One and a half to two years, two giants are faced with foreheads-AMD and NVIDIA. The last time we tried them when the GeForce GTX 2×0 and Radeon HD 4XX0 came out – and recognized a draw. But today, AMD strikes again: Radeon HD 5870 appeared on the market. Samples of a new video card in Russia can be counted on the fingers, but we managed to get the pre -sale version of the PowerColor Radeon HD 5870 – in full configuration and with fresh drivers. We compared the novelty to Radeon HD 4870 X2 and GeForce GTX 295 – and this is what we got.
Rustled packaging
Radeon HD 5870 Longer Radeon HD 4870 x2 a couple of centimeters – this can be seen even with the naked eye. The cooling system is to blame: it remains the same – the usual turbine takes the air from the case, drives it through the entire map and throws it outside the computer – but the cooler’s metal casing does not just close the entire surface of the board, but stands a little beyond it, adding extra centimeters.
At the same time, the hole for ejecting hot air doubled due to the fact that additional exits to the monitors appeared. Radeon HD 5870 is equipped with two DVI, one DisplayPort and one HDMI – you can connect up to three displays in total. They are supported by Ati Eyefinity technology – and today it is the only way to play three screens at once. At the peak, the video card is able to provide a resolution of 8192×8192-this is more than three 30-inch monitors can give out.
AMD HDMI output drivers promises to update soon. Today not a single video card (and even audio payment!) cannot display the sound from licensed Blu-ray to home cinemas: on Blu-ray and sound and video are protected from copying. Nobody can agree with players' manufacturers, but AMD is going.
There are also two connectors for the organization of Crossfire and a couple of six -pine contacts. The miracle, by the way, did not happen: the new AMD flagship consumes a huge amount of energy – 188 watts. Even more than the Radeon HD 4890, which required from 150 to 180 watts under load. But in 2D mode, the new card reduces the appetite to 27 watts, which is very good compared to 90 watts of Radeon HD 4890.
Technology parade
For the release of Radeon HD 5870 in AMD were prepared in advance. So, for the new RV870 graphic processor, also known as CYPRESS, they prepared 40-nm technology process and even drove it to RADEON HD 4770 chips-as a result, AMD receives a decent percentage of crystals suitable for work.
Despite the advanced process, the physical dimensions of the graphic processor have grown. The chip area increased from 263 mm2 in RV770 to 334 mm2. This is explained simply: if in the last crystal there were 0.959 billion transistors, then Cypress installed 2.15 billion. Today it is an absolute record, even the NVIDIA GT200 is a only 1.4 billion transistors in its 576 mm2.
Further numbers only grow. The number of SIMDs was increased from 10 to 20 pieces, they gave each of them 16 stream processors equipped with 5 cores-and received 1600 streaming nuclei. So, CyPress, operating at a frequency of 850 MHz, in perfect conditions provides computing power up to 2.72 theraflops. In addition, the number of texture blocks has also doubled with the doubling of the amount of SIMD – now there are 80. Here we will also write down a double set of rasterization blocks responsible for filling the picture with pixels – in Radeon HD 5870 there are 32 of them.
The throughput also increased-to 153.6 Gb/s (it could be more, but the old 256-bit tire imposes its restrictions). AMD continued to produce boards with GDDR 5 and installed on the video card 1 GB of memory, adding it 1200 MHz speeds. Cypress also received blocks for working with Tesselia, one of the main innovations DirectX 11. But the blocks responsible for interpolation have disappeared – now stream processors will do this. True, AMD was not able to teach the Radeon HD 5870 to process more than one triangle for the passage, and this is the main restriction of all video cards, into which performance rests on.
Great and terrible
Of course, we, of course, evaluated the Radeon HD 5870 performance by old benchmarks left from games released under DirectX 9 and 10. However, do not talk about DirectX 11, rumors about which they go from the very announcement of Windows 7, cannot be.
It was assumed that Microsoft will launch Windows 7 and DirectX 11 at the same time. But the company remembers old misses: DX11 will appear only after a few months, and you will not have to change the OS – the API will work with Windows Vista.
What is DirectX 11 good? It’s easier to program it: “Microsoft” simplified the API language itself. In addition, DX11 even draws graphics cards of medium -level, so you will not have to collect two versions for different APIs, as before. For example, Devil May Cry 4 had a separate version for DX9 and separate – for DX10. But the main thing is, of course, the tessellation that will facilitate the miscalculation of multi -polygonal objects.
Games from year to year become more complicated and detailed. The textures and lighting improve, the number of landfills increases, but at the same time, the cost of development is growing, and the performance of medium computers is not enough to calculate all this beauty. The worst of all is video memory – she lacks a bandwidth. Tesselia will facilitate video memory life.
The idea is as follows. Instead of making a complex detailed model weighing several megabytes, the developers create a light low -polygonal object and load it into the memory of the video card. From there, the object enters special teseling blocks, which independently read the model for certain algorithms and pump it with landfills – as a result, the memory is unloaded, because it no longer needs to store unnecessary information.
In addition, Tessellation will solve the problem with the level of detail. Imagine that in the game you are in the forest. Worms are visible on the bark of the trees, the leaves sway in the wind, proteins jump on the branches. But here you go into the field. The forest remains far behind, the squirrels have long disappeared from sight. The game knows this and, in order to save a video card resources, replaces the highly polygonal models of trees more simple – you have already moved over a decent distance and obviously will not consider the wormhole. It would seem great! But, firstly, the developers have to create and store several versions of one object at once. Secondly, if you slowly move away from the forest and look at this time at the trees, the replacement moment is easy to notice.
DirectX 11 will cope with this. Now developers will be able to download the simplest model, and its scaling will be engaged in the video card, smoothly reducing or increasing the detail. Plus, she herself will change the lighting and apply textures on the model, which, again, will save the artists from excess work.
It is interesting that the tesselar appeared back in 2001 – the first technology was presented by ATI, then NVIDIA pulled up. But since the implementation of the technology from companies differed, the method itself did not gain particular popularity, because the developers had to write code for two graphics lines at once – with the output of DX11, the process is unified by the process. By the way, all the ports with the Xbox 360 will have to support the tessellation, where it is used from the very release of the console.
Another processor
Another Innovation DX11 – Compute Shader, a kind of NVIDIA CUDA version. CUDA technology, if you remember, taught how to take on a part of the tasks of the central processor: encode videos, engage in calculations in mathematical programs, etc.D. AMD has a similar solution – called Brook+ – but it is not moving so intense. Naturally, Microsoft offered its own option to use GPU for general calculations – and its first implementation appeared just in DX11.
Compute Shader is primarily focused on games: it is planned that he will fall on the shoulders of the video card the calculations of physics or artificial intelligence, will take up post -strobot – for example, it will at the final stage apply HDR effects. And since NVIDIA, together with Intel, actively promote the use of weak video cards for physics calculations, Compute Shader may come to the place.
Information flows
The latest innovation DX11 is parallel flows. Oddly enough, but the developers of the games have not learned how to download the quad -core Core i7 to the fullest. Typically, the main stream of data falls on one core, and the rest are engaged in "small things" like sound or physics. DX11 will appear a function that will allow you to load one core with the calculation of rendering results, the other – HDR, the third – work with drivers, etc.D. Interestingly, they are going to connect this function to the old DX10 and DX10.1 – after patches, some games will earn much faster.
There are also many smaller functions in DX11. API taught to more effectively compress textures; Added support for HDR, LDR and Shader Model 5.0 – with them the water in games will become even more beautiful.
Test stand
For testing Radeon HD 5870, we have collected a very powerful computer. It was based on the FoxConn motherboard based on Intel X58 Chipset. We equipped it with the Core i7 920 processor, accelerated from 2.67 to 3.2 GHz. Cooler Master Hyper N620 was put on cooling the processor and added three planks of overclocking memory Kingston Hyperx DDR3-1600 with a total volume of 6 GB. We have an open stand, so there were no problems with installing a long PowerColor Radeon HD 5870: it was only necessary to insert it into the PCIe X16 and connect a couple of six -side power cables. All this was running 64-bit Windows Vista Ultimate.
The novelty was compared with the AMD Radeon HD 4870 X2 and Nvidia GeForce GTX 295. Unfortunately, not all benchmarks were identified by the new video card and we had to abandon 3DMark Vantage in favor of a less fundamental 3DMark06 (we drove synthetic tests in the standard mode established by the developers). And we had four for the games: Crysis, Resident Evil 5, Call of Juarez: Bound in Blood, Devil May Cry 4 – We came up with special responsibility. We refused the run in the modes without smoothing and filtering, permissions lower, and the easiest option was installed 1680×1050, AF X16, AA X4.
We work
Before we move on to the analysis of test results, we will talk about our impressions of working with Radeon HD 5870. The first thing we noticed is silence when starting the system. If earlier, all AMD-video cards, when the computer turned on, spun the fans with might and main and reduced the speed only after loading Windows, then the Radeon HD 5870 works silently.
But the ATI Catalyst Control Center has remained the same. We have one main window, and all options are removed in two drop -down lists. The first is the settings of graphics and monitor, in which, in essence, nothing has changed. The second list, Hydra, appeared not so long ago. Here you can configure several desktop at once and tie each program to a specific place on the screen.
Finally, we note that we established the final version of the drivers (at that time – 8.66 RC6) and no problems during tests arose.
For sweets
Finally we approached the most interesting – performance tests. The first we launched Crysis, putting 1920×1080 with AFX16 and AAX8 at once – and the game without brakes passed a test level with a indicator of 50 frames per second. And with such settings, this is no doubt a record. The rest of the tests only confirmed the unambiguous leadership of the Radeon HD 5870.
The most fully card is revealed at new games: in Call of Juarez: Bound In Blood, our Radeon HD 5870 overtook GTX 295 by 11 frames. But this is attention! – 35% faster than two GeForce GTX 285! In other games, the speed growth was about the same-10-15%. Only the Devil May Cry 4 test is not representative: the version of the DX10 refused to work with the Radeon HD 4870 X2 and GeForce GTX 295, and they had to run the DX9 version on these cards – they scored significantly more.
It is too early to rejoice?
AMD, of course, took a huge step forward. Radeon HD 5870 can be compared with the Radeon HD 4870 X2 – cards have almost the same characteristics, but at the same time HD 5870 gives out much higher results. Of course, a double increase in performance did not occur, but the Radeon HD 5870 is the first one -chip solution on which Crysis works without brakes, and this says a lot.
In addition, AMD has a competent pricing policy. The official price of a top video card is $ 400 (about 12,000 rubles.). Of course, we will raise this figure to 14,000-15,000 rubles., But only for the first time. Remember that the flagships from Nvidia cost at the beginning of sales no less than 18,000 rubles.
However, the victory of AMD cannot be called unequivocal. The new generation of Radeon HD 5xx0 came out too early – from the appearance of the last series, Radeon HD 4XX0, less than a year and a half has passed. It is in a hurry by AMD – it is clear: she tries to gain a foothold in the new generation video cards market, because very soon she will have to take the NVIDIA blow, which is preparing for the launch of GT300 and rolls up a new process on the GeForce GT 220 and 210.
Most likely, AMD will prepare the Radeon HD 5870 X2 for GT300. Plus, the AMD will have a established production of video cards and stable drivers, while NVIDIA will have to fight small errors for a long time. In addition, AMD will try to settle in the average and low price segment: very soon the Radeon HD 5850 and HD 5770 should go on sale. In general, the beginning of the year will be hot – and at JackBit Casino the end of this year the video card is faster than the PowerColor Radeon HD 5870.
P.S. We tested the pre -sale version of the Radeon HD 5870, which has already appeared in stores today. During the tests, the Iron Workshop used the latest version of drivers 8.66 RC6 provided to us by the manufacturer. Compared to beta version of the month ago, a huge number of errors have been corrected in this release. In this regard, the results of our tests may be very different from what you can find in other sources of information.
Table 1
Technical characteristics
PowerColor Radeon HD 5870 Radeon HD 4870 X2 GeForce GTX 295
The core RV870, CyPress RV780 GT200B
The number of transistors 2.15 billion 2x 0.956 billion 2x 1.4 billion
Process process 40-nm 55-nm 55-nm
The number of stream processors 1600 2x 800 2x 240
The frequency of the graphic nucleus 850 MHz 750 MHz 576 MHz
Frequency of stream processors 850 MHz 750 MHz 1242 MHz
Type, memory volume GDDR5, 1 GB GDDR5, 2x 1 GB GDDR3, 1792 MB
Memory frequency 4800 MHz 3600 MHz 1998 MHz
Data tire 256 bits 256 bits 448 bits
The number of texture blocks 80 2x 40 2x 80
The number of rasterization blocks 32 2x 16 2x 28
Interface PCIE 2.0 x16 PCIE 2.0 x16 PCIE 2.0 x16
Price (for October 2009) from 12,000 rubles from 15,000 rubles from 14,500 rubles
Table 2
Synthetic tests
3DMark06
SM2.0
SM3.0
CPU
Overall
PowerColor Radeon HD 5870
6678
8362
4766
17300
Radeon HD 4870 X2
6546
8290
4752
17106
GeForce GTX 295
6521
9063
4763
17784
Table 3
Game tests
Crysis 1.2.1
PowerColor Radeon HD 5870 Radeon HD 4870 X2 GeForce GTX 295
High, 1680×1050, AA X4, AF X16 50 40.01 51.34
High, 1920×1080, AA X4, AF X16 45.87 35.52 45.02
High, 1680×1050, AA X8, AF X16 54.86 42.25 49
High, 1920×1080, AA X8, AF X16 47.8 37.69 44.34
Devil May Cry 4 (SC2)
1680×1050, AA X4, AF X16 148.06 184.1 197.9
1680×1050, AA X8, AF X16 140.46 173.59 165,13
Call of Juarez: Bound in Blood
1680×1050, AA X4, AF X16 72 65.5 44.3
1920×1080, AA X4, AF X16 67.1 20 60.1
1680×1050, AA X8, AF X16 70.3 39.9 53.1
1920×1080, AA X8, AF X16 65.7 22.1 54.3
Resident Evil 5
1680×1050, AA X4, AF X16 101.8 96.3 93.5
1920×1080, AA X4, AF X16 98.2 95.2 91.8
1680×1050, AA X8, AF X16 96.2 96 93.9
1920×1080, AA X8, AF X16 95.5 95.1 90.6
[[Break]] what's in the forehead, on the forehead
With the release of the new generation of video cards, AMD changes the marking of its graphic processors: now there are no games with numbers, only its own names. The series is called Evergreen ("evergreen"), and it includes CyPress processors ("cypress"), Juniper (Juniper), Redwood (Sequoia), Cedar ("Cedar") and Hemlock ("Tsuga" – this is an American conifer).
So, the fastest processor that is placed on the Radeon HD 5870 and HD 5850 – CyPress. Juniper is a little weaker, they will make boards of the middle price category (about 7500 rubles.) – this is Radeon HD 5770 and HD 5750. They will also support Eyefinity and DX11, but at the same time will lose part of the computing power: they will receive only 1120 stream processors, 56 texture blocks and a 128-bit memory bus.
In 2010, the division will continue. So, Juniper will be divided into Redwood and Cedar, which are planned to be transferred to a 32-nm process of process. Redwood will work on the cards of the Radeon HD 5600 series (about 4500 rubles.), Cedar – on Radeon HD 5500 (3000 rubles.). We should also be presented by Hemlock, which is supposedly installed on the Radeon HD 5870 X2.
Radeon HD 5850
Not everyone can afford a top video card: Radeon HD 5870 is good for everyone, but at a price of 12,000 rubles – an expensive pleasure. Fortunately, AMD has a more acceptable solution – Radeon HD 5850 for 7500 rubles. The HD 5850 is based on the same CyPress, which has lost quite a bit: two SIMD processors and 125 MHz at a crystal speed; The speed of memory of memory fell to 2 GHz. But electricity consumes a little less – 160 watts. Unfortunately, we did not have time to get a working sample of this video card for tests, but, according to some reports, it should overtake the GeForce GTX 285.
What is NVIDIA preparing us
At the beginning of 2010, Radeon HD 5870 will receive a powerful competitor – NVIDIA graphics card on the new architecture Fermi , named after the Laureate of the Nobel Prize of Italian physicist Enrico Fermi. Nvidia Fermi is unofficially called GT300, but we were confirmed in NVIDIA that this is not the final marking.
About the new line of video cards from NVIDIA is known very little. Fermi is a new platform that was originally created for calculations, and not just for processing graphics. Like the Radeon HD 5870, GT300 will be created at 40-nm process, but it will have almost one and a half times more transistors-3 billion. In the top version there will also be 512 stream processors (in Fermi they are called Cuda Cores – The nuclei are optimized for GPGPU calculations). Stream processors will be combined into clusters of 32 pieces in each (in GT200 there were 8 processors each). Each of the arrays will be equipped with its cache, and it will be used not only for storing textures, but also for general purpose information. Work with a cache is generally very processed: NVIDIA suggests that they suggest that a multi-level cache will allow you to increase the effectiveness of computing several times.
In addition, NVIDIA will finally translate its video cards to GDDR5-for this, GT300 was taught to understand 64-bit addressing. GDDR5 will be able to work with volumes up to 6 GB at a frequency of 4000 MHz. Of course, this will not be implemented soon, the first GT300 will appear with only 1.5 GB GDDR5 on board, but the prospect is still impressive.
Of course, the GT300 is friends with the DX11 and will continue to support all the branded NVIDIA technologies – the hardware acceleration of physics, the creation of stereoscopic effects, Ray Tracing, etc.D.
Unfortunately, these information crumbs are all that is known for today about GT300. Nvidia does not even disclose information about the speed of work in games. But then she attracted attention to the GT300, be healthy: just think, even manufacturers of supercomputers say that Fermi can achieve a much higher speed of work than today's computing centers provide.
The latest story
After grandiose success GeForce 8800 GTX Nvidia missed a little marketing policy and arranged a wild confusion in the video cards market. To begin with, she released a series GeForce 9800 , which, in fact, was just a slightly updated version of the previous generation. Having successfully sold the reserves of old crystals, just a few months later, the Californians presented a full -fledged new generation with the flagship GeForce GTX 280. The card turned out to be surprisingly powerful, and NVIDIA, not expecting a trick from competitors, inflated prices to a prohibitive level.
AMD at that moment was, to put it mildly, not in the best position. Series Radeon HD 38×0 did not differ in high performance and was sold only in the average price segment. Radeon HD 3870 x2 There were also not enough stars from the sky, nothing special was expected from AMD. And here, immediately after the release of the GTX 280, AMD released the Radeon HD 4870. And even if these were not the fastest video cards in the world, but they were sold at a very attractive price. In terms of speed, the Radeon HD 4870 by 25-30% lagged behind the top-end GTX 280, but cost almost one and a half times cheaper-and this is despite the fact that the excess performance of the GTX 280 in games at that time was not particularly required. It is easy to guess that the Radeon HD 4870.
Such a blow to Nvidia survived with difficulty. In the shortest possible time, she dropped prices for her flagships and missed again – Radeon HD 4870 X2 came out, which immediately selected the title of the fastest video card in the world from the GTX 280. Of course, then the GeForce GTX 295 appeared and Nvidia sighed more freely, but the fact remains – once intercepting the leadership of Nvidia, AMD will no longer stop.