similarly to the Radeon HD 3870 Crossfire. The advantage is that the Radeon HD 3870 X2 costs slightly less than a pair of $450 Radeon HD 3870s and will work with just about any motherboard, with PCI Express x16 slot.
Aside from a strange crackling sound here and there, which we believe was purely driver related, the ATI Radeon HD 3870 X2 worked perfectly. Of course, we hope that these minor stability flaws will be fixed in the next release of the Catalyst drivers. Further, another disadvantage is that the card heats up more than we would like, leaving a lot of hot air in the case. Not surprising though, since the new generation cards and the X2 cards produce almost the same amount of heat (hot air).
We're pleased with ATI's accomplishments, although it's rather ironic that they showed up here using an already existing product, which makes their victory less impressive. The way things are now, nVidia may find a way to turn things around and create a dual card based on the G92, which will put nVidia back in the lead.
But speaking strictly about the present, in terms of material valuation, we recently bought two ATI Radeon HD 3870 X2 cards for about $460. And, giving a list of the advantages that installing SLI had over the Radeon HD 3870 X2 in some tests, we have to consider what is better (more profitable) to buy. What's more, GeForce 8800 GT 256MB cards cost just over $400, which makes our choice even more difficult.
The Radeon HD 3870 X2 is currently the fastest card (when using a single card) as well as a convenient alternative for both Crossfire and SLI, which require two suitable cards and a specific supporting motherboard. Let's see what nVidia will show us in a few months.
some weird projects...
Now we can lift the veil of secrecy. During the May holidays, when everyone was resting, ATi announced a new line of GPU-based graphics cards Radeon X800 codenamed R420. If you thought that " X” in the name of the chip means support DirectX 10 then you are wrong. X is the usual Roman numeral "10". Just after the line 9xxx I had to come up with a new designation. So the X800 appeared.
R420: old new friend
Monster from nVidia named NV40 consists of 222
million transistors. R420 turned out to be much more modest - “only” 160
million transistors. The ATi GPU is manufactured by 0,13
micron process. So far, there will be only two models in the new line of video cards from ATi - X800Pro And X800XT Platinum Edition (PE). They differ from each other in core and memory frequencies, as well as in the number of pixel pipelines - 12
for X800 Pro and 16
for X800XTPE. X800 series cards use memory GDDR3, which has low heat dissipation. Unlike the GeForce 6800 Ultra, X800-based video cards consume no more energy than Radeon 9800XT And GeForce 5950 Ultra. Therefore, only one additional connector is needed to power the video card. The graphics processor does not get very hot, so the X800 uses the same cooling system as the Radeon 9800XT. Recall that it occupies only one adjacent slot.
Next to the power connector on the board is a video input that can be moved to the front panel system block, video input connector ( 3,5
- or 5,25
-inch compartment). As you may have guessed, the video capture and output function ( VIVO) is now standard. The ATi chip is responsible for it. Rage Theater.
| Technological characteristics of video cards from ATi and nVidia | |||||
| Map | ATI Radeon 9800XT | ATi X800 Pro | ATi X800XT Platinum Edition | nVidia GeForce FX 5950 Ultra | nVidia GeForce 6800 Ultra |
| code name | R360 | R420 | R420 | NV38 | NV40 |
| Chip technology | 256 bit | 256 bit | 256 bit | 256 bit | 256 bit |
| Process technology | 0.15 µm | 0.13 µm low-k | 0.13 µm low-k | 0.13 µm | 0.13 µm |
| Number of transistors | ~107 million | 160 million | 160 million | 130 million | 222 million |
| Memory bus | 256bit GDDR | 256bit GDDR3 | 256bit GDDR3 | 256bit GDDR | 256bit GDDR3 |
| Bandwidth | 23.4 GB/s | 28.8 GB/s | 35.84 GB/s | 30.4 GB/s | 35.2 GB/s |
| AGP | 2x/4x/8x | 2x/4x/8x | 2x/4x/8x | 2x/4x/8x | 2x/4x/8x |
| Memory | 256 MB | 128/256 MB | 128/256 MB | 128/256 MB | 128/256/512 MB |
| GPU frequency | 412 MHz | 475 MHz | 520 MHz | 475 MHz | 400 MHz |
| Memory frequency | 365 MHz (730 DDR) | 450 MHz (900 MHz DDR) | 560 MHz (1120 MHz DDR) | 475 MHz (950 DDR) | 550 MHz (1100 DDR) |
| Number of blocks of vertex programs | 4 | 6 | 6 | FP array | 6 |
| Number of pixel pipelines | 8x1 | 12x1 | 16x1 | 4x2 / 8x0 | 16x1 / 32x0 |
| Version of vertex/pixel programs | 2.0/2.0 | 2.0/2.0 | 2.0/2.0 | 2.0/2.0 | 3.0/3.0 |
| DirectX Version | 9.0 | 9.0 | 9.0 | 9.0 | 9.0c |
| Number of display outputs | 2 | 2 | 2 | 2 | 2 |
| Additionally | TV encoder on a chip; fullstream; adaptive filtering; F-Buffer | TV encoder on a chip; fullstream; adaptive filtering; F-Buffer; compression of 3Dc normal maps; temporal smoothing; VIVO; Smart Shader HD; Smoothvision HD; Hyper Z HD | TV encoder on a chip; fullstream; adaptive filtering; F-Buffer; compression of 3Dc normal maps; temporal smoothing; VIVO; Smart Shader HD; Smoothvision HD; Hyper Z HD | TV encoder on a chip; adaptive filtering; UltraShadow | Video processor and TV encoder on a chip; advanced programmability; adaptive filtering; true trilinear filtering; UltraShadow II |
| Price at the time of release | $499 | $399 | $499 | $499 | $499 |
| Retail price | $440 | $420 | |||
The X800 doesn't have many new features. ATi decided to take the path of further improvement of the proven architecture R3xx. The recipe for success is simple: more vertex and pixel blocks plus some optimizations in the core. The R420 has two really new features: 3Dc And temporal smoothing (Temporal FSAA). We will talk about them later.
Radeon X800 Pro will go on sale in May-June, a bit later ATi will release an older model - X800 XT Platinum Edition. IN Pro versions uses the same graphics chip as the XT PE. But she has 4 pipelines disabled. ATi High Definition Gaming - Hi-Fi in the gaming world
In the world of television today there is a shift towards HDTV (High Definition Television) - high-definition television. ATi decided to use the term HD in their updated technologies: Smart Shader HD, Smoothvision HD And Hyper Z HD.
In fact, the R420 core is a development of the successful and powerful DX9 chip R300 ( Radeon 9700Pro). Like its predecessors, the X800 supports DirectX 9.0 And pixel-vertex programs versions 2.0 . While nVidia added support for pixel and vertex shaders to the GeForce 6800 Ultra 3.0 . In addition, the floating point accuracy of the R420 is limited. 24 bits (recall that NV40 pixel programs can now quickly work with 32 -bit numbers). The X800 Pro/XT PE uses a 256-bit bus divided into four 64-bit channels. ATi has increased the number of vertex units from four (in the Radeon 9800XT) to six (in the X800). It turns out that the X800 is technologically behind the GeForce 6800, but today it is hardly possible to reliably declare the weakness of ATi until DirectX appears. 9.0c and games using shaders 3.0.
3Dc - a new technology for compressing normal maps
In the new R420, ATi engineers used a new technology - 3Dc(For more information about normal maps, see the section “Normal Maps and 3Dc”). It allows you to reduce the size of the normal map file, saving memory. Developers now have two options: they can improve the performance of games by introducing support for compression of these maps, or increase the detail of the game world by using more complex and detailed maps with compression applied. Adding new support
3Dc is hardware supported by all cards based on the R420 core. But you will have to forget about the support of this function in old chips. There is a high probability that in new version DirectX will support 3Dc. In any case, the company is pushing the new technology as an open standard, and we will soon see several games with 3Dc support ( Doom III, Half Life 2 And Serious Sam 2).
For example, texture compression ( S3TC, DXTC) has been used for a long time and allows you to reduce the size of high-resolution textures. Whereintextures are stored in a compressed format.
Many modern games, such as Far Cry, use an improved bump rendering method called “ normal maps” (normal mapping). They are special textures containing information about the details of an object. Using Normal Maps Like Maps irregularities, allows you to increase the detail of the object without resorting to an increase in the number of polygons in the models. In the X800, the company decided to use a new technology for hardware compression of normal map textures - 3Dc.
The essence of the normal map technology is that first the game developer needs to create a very detailed character model using a large number of polygons. The real game character is then created using a simplified model with fewer polygons. After that, the differences between the two models are calculated, which are recorded as a kind of texture (normal maps). They contain details lost in the transition from one model to another. The normal map can then be applied to the simplified model, making it look exactly like a model with a lot of polygons. One hundred percent similarity with the original cannot be achieved, since the normal map does not contain geometric information.
| |
However, there are several disadvantages associated with using normal maps. Firstly, the load on the GPU increases, since normal maps, in fact, are just another texture applied to polygons. Second, more data is needed. The more detail a developer wants to use, the higher the resolution of the normal map used will be - and the more memory bandwidth will be required. Although the normal map can be compressed using the DXTC algorithm, this usually results in noticeable image artifacts. Just as S3 developed its own S3TC technology when problems with large textures arose, ATi came up with a new 3Dc compression technology specifically designed for normal maps. According to ATi, the new method is able to reduce the size of normal maps by a factor of four without any noticeable impact on quality. Smoothvision HD - new full screen smoothing
Video cards from ATi have always been famous for their high-quality implementation full screen anti-aliasing (FSAA - Full Screen Anti-Aliasing). The company's graphics chips are able to support sampling rates up to 6x in combination with color gamma correction at the edges of objects. This gives excellent picture quality.
With the release of the X800 line, the company has implemented a new anti-aliasing technology called “temporal anti-aliasing” (or “temporal” - Temporal AA).
The human eye perceives the sequence of frames on the screen as a constantly moving picture, since the eye cannot notice the change of frames that occurs in milliseconds.
When drawing a frame, TAA changes the arrangement of subpixels - it changes taking into account the inertia of our eye. This allows you to get a higher quality image than with using the regular FSAA.
But temporal smoothing has certain limitations. Let's start with the fact that when using it vertical sync (v-sync) must be enabled. The minimum frame refresh rate should be 58 fps. If the frame rate falls below this limit, then temporal anti-aliasing will automatically change to normal until fps grows again. The thing is that at a lower refresh rate, the differences between frames will become noticeable to the eye. This will degrade the picture quality.
The idea behind the new feature is obvious. 2xTAA(temporal smoothing) provides the same level of quality as 4xFSAA. But the most important thing is that in this case a few resources of the video card are expended (no more than for 2xAA). Temporal smoothing is already implemented in the new drivers. It is possible that this feature will also be supported in 9x00 generation cards in future driver versions. Catalyst(with the exception of 9000 And 9200 , which do not have DX9 support).
Test configuration
Let's not delve further into the theory, but move on to testing the video cards themselves. For testing, we used Catalyst 4.4 driver for cards from ATi, and for nVidia products - the driver ForceWare 60.72.
| test system | |
| CPU | Intel Pentium 4 3.2 GHz |
| FSB frequency | 200 MHz (800 MHz QDR) |
| Motherboard | Gigabyte GA-8IG 1000 PRO (i865) |
| Memory | 2x Kingston PC3500, 1024 MB |
| HDD | Seagate Barracuda 7200.7 120 GB S-ATA (8 MB) |
| DVD | Hitachi GD-7000 |
| LAN | Netgear FA-312 |
| Power Supply | Antec True Control 550W |
| Drivers and settings | |
| Graphics | ATI Catalyst 4.4 NVIDIA 60.72 |
| Chipset | Intel Inf. update |
| OS | Windows XP Pro. SP1a |
| DirectX | DirectX 9.0b |
| Graphics cards used | |
| ATi | Radeon 9800XT (Sapphire) Radeon X800 Pro (ATi) Radeon X800 XT Platinum Edition (ATi) |
| nVidia | GeForce FX 5950 Ultra (Nvidia) GeForce 6800 Ultra (Nvidia) |
Test results
We tested the graphics cards in a variety of games and test applications, including AquaMark3, call of duty, Colin McRae Rally 04, Far Cry, Unreal Tournament 2004, X2: The Threat. We conducted all tests both in normal mode and in “heavy” mode - with anisotropic filtering and full-screen anti-aliasing turned on (except for AquaMark3).
In the AquaMark3 test from Massive Development GeForce 6800 Ultra became the absolute winner. Continuing the winning pace, the NV40 showed the best results in Call of Duty as well. At the same time, it overtook the X800 XT PE in all tests, even in “heavy” modes.
| Test results | |||||
| Radeon 9800XT | Radeon X800 Pro | Radeon X800XT PE | GeForce FX 5950 Ultra | GeForce 6800 Ultra | |
| AquaMark-Normal Quality | |||||
| Score | 46569 | 54080 | 58006 | 44851 | 61873 |
| Call of Duty - Normal Quality | |||||
| 1024x768 | 146,5 | 218,5 | 253,4 | 141,0 | 256,4 |
| 1280x1024 | 101,2 | 156,0 | 195,8 | 97,4 | 219,5 |
| 1600x1200 | 70,7 | 113,5 | 145,5 | 69,6 | 175,2 |
| Call of Duty - 4xFSAA, 8x Aniso | |||||
| 1024x768 | 70,1 | 110,2 | 146,9 | 63,1 | 157,4 |
| 1280x1024 | 47,6 | 75,7 | 100,4 | 42,8 | 110,8 |
| 1600x1200 | 33,1 | 53,7 | 71,3 | 30,5 | 82,1 |
| Colin McRae Rally 04 | |||||
| 1024x768 | 130,5 | 172,5 | 174,8 | 91,2 | 166,0 |
| 1280x1024 | 95,8 | 133,8 | 172,8 | 68,5 | 163,2 |
| 1600x1200 | 67,6 | 95,1 | 141,4 | 49,5 | 132,1 |
| Colin McRae Rally 04 - 4xFSAA, 8x Aniso | |||||
| 1024x768 | 70,5 | 107,6 | 142,2 | 52,3 | 118,0 |
| 1280x1024 | 53,3 | 81,1 | 105,7 | 40,6 | 92,5 |
| 1600x1200 | 39,1 | 59,9 | 76,7 | 30,5 | 70,2 |
| Far Cry 1024x768 | |||||
| normal quality | 55,0 | 75,3 | 81,2 | 48,6 | 66,8 |
| FSAA High, Aniso 4 | 30,3 | 49,0 | 68,8 | 30,7 | 50,5 |
| Far Cry 1024x768 | |||||
| normal quality | 45,1 | 69,6 | 90,8 | 28,5 | 74,7 |
| FSAA High, Aniso 4 | 25,9 | 41,5 | 59,6 | 20,9 | 53,1 |
| Unreal Tournament 2004 - Normal Quality | |||||
| 1024x768 | 106,9 | 104,6 | 105,3 | 104,1 | 103,7 |
| 1280x1024 | 94,4 | 105,0 | 104,9 | 95,7 | 103,6 |
| 1600x1200 | 69,1 | 97,1 | 104,5 | 72,8 | 102,9 |
| Unreal Tournament 2004 - 4xFSAA, 8x Aniso | |||||
| 1024x768 | 75,1 | 104,6 | 105,0 | 80,5 | 102,7 |
| 1280x1024 | 52,5 | 92,2 | 101,9 | 54,7 | 84,9 |
| 1600x1200 | 38,2 | 68,5 | 82,3 | 39,1 | 64,1 |
| X2 - The Threat - Normal Quality | |||||
| 1024x768 | 67,9 | 80,0 | 83,4 | 74,3 | 84,6 |
| 1280x1024 | 54,7 | 68,5 | 76,7 | 61,1 | 75,3 |
| 1600x1200 | 44,2 | 58,9 | 68,4 | 50,5 | 67,1 |
| X2 - The Threat - 4xFSAA, 8x Aniso | |||||
| 1024x768 | 48,9 | 62,4 | 69,7 | 53,9 | 73,2 |
| 1280x1024 | 36,1 | 51,1 | 58,9 | 40,1 | 61,1 |
| 1600x1200 | 28,4 | 42,6 | 49,8 | 30,6 | 51,8 |
In the next test, ATi bounced back in full. In Colin McRae Rally 04, the X800 XT PE turned out to be head and shoulders above its rivals, especially in the mode with anisotropic filtering and full-screen anti-aliasing enabled. The situation repeated itself in the game Far Cry - the victory was again for the flagship from ATi. The next game in which we tested video cards was Unreal Tournament 2004. In normal mode, all three cards showed approximately equal results. Inclusion ANISO And FSAA completely changed the picture: the X800 Pro and X800 XT PE just went into the lead! At the same time, even the Pro-version managed to overtake the GeForce 6800 Ultra. In the last test - X2: The Threat - the test results for NV40 and X800 XT PE turned out to be approximately equal.
Conclusion
We didn't have time to fully recover from the impressive results shown by nVidia GeForce 6800 Ultra, when ATi surprised us now. The Radeon X800 XT Platinum Edition showed very high performance, even the X800 Pro with 12 pipelines outperformed the GeForce 6800 Ultra in some tests.
Canadians from ATi did a great job. The power consumption of the X800 series cards turned out to be almost at the same level as that of the predecessor 9800XT. That is why the new cards from ATi require only one power connector, unlike the GeForce 6800 Ultra, which needs two. The R420 core also turned out to be less hot. It is cooled by a standard cooler that occupies only one adjacent slot (GeForce 6800 Ultra has two). The R420 core has many innovations, including the ATi Rage Theater chip with VIVO support, innovative 3Dc technology that can improve graphics quality in games, and the original Temporal Full Screen Anti-Aliasing (Temporal FSAA) technology.
| |
X800 XT Platinum Edition and GeForce 6800 Ultra show incredible performance. But the X800 XT PE looks better. This graphics card from ATi showed very high performance in high-tech modern games such as Unreal Tournament 2004, Far Cry and Colin McRae Rally 04.
A new round of confrontation between the two companies has just begun, and it is too early to sum up the final results. In the near future, budget options for video cards from two companies will appear, as well as cards with support for PCI Express. So we will definitely return to the topic of confrontation between the Canadian company ATi and the American nVidia, and more than once.
| power usage In the article about NV40 we talked about the high gluttony of the GeForce 6800 Ultra. In this article, we conducted a test in which we found out how much energy modern video cards consume. Since this cannot be done separately for cards, our table shows the power consumption values \u200b\u200bof the entire computer. We used the same system configuration for all cards.
|
||||||||||||||||||
Loading...