Asus AMD Radeon HD7970 is the "dark knight" of the gaming world. Typical model of a well-known manufacturer

Radeon HD 7970: transition to 28nm process technology

At the beginning of December, we did not expect that the next generation video card would be with us until 2012. Even in the middle of the month, after we received the official notification from AMD, we planned to release this review in January. Windows 8 and the DirectX 11.1 API that comes with the new OS won't be available for the next few months, and current high-end graphics cards are great for today's games. Despite the fact that AMD, according to rumors, a few months ago stopped the release of the flagship model Radeon HD 6990, we assumed that the difficulty of moving to 28nm at TSMC, whose manufacturing facility AMD uses, left little chance of new GPUs being introduced on time.

However, AMD eventually set an announcement date for mid-December, which surprised us even more. Here's how the official sounded AMD version: "After collecting feedback from our partners and evaluating our overall readiness… we are convinced that the new announcement date will allow us to capture the Christmas holiday season and CES." Trying to capture the Christmas season by announcing the map a couple of days before Catholic Christmas is an extremely optimistic strategy, especially considering that the first official shipments will begin after January 9th. The unfortunate result of all this is that many developers collaborating with AMD did not have time to prepare the appropriate software updates needed even to properly test the HD 7970.

Meet the Radeon HD 7970

Whether you're ready Radeon HD 7970 meet the world and is the world ready to meet her, she got into the laboratory and passed a series of tests. This card is the first representative of the new AMD generations and has serious differences from the models of the current 6000 series.

AMD uses the new Southern Islands architecture and has provided support for new features and technologies, including compatibility with DirectX 11.1. The flagship GPU Tahiti (Tahiti) consists of 4.31 billion transistors and is thus approximately 160% given parameter outperforms its predecessor on the Cayman core.

At the same time, AMD made the transition to the 28 nm process technology, which made it possible to cram so many transistors into a 365 mm² silicon wafer - the Cayman chip occupies an even larger area (389 mm 2).

Before diving into the intricacies of the Southern Islands, let's take a closer look at the main specs. Radeon HD 7970 compared to its predecessors and competitors.

Compared to the Radeon HD 6970, the new product has clear advantages: the number of stream processors and texture processing units has increased by a third, throughput memory increased by 65% ​​due to the use of a 384-bit memory bus. Only two rows of model specifications Radeon HD 7970 and HD 6970 are the same: number of Color ROPs and TDP 250W.

Based on these specs (and the fact that the new card will cost around $550), we can assume that Radeon HD 7970 surpass 6970, slightly ahead GeForce GTX 580, but will be inferior to a dual-processor card Radeon HD 6990. Undoubtedly, purely technical characteristics in practice are less important than gaming performance tests. But first, let's take a closer look at the functional features of the new generation. Radeon video cards and the Graphics Core Next architecture.

First, let's talk about what we know about the Radeon HD 7000 series. Despite rumors to the contrary, all 28nm GPUs in the Radeon 7000 line, formerly codenamed Southern Islands, are based on the Graphics Core Next architecture. It includes the Radeon HD 7700 series (Cape Verde core), 7800 (Pitcairn) and 7900 (Tahiti). In addition, AMD may include some 40nm products in the new 7000 line by rebranding older models that use the current VLIW4/5 architecture.

The models implemented on the basis of the new architecture have the same functionality, which is good news. The following figure shows the position of the new cards in the lineup relative to the Radeon HD 6000 series.

As you can see from our tests, the Radeon HD 7900 series is positioned as a powerful single-GPU gaming card, approaching the current dual-GPU flagship, the HD 6990, in performance. based on two Tahiti GPUs, designed to be at the top of the Radeon food chain.

Now we know what position the representatives of the Radeon HD 7000 series will take in accordance with the plans of AMD's marketing department. Next, we will elaborate on unique features new line, but let's start with the main question - what is the architecture of the Southern Islands?

Graphics Core Next and Southern Islands architecture

Radeon HD 7970 is the first commercial product on the Graphics Core Next architecture, but the design of the new GPUs is no secret in itself. To give software developers the opportunity to better understand the features of future hardware, AMD revealed the features of the Graphics Core Next architecture at the AMD Fusion'11 Developer Summit conference, held from June 13 to 16 in Bellevue, USA.

According to Eric Demers, CTO of AMD's Graphics Products Division, the current VLIW architecture, originating from the Radeon 2000 line, still has the potential to build powerful graphics cards. But it is limited in terms of functionality. Instead of keeping the old architecture alive, AMD has chosen a different path - to invest in the development of a completely new architecture. For high-end cards, support for the latest games and image quality are at the forefront. To match this setup, AMD dropped further work over the VLIW (Very Long Instruction Word) architecture in favor of new Graphics CoreNext.

The Efficiency Benefit of the Graphics Core Next Architecture

The VLIW architecture is very efficient in processing graphics instructions. Its compiler is optimized to markup the scalar mathematical transformations that underlie 3D graphics calculations. The weakness of VLIW is when the GPU has to queue jobs with scalar instructions in general purpose applications.

Sometimes it happens that a set of instructions, called an instruction package (wavefront - "front of instructions"), cannot be executed until calculations are performed on another similar set. This is called "dependencies". The problem is that the compiler cannot change the execution order of a batch of instructions once the job queue has been defined. Therefore, the precious potential of an ALU is often wasted when instructions are waiting in line and the dependencies that pertain to them are not fulfilled.

Here is a theoretical example of how the SIMD engine in the VLIW4 architecture radeon cards The HD 6970 and its 16 banks of shader processors (each shader processor has four ALUs, for a total of 64 ALUs per SIMD engine) process an instruction packet that includes dependencies.

As long as three ALUs remain free, the next packet is idle in the queue. As you can see, there is an inefficient use of computing resources, due to the overly simple logic of processing instructions in the outdated architecture.

How to optimize a set of scalar operations that are performed per cycle? It is necessary to introduce a calculation unit - Copmute Unit or CU, which will replace the traditional SIMD engines for AMD.

Each CU has four Vector Units (VUs), which in turn include 16 ALUs, for a total of 64 ALUs per compute unit. Thus, the number of ALUs in the calculation unit (CU) is exactly the same as in the SIMD engine. The difference is that, unlike the shader processors in the SIMD engine, each of the four vector operation units included in the Copmute Unit can have an independent task queue. The CU has its own queue scheduler, which allows free VUs to work with new batches of instructions, eliminating the "bottleneck" in the processing of dependencies. This is the key difference of the new architecture, thanks to which each VU can work with new packets of instructions, even if there is a dependency queue.

In our example, the same instruction queue can be executed in six cycles on the VLIW4 architecture, but in four cycles on the Graphics Core Next. AMD suggests that Radeon HD 7970 can achieve up to 7.5 times the theoretical peak performance of the Radeon HD 6970 through more efficient use of computing resources.

The difference in real applications depends on the efficiency of the code compiler. In some computing problems Radeon HD 7970 will be better only due to more ALU and higher core and memory frequencies. In the course of testing the novelty, we saw many variations, and you will see them too. But, based on the results of synthetic benchmarks, we can confidently say that the computing potential of the Graphics Core Next is superior to VLIW4.

graphics card temperature chart

Even if the novelty showed not the best "temperatures", after such high performance results, we could forgive her for that. But our generosity is not necessary. The Radeon HD 7970 runs cooler than its predecessor (Radeon HD 6970) with much better performance. At the same time, as I said above, if you do not change the fan speed yourself, then you will have to try very hard to hear the card. I don't think it's possible with the case closed.

When overclocked, the temperatures are lower than at standard settings, due to the same manual increase in speed for stability.

By the way, we have a few more tests of the same video card with an alternative OS on another system. You can view the results to understand how modern processors affect performance in popular apps. It is worth noting that the stock cooling system on the video card manufactured by ASUS lost only one degree in nominal value to the dual-fan system from XFX, this is a good result.

Conclusion

The overclocking of the ASUS HD 7970 video card was a pleasant surprise. The GPU overclocked from the default 925 MHz to 1200 without increasing the voltage. This is the reason for such a high increase in performance during overclocking. To be honest, this is a very pleasant moment. Recently, video cards have been accelerating worse and worse, so the need for overclocking disappears: if the video card does not pull the game in default, then poor overclocking by 50 MHz will not correct the situation. This is clearly a different case.

The only thing that confuses the new product is the exorbitant noise level when manually increasing the speed. And it is necessary for successful overclocking, otherwise you have every chance to burn the video card. But, I think, over time, this problem will be solved - there will be many alternative cooling systems. The main thing is that the overclocking potential does not change for the worse.

High performance, low heat generation, and new proprietary technologies make the AMD Radeon HD 7970 graphics card the best single-GPU graphics card available today. Moreover, when overclocked, the new product easily competes with the dual-processor monster Radeon HD 6990. Therefore, if you have a question: “which is better: HD 7970 or HD 6990?”, Then I recommend choosing the first option. With it, you will not have problems when activating CrossFireX, there will be no sudden drops in performance, as is the case with dual-headed cards. Three gigabytes of video memory - nice bonus to purchase. Unfortunately, they will be used extremely rarely, but it is better with them than without them.

In general, the ASUS HD 7970 video card was a success and leaves a very pleasant impression. Let's hope that the continuation of the new line, which will be released very soon, will please us no less.

Mining Bitcoin and other cryptocurrencies using a video card is a promising way to make money for users who know how to take risks. The demand for this equipment is constantly growing. New high-performance models are ahead of their predecessors, among which the powerful Radeon HD 7970 video card is currently in the lead. For mining, this is perfect option. A single-chip graphics card from AMD is suitable for mining Bitcoin, Ethereum and ZCach.

The energy-efficient and feature-rich product is based on an advanced architecture chip that is manufactured using a 28nm process. This video card definitely deserves praise. It greatly outperforms previous models. In addition to the high price, it has practically no drawbacks.

Mining on a video card: pros and cons

Miners with experience prefer to mine cryptocurrency on video cards. They assemble the hardware themselves, completing the device with the necessary details. The resulting power allows you to achieve high mining efficiency. The profit is excellent, the investment pays off quickly. The advantages of mining cryptocurrencies in this way are obvious:

• Passive income. One video card or four together with other devices provides high power and performance.
• Saving time. Quality equipment rarely fails. Checking its performance and temperature takes a couple of minutes.
• Stable income. You can mine not only Bitcoin, but also other cryptocurrencies.

If you insure yourself against trouble, mining on the Radeon HD 7970, like on its predecessor video cards, will bring a pleasure. For uninterrupted supply electricity, you need to install special batteries. This will prevent malfunctions.

When manually increasing the speed of the turbine, there is a strong noise. This is a significant disadvantage. But there is a way out - installing an air conditioner in the room. It will drown out the noise, creating a comfortable atmosphere.

Video card features

AMD for graphic solutions used the VLIW architecture for a long time. And until the advent of video cards, the HD 6000 did not change anything from the beginning of the existence of the Radeon HD 2000. And only with the beginning of the history of the HD 7000 series did the GCN architecture begin to be used. It was developed specifically to improve the efficiency of calculations that are not directly related to 3D.

On the new microarchitecture, the GPU executes, for example, certain instructions already in 4 cycles, and not in 6, as on VLIW. Thanks to this, the performance of the innovative graphics card Radeon HD 7970 for mining is several times greater compared to the HD 6970 - GCN computing resources are more efficiently used and distributed.

Briefly about AMD PowerTune and ZeroCore

Another innovation. A technology that controls the power consumption of the GPU. Helps to achieve good performance without fear that consumption will go beyond the norm. A similar technology was used before, but in the new video card it went to a completely different level.

ZeroCore is also an innovation. This technology reduces GPU power consumption during idle. And it puts the processor into "sleep" if there is no signal from the PC to the monitor.

PCI Express 3.0

Support for Gen 3 PCI Express, introduced in the new 7970. Helpful in eliminating weaknesses in the process of working with complex applications. New technologies CrossFire, AMD HD3D and Eyefinity - their video card also supports (simultaneous operation of these technologies).

Other features of the new AMD graphics card include:

• RAM - 3 GB
• Power consumption - 250 W
• Increased GPU clock speed - up to 925 MHz
• Computing power - 3.79 TFLOPS (increased by 40%)
• Texture blocks - 128 pieces (predecessor had 96)
• Texturing speed - 118.4 GT / s
• The number of transistors GPU "Tahiti" - 4.31 billion

But one of the most significant innovations is the production of the Tahiti GPU using a 28-nanometer process technology. As for the memory, almost nothing has changed here. The HD 7970 uses 5500 MHz of memory. The bus is 384 bits. This increased the memory bandwidth from 176 GB/s to 264 GB/s.

The potential of the 7970 video card

Mining Ethereum, Bitcoin and Zikesh on the Radeon 7970 remains relevant today. The video card has excellent performance and excellent overclocking potential. If you try to overclock it, you can achieve an impressive result: the core frequency is 1160 MHz, and the memory frequency is 7160 MHz. But this is when the turbine is controlled in automatic mode.

You can overclock the video card even more. To do this, you need to wind the turbine speed manually. But the resulting noise will exceed all expectations. It is better to give preference to the core frequency of 1160 MHz for acoustic comfort.

Users often have a question, will the video card burn out with strong overclocking? Won't burn. It is only necessary not to increase the voltage supplied to the GPU. In case of failure, all frequencies will be reset to initial values and the video card will continue to work.

Mining Ethereum

And now it's time for the main point. How to mine cryptocurrency on a Radeon video card with such excellent characteristics? The performance of the model is high, the overclocking potential is excellent, and the power consumption cannot be called too high - only 250 watts. It remains only to connect the device and start earning by replenishing with the help of the unique Radeon HD 7970 ETH cryptocurrency wallet. Mining starts after setting up the program.

New AMD graphics card recognized as the best for Ethereum mining. The Ethereum algorithm is extremely demanding on memory. In this model, it is 3GB, although 2GB is enough. So you can safely start mining a promising coin, the rate of which never ceases to please.

Ethereum mining instructions for 7970

The first step is to select a pool. You can mine alone too, but it's too expensive. Achieving a good result will help to unite with other participants. And the more users in the pool, the better. The operation of this service is simple. It distributes the calculation task among all users. One hit the target - the block was formed. All participants received a reward in cryptocurrency video. So, let's start mining ether (using the Ethminer program):

1. We enter the pool, fill out the form and activate the account through the link in the letter that will be sent to the email. Mail.
2. Having entered the account, we create a worker (a worker is a key that connects the pool and the processor). To do this, select the tab with such keys, come up with a username and password, then click “Add”.
3. Set up automatic payments. Specify the number (from 0.01 ETH) and wallet data.
4. We choose the program. When the worker appears, go to the Ethereum settings through the “Help” tab and click on ethereum/tree/master/releases. A list of programs will appear, which is constantly updated. You need to download the archive located at the very end.
5. Transfer the data from the downloaded file to drive C.
6. Let's start mining Ethereum. Pre-call command line and set the path to the repository and the name in the console that appears. Press Enter. Next, enter this: ethminer -G -S coinotron.com:3344 -O. Press Enter again.

For Ethereum mining on the Radeon HD 7970 video card, it is convenient to use Ethminer. There are several programs for mining this cryptocurrency, but this one is the simplest. Especially suitable for inexperienced users. Experienced users, of course, can connect through other pools.

Is a single-chip model better than a two-chip counterpart?

It was decided to test the novelty and compare it with other models. The results are impressive. The 7970 video card turned out to be the most productive to date. Despite the fact that it is single-chip.

In terms of overclocking, the new Radeon is the fastest graphics card. Outstripped the overclocked dual-chip counterparts! From the standard 925 MHz, the GPU overclocked to 1160 without increasing the voltage. This is a very good indicator. Recently, video cards have been overclocking badly.

There are no shortcomings in this model. Well, unless the price is high. But it is fully compensated by good overclocking potential and high performance. As for the cooling system, when manually increasing the speed to 70% and beyond, a lot of noise rises. But if you do not touch the video card, allowing it to function in standard mode, it works almost silently.

Appearance

Video card Radeon HD 7970, with which Ethereum mining can bring good profit, practically does not differ from its predecessor 6970. The dimensions are almost the same: length - 26.25 cm, width - 11.25 cm. However, there are differences.

The front side of the video card is covered with glossy plastic with a scarlet insert in the middle. Stylish design looks more refined than the previous counterpart. There is no heat-dissipating plate on the back side, unlike the Radeon 6970. There are also changes in the cooling system. The fan comes with wider blades than the previous model and a massive impeller. Other changes: one DVI port remained on the back plate. The hot air will now exit through the expanded grille.

Summing up

The new graphics card from AMD is different:

• high performance
• Low heat
• Low energy consumption
• Innovative proprietary technologies
• Large RAM

All this makes HD 7970 the best single processor solution. The new product surpasses the dual-processor HD 6970 in many respects. When choosing among these models, it is better to give preference to the HD 7970.

If you have enough money and great overclocking potential, get AMD's new Radeon HD. This is the most productive uniprocessor video adapter. This model has proven itself in cryptocurrency mining and is still used today.

High-End video cards are no longer surprising, because manufacturers have realized that for most fans of playing resource-intensive computer games of the new generation, the power of the video adapter comes first. That's just the price of the device can stop potential buyer. To attract attention, intermediate modifications of devices are created that are aimed at top segment, but their price is much lower than analogues.

In this article, we will talk about one of these gaming video devices - the HD 7970 video card from famous brand who built their solutions on the AMD chip. The primary task of the manufacturer was to reduce the cost of the video adapter while maintaining decent gaming power.

Worthy competitor

When it comes to affordable Nvidia products, it turns out that the manufacturer simply offers cut-down versions of its flagship Geforce Titan, you can verify this by comparing the chips of the devices and studying their architecture. AMD could offer cheaper modifications of the Radeon R9 chip in the same way, but they did not do this. The HD 7970 video card received a completely new platform using all modern technologies, thanks to which it outperformed all devices offered by a competitor in the same price niche (15-20 thousand rubles).

The fact that the result is a decent gaming device at an affordable price can be judged by the abundance of devices built on this chipset. All the well-known ones have taken up the production of products based on the AMD Radeon HD 7970. On the market, the buyer can even find several modifications from one manufacturer - this is a worthy indicator for many users.

In original packaging

An affordable price on the video card market is provided by 28 nm technology with the GCN architecture, which has only one graphics core (all top-end devices have two). Thanks to modern technologies, the manufacturer managed to install about 4.3 billion transistors on the HD 7970 chip.

The graphics core operates at a frequency of 925 megahertz, but as the owners assure in their reviews, this is the minimum figure that can be easily increased, it would be decent cooling. 128 texture units and 32 ROPs installed. The memory operates at a frequency of 1375 MHz (effective 5500), has a capacity of 3 gigabytes and communicates with the computer via a 384-bit bus. By the way, the board has a pinout for installing additional memory modules, so we should expect new products from manufacturers with 6 gigabytes on board (possibly with a 512-bit bus).

The high performance of the video adapter led to an increase in power consumption - 250 watts at maximum load is very reminiscent of Nvidia's competitor products. It's rather strange to see AMD products with such resource-intensive requests (previously the limits were up to 130W).

overclocked version

Following the trends in the video adapter market, the manufacturer decided to create an improved modification of their device for demanding users (Nvidia has long used such a scheme with the creation of “Ti” modifications). The video card HD 7970 GHz Edition went on sale in an overclocked form: the graphics core operates at a frequency of 1050 MHz, and the video memory has 1500 MHz (6000 MHz). It is worth noting that the manufacturer has not made any other changes: the expected increase in bandwidth over 384 bits and an increase in memory to 6 gigabytes did not happen. Only the price has increased. As experts note in their reviews, the purchase will clearly turn out to be irrational, because the prefix “Ti” is unambiguously assigned to devices with increased memory bus bandwidth in competitor products.

Appearance and cooling

Still, the moment has come when for installation gaming device AMD needs a roomy Tower ATX chassis. After all, the length of the adapter is 38 centimeters, and it occupies two slots in the system unit, clearly claiming the third for a decent air intake for cooling. Elements are arranged compactly: around graphics core memory chips are installed, and power elements are not scattered around the entire perimeter of the device, but neatly soldered into one block.

The video card based on the AMD Radeon HD 7970 chip has a nice-looking cooling system. The aluminum section of the heatsink is solid and covers two-thirds of the PCB; its base is made of copper, which ensures better heat dissipation from the heating elements. A branded turbofan is installed in the tail of the video adapter. The entire cooling system is tightly closed with a protective casing, which does not allow heated air to return to the device case, but is brought out through the grill

Inexpensive solution

The cheapest on Russian market HD 7970 video card, reviews of which are only positive, is represented by the well-known Taiwanese brand ASUS. For the first time in recent years, the manufacturer did not install its own cooling system, trusting the manufacturer. Generally, ASUS greatly surprised its fans by providing the device with a decent package: all kinds of adapters for monitors, a power splitter, a CrossFireX bridge for connecting several video cards.

Externally, the ASUS video adapter differs from the initial AMD product only in the inscription on the radiator casing, so all the user's attention is riveted to the technical characteristics of the device. Supplied in a kit ASUS utility GPU Tweak put everything in its place: excellent overclocking, demonstrating the high performance of the video adapter, is the main advantage of the manufacturer. The graphics core shows stable operation at a frequency of 1200 MHz, and the memory module practically does not heat up at a frequency of 1635 MHz (6500 MHz).

ice freshness

The HIS 7970 video adapter, as always, attracts buyers to its products with an exclusive appearance and a decent cooling system. The noise level declared by the manufacturer at 15 dB is clearly not marketing ploy, because the company's technologists completely modified the power supply system of the video card by installing a PWM controller of their own production and increasing the number of phases.

The aluminum heatsink covers the entire surface of the PCB and is pierced with copper tubes, which eventually converge on the core adjacent to the graphics core. A metal protective casing with two huge fans is installed on top, at low speeds they very effectively create a powerful air flow that can quickly remove heat from heating elements.

The HD 7970 is factory clocked at 1050 MHz core and memory runs at 1400 MHz, however, according to users in their reviews, these figures are nothing compared to the overclocking potential: 1125 MHz core and 1575 MHz memory provide excellent performance in games.

Typical model of a well-known manufacturer

But the Sapphire company disappointed its fans, because the manufacturer simply took the AMD HD 7970 product and attached its sticker to the casing of the device. Naturally, no additional modifications were made to the printed circuit board. The device attracts attention on the market only for its price (15,000 rubles), which is affordable for many fans of dynamic games.

Standard core overclocking up to 1200 MHz is no surprise, but the memory shows stability at a frequency of 1785 MHz (effective 7140 MHz). Pretty good performance in test applications and games overshadowed by the work of the cooling system. Apparently, the strong heating of the core and memory chips causes the turbofan to operate at maximum power.

Urgent rehabilitation

Apparently, realizing their mistakes, Sapphire decided to urgently rehabilitate itself in the video adapter market. Marketers presented two products at once: Radeon HD 7970 Vapor-X and Toxic Edition video cards, which went on sale in two variations each: with three and six gigabytes of RAM on board. It is quite another matter - it is immediately clear that a serious brand is in front of buyers.

The entire power system was modified: 8 + 2 phases were installed, which significantly reduced the power consumption of the video adapter. It is worth noting that the manufacturer installed branded power controllers and changed all the capacities of the printed circuit board to solid state drives. Copper was not spared in production: the base of the radiator, sections and pipes for heat removal are made of this noble metal.

Decent cooling is the key to the success of any overclocking. The graphics core shows stable operation at 1250 MHz, and the memory works fine at 1600 MHz. There is potential for further overclocking, but noise immediately appears in the operation of two small fans.

silent monster

PowerColor is behaving rather strangely in the video adapter market. She sells one series of devices in the factory version, ignores the other altogether, and endows the third with superpowers. The HD 7970 video card, which can be found in literally all computer magazines, deserves special attention of buyers, because we are talking about water cooling. It should be noted right away that the water block is not free, and the buyer will have to pay about 25,000 rubles for the assembled device (almost the same is the flagship on an AMD chip).

As experts note in their reviews, the water cooling system supplied with the video card strongly resembles a well-known Zalman brand water block, and the principle of operation is identical. Accordingly, questions about the efficiency of the system should not arise, as with the assembly of the cooling unit. There is an instruction, a set of accessories and even a key.

Naturally, such a monster can be easily overclocked, because the efficiency of water cooling is much higher than air blowing. The graphics core runs stably at 1290 MHz, and the video memory can handle 1750 MHz (7000 effective).

Double punch

Fans were also pleased by XFX, which presented two modifications of the HD 7970 chipset on the market at once. The review of an inexpensive device should not even begin, because taking a factory video card, the manufacturer made a proprietary engraving on the casing of the device and brought the product to the market. Low price without any modifications - not at all like a serious brand.

But the second XFX device will be of interest to fans of overclocking, because the manufacturer allowed himself to completely replace the power supply system on the video adapter printed circuit board. Branded controllers of the power phase control unit, microcircuits of our own production and many other changes have significantly reduced the power consumption of the video card. With the cooling system, they did not reinvent the wheel, supplying the device with a cooler, which is installed on the top R9 accelerator.

All these innovations have resulted in effective work video adapter in overclocking at a graphics core frequency of 1200 MHz. The memory shows stability at 1700 MHz. Excellent performance for a device whose temperature is kept within 70 degrees Celsius.

In the race for performance

MSI has launched its HD 7970 Lightning product on the market. The main feature of this video card is a proprietary cooling system. The manufacturer did not begin to refine the printed circuit board, trusting completely AMD technologists, but spent all his time creating an efficient heat removal system.

The aluminum base covers the PCB case on both sides - a rather strange solution, because on reverse side The device has no power elements. An aluminum radiator with copper tubes is installed around the entire perimeter of the video card, which come down to the contact pad of the graphics core. In the places where the fans are installed, the aluminum grilles are slightly cut off in height, creating small recesses for installing coolers (it vaguely resembles Gainward Phantom). As a result of overclocking, you can achieve the stability of the graphics chip at a frequency of 1190 MHz, and the memory - 1650 MHz.

Finally

There are many products based on the HD 7970 core on the Russian market, but their overclocking efficiency is so low that it makes no sense to describe their merits. The review presents devices that will be of interest to buyers who, when buying, are guided by either low cost or high potential for the Golden mean has recently gone out of fashion.

The first graphics adapter with a 28nm GPU. lump high technology, a fusion of extreme performance, exemplary functionality and exemplary economy. Or simply put, the world's fastest single-chip video card from AMD - Radeon HD 7970.

We have been waiting for something extraordinary, bright and fast from AMD. A little over a year ago, when the Radeon HD 6900 line of video cards was introduced, they had no chance to compete with the GeForce GTX 580. Then the company deliberately gave leadership in the top segment, focusing on the fact that solutions based on Cayman chips are better than competitor, price/performance ratio and more energy efficient. But, like any serious player in the market, the company can long time very successful in offering decent devices at a good price. But somewhere in the back of her mind… she certainly wants to be the technological leader in her segment.

Radeon HD 7970 gives her the opportunity. The novelty is based on a chip with a fundamentally new architecture, which is produced according to the most advanced 28-nanometer process technology. Does such a potentially very attractive combination guarantee success?

Graphics Core Next architecture

For a long time, AMD used the VLIW (Very Long Instruction Words) architecture for its graphics solutions, which is well optimized for 3D, but inefficient for general-purpose computing. In such tasks, the rather high theoretical power of the GPU did not give the expected results. The new architecture, called Graphics Core Next, was developed primarily to improve the efficiency of non-3D computing.

The building blocks of GCN are the so-called Compute Unit (CU), which are based on vector blocks, a scalar coprocessor and an independent scheduler.

This model of organization is more universal, easier to program, debug and analyze. But the main advantage of GCN is the good parallelism of calculations in a multitasking environment and the ability to dynamically distribute the load. For this reason, AMD emphasizes that the presented GCN architecture solutions are more than just powerful gaming graphics cards.

In the maximum configuration, the chip has 32 Compute Units (2048 stream processors in total), six 64-bit memory controllers, and an advanced two-level caching system. The new core still uses two geometry processing units, which have been significantly upgraded.

Functional nuances

In addition to low-level architectural changes, AMD's new products will have a whole range of different innovations that improve their functionality.

DirectX 11.1

First of all, we note support for DirectX 11.1, the capabilities of which will be implemented in Windows 8. After the release of this operating system from Microsoft, the new version of the API promises many interesting features. A preliminary list of DirectX 11.1 features can be found here.

AMD ZeroCore Power

The PowerTune function, which appeared in the HD 6900 series adapters, allows you to optimally use the resources of the video card, squeezing out the maximum performance within the specified thermal package. GPUs with GCN architecture have at their disposal a new tool for optimizing power consumption - AMD ZeroCore Power.

This function, inherited from mobile version Radeon, allows you to put the video card into a deep sleep state, in which its power consumption does not exceed 3 W, and the cooling fan stops. This happens when the monitor goes into stand-by mode (this period is changed in the OS settings). The technology is implemented at the hardware level and does not depend on the operating system used.

Such an accelerator operation algorithm is so obvious that the only question is why it is implemented in hardware only now. One way or another, AMD's desire to maximize the efficiency of their solutions can only be welcomed.

Curiously, in CrossFire mode, all additional adapters that are not currently used are automatically set to the ZeroCore Power state and do not emit noise.

Video Codec Engine

For video decoding, a block is used that is functionally identical to UVD3, which is implemented in Cayman chips. This is the most functional solution on the market, so this fact does not cause any complaints. At the same time, the GCN architecture assumes the presence of a Video Codec Engine hardware unit, which is designed to encode video in the H.264 format. To engage this development, requires appropriate software, which is currently under development.

In general, the ability to encode video using GPUs has existed for a long time, but earlier stream processors were used for this. In addition, the customization options, as well as the final picture quality, did not always suit demanding users. Now AMD promises to provide the possibility of flexible configuration of parameters, speed, and encoding method (a hybrid version is possible with the use of stream computers).

Partially Resident Textures

The innovative Partially Resident Textures feature allows you to process huge textures (up to 32 TB) using the adapter's local memory as a kind of cache buffer. The block principle of working with virtual textures and a special algorithm for using mip-levels make it possible to compensate for delays when loading the parts necessary for rendering.

This function can be useful in cases where MegaTexture technology is used. illustrative example is Rage by id Software. In the future, the number of projects using this visualization principle will only increase.

Especially for perfectionists, the anisotropic filtering algorithm has been improved. The new method eliminates the occurrence of small artifacts in rare situations that occur when processing certain types of textures. The function is activated automatically and does not affect the overall performance.

HDMI 1.4a (3GHz)

Another innovation Radeon HD 7970 is an speed version HDMI 1.4a (3 GHz) with support for Frame Packing, which allows you to get a stereo image on a suitable screen not only for watching 3D video, but also for playing games in 1080p mode with a refresh rate of 60 Hz for each eye.

At the same time, the maximum resolution when connected via HDMI is 4096 × 3112. Thus, the adapter is fully prepared for the coming era of even higher quality Quad HD/4k video.

Eyefinity 2.0

Recently, several important innovations have been received by Eyefinity's proprietary multi-display technology. The second revision greatly increases the ability to create multi-monitor configurations. In particular, 5×1 modes with landscape or portrait screen orientations are now available, and the maximum supported resolution has increased to 16384×16384.

Adapters with 28nm chips also received Discrete Digital Multi-Point Audio (DDMA) technology, which allows to allocate several independent audio streams for various displays working within Eyefinity.

With the advent of Catalyst 11.12, stereoscopic HD3D technology on three monitors became possible. Driver 12.2 is expected to be released in February to further enhance the multi-display configuration options.

PCI Express 3.0

Video cards with GCN were the first to receive support for PCI Express 3.0. Thanks to an improved transmission coding algorithm, the new generation bus's throughput is twice that of version 2.0. In particular, for the x16 mode, this is 32 GB / s.

Recall that in the case of Intel platform for the full operation of this high-speed interface, the appropriate support must have CPU, which actually contains the PCI-E 3.0 controller, motherboard with special switching "gates" that allow you to get the required speed, and a graphics adapter.

As for processors, formally the first CPUs to receive PCI-E 3.0 support should be chips designed for the LGA1155 platform. Ivy Bridge, which are expected to be announced in early April. However, in reality, there are already solutions on the market with a new standard bus controller. We are talking about processors for LGA2011. Officially, PCI-E 3.0 support has not been announced for them, but in fact, in the presence of other required components of the ecosystem, the bus operates in this mode.

Curiously, AMD is in no hurry to implement PCI Express 3.0 support for its desktop platform. At the moment chipsets for AM3+ and FM1 are content with version 2.0 bus. To some extent, this pragmatic approach is justified. The capabilities of the current version of the interface are quite enough for the needs of even the most productive video cards. Moreover, in configurations with multiple adapters, when slots operate in x8 (PCI-E 2.0) mode, there is no performance degradation in gaming applications, so bus bandwidth is not a platform bottleneck. On the other hand, in addition to a purely marketing effect (and, admittedly, a considerable one), the practical benefits of PCI Express 3.0 may be obtained for GPGPU.

Southern Islands processor series

When forming a product line based on solutions with GCN, three types of graphics chips belonging to the Southern Islands family will be used.

Tahiti is a solution for maximum performance adapters that will have to satisfy the most demanding enthusiasts. For models of the Radeon HD 7900 line, this core is used. Pitcairn- the basis of the adapters that will be offered for quite demanding players. These GPUs are used for Radeon HD 7800 series graphics cards. Cape Verde– a chip for solutions with good energy efficiency and sufficient performance. These should be the models of the Radeon HD 7700 line.

Here is a slide that will help you understand the positioning of future Radeon HD 7000 series adapters.

On the initial stage 28nm chips with GCN architecture will be used only for high-end and mid-range graphics cards. In particular, for Radeon HD 7900/7800/7700 series devices. The lower performing graphics cards in the HD 7000 line will be virtually identical to those currently included in the HD 6500/6600 and HD 6450/6350 series. It is obvious that the contract manufacturer of crystals, TSMC, is currently unable to provide AMD with such a number of 28 nm chips that they can also be used for budget solutions. In such conditions, more complex, but also more profitable devices take precedence.

Radeon HD 7970

The first of the family of video cards with a new architecture was introduced Radeon HD 7970. Let's look at its technical characteristics and compare it with the older single-chip model of the previous line - Radeon HD 6970.

The graphics adapter is based on the Tahiti chip, which, as already mentioned, is manufactured using 28nm technology. The new die contains 4.31 billion transistors, which is 63% more than its predecessor, while its area is even slightly smaller than that of the Cayman (365 vs. 389 mm2). The clock frequency of the GPU increased slightly - up to 925 MHz. AMD decided not to force things, breaking the milestone of 1 GHz. In the case of the Radeon HD 7970, the Tahiti version is used with everything active blocks, including 2048 stream processors. Computing performance has increased by 40% and is 3.79 TFLOPS. The number of texture units increased from 96 to 128, which, taking into account the increased core frequency, made it possible to obtain a filter rate of 118.4 GT/s. The number of ROPs has not changed, so the pixel fill rate is almost the same.

Like the predecessor Radeon HD 7970 uses GDDR5 memory running at 1375 MHz (5500 MHz). However, now data transfer is performed over a 384-bit bus, due to which the throughput has been increased from 176 to 264 GB / s. At the same time, the amount of local RAM for the top solution is 3 GB. For a traditional single-monitor configuration, this capacity is clearly excessive, even if we talk about a 27- or 30-inch panel with a resolution of 2560x1600. However, in configurations with multiple displays, the frame buffer is significantly increased, so this amount will not be superfluous.

Having become acquainted with technical specifications we can safely say that the Radeon HD 7970 will be noticeably more productive than its predecessor with the Cayman core. The novelty has been improved in all parameters that affect the performance of the GPU. With a constant maximum TDP of 250 W, the developers managed to squeeze the maximum.

Reference video card

Outwardly, the novelty only partly resembles the top model of the previous series. The strict rectangular shape of the outer casing of the cooling system has given way to a design with rounded corners, bright red inserts and a glossy finish on the elements of the top protective cover.

The PCB is 270mm long, however the shape of the cooling system increases the device to 280mm. Middle-tower cases, even in the middle price category, as a rule, provide a hassle-free installation of adapters of similar dimensions, in the case of more expensive ones. game models There will definitely be no problems with the placement of the Radeon HD 7970.

This time, AMD did not install a heatsink plate on the back of the PCB. Heating elements that require additional cooling are located on the PCB front panel, so such a decorative element would only worsen heat dissipation. Although, it must be admitted that the anodized plate adds points to the external design, but this time practicality won out.

On the upper edge of the video card there is a pair of connectors for connecting additional power: six and eight-pin. Appropriate adapters from Molex connectors must be supplied with the video card.

The use of two BIOS chips has become a good tradition for older models of adapters from AMD, and in this case the company did not abandon the Dual BIOS function. A large part of the owners of such solutions are potential experimenters who want to play around with the chip / memory frequencies, including using various firmware versions for this. To protect users from possible problems on this side, AMD offers two BIOS versions, which you can switch between using a miniature toggle switch located on the top edge of the printed circuit board.

Thickness Radeon HD 7970 occupies two expansion slots. There are four connectors on the interface panel: a pair of Mini-DisplayPort, HDMI and DVI (Dual-Link). It is noteworthy that all connectors are placed in one row, while most of the mounting plate is occupied by the exhaust grille of the cooling system. Recall that Radeon HD 6900 have one more DVI port located on the second tier, which significantly reduces the area of ​​the grille, thereby slowing down the output of heated air to the outside.

The overall design of the cooling system is in many ways similar to that used for reference video cards based on Cayman chips. A large evaporation chamber is attached to a massive aluminum base, on which a cassette with a set of aluminum plates is installed. This unit is blown by a 70mm centrifugal fan. The memory chips, as well as power elements, are in contact with the heatsink plate through heat-conducting pads, while the GPU communicates directly with the heat sink of the evaporation chamber. The top of the cooler is covered with a protective plastic cover. This concept of CO has already proved its effectiveness, so it is quite logical that AMD preferred this cooler for the Radeon HD 7970. An important advantage such a system is that almost all of the heated air is blown out of the system unit. This is important nuance, especially when we are talking about productive solutions with high TDP.

Now a few words about the impressions of the video card. In rest mode, the graphics adapter operates almost silently, reducing the chip frequency to 300 MHz, and the memory to 600 MHz. The fan rotates at a speed of 1000-1100 rpm (20% of the maximum value). On an open bench, this was enough to keep the GPU temperature from rising above 35 C.

Under load, the fan comes to life, smoothly increasing speed. The algorithm is linear, so there are no sharp jerks or drops in the process of change. After a long warm-up, the temperature of the chip increased to 77 degrees, the propeller rotated at 43% of the maximum (~2300 rpm). In this mode, the video card is certainly audible, but we are not talking about serious discomfort.

We've also seen AMD's ZeroCore Power technology work in practice. After 20-30 seconds after the monitor goes into standby mode, the video card cooler fan actually stops. To cool the adapter in this state, it is enough for the CO to work in the passive mode.

Generally, regular system cooling Radeon HD 7970 demonstrates decent efficiency with a fairly moderate sound under load. For this reason, there is no regret that at the initial stage the vast majority of models from various manufacturers graphics cards will be based on reference printed circuit boards and coolers, differing from each other in fact only by stickers on the top cover.

Of course, companies are working on devices with alternative COs. In particular, Gigabyte and XFX have already prepared their options.

Similar steps should be expected from other significant players in this market. Another question is how effective such solutions will be. Designs with overall low-speed axial-type fans look advantageous on open stands, but require much more attention to the organization of cooling inside the cases, since in such cases hot air is not actually removed from the system.

Overclocking

For those who have already joined Radeon HD 797 0 to liquid nitrogen, Tahiti opened up in all its glory, surprising with stable operation at 1500-1600 MHz even in CrossFire configurations. Overclockers who have managed to acquire several Radeon HD 7970s, to the envy of their less fortunate colleagues, are actively setting new world records in classic applications from Futuremark.

Unfortunately, at the time of testing, the available overclocking tools did not allow us to change the supply voltage, as well as increase the GPU frequency above 1125 MHz - the limit set in the driver. Reach specified value I didn't have any problems even with stock voltage. +200 MHz is already a very good increase compared to the recommended value, but this is certainly not the limit for the new chip. The question of the possibility of overclocking remains open, and we will definitely return to it. It will be especially interesting to evaluate the potential of the final Radeon HD 7970, which will be available for retail sale. Nevertheless, an engineering sample of a video card has been in our hands. However, we can already say that 1 GHz for Tahiti is only the beginning. Considering the ease with which such frequencies are given to a new crystal, video card manufacturers in the future will most likely use more impressive starting values ​​for the graphics core for their turbocharged modifications.

As for the memory, the considered video card uses 2-gigabit Hynix H5GQ2H24MFR-ROC chips with a base frequency of 6 GHz. Since the standard value for the Radeon HD 7970 is 5500 MHz, the microcircuit initially has a certain margin for overclocking.

The 28nm process works wonders. Despite the significantly increased chip complexity, as well as the increased number of functional blocks and transistors, Tahiti operates at very high clock speeds. In this regard, their 40-nm ancestors "blown away" much earlier.

Test stand configuration

Processor: Intel Core i7-3930K

Motherboard: MSI X79A-GD65 (8D)

RAM: GeIL EVO TWO DDR3-2133 16GB (GET316GB2133C11QC)

Drive: HDD WD WD1001FALS (1TB, 7200 rpm)

Power supply: be Quiet! Dark PowerPro 1000W

Test results

In this case, it was quite easy to select competitors for AMD's new product. First of all, this is the previous flagship - Radeon HD 6970. Even before the start of practical tests, it was obvious that Radeon HD 7970 will be faster than its predecessor, the only question was how significant the difference in performance of these solutions will be in real conditions. But the GeForce GTX 580 is a much more interesting opponent for the new top-end AMD solution. For more than a year, this video card has been the owner of the heavyweight championship belt. And the rivalry with this adapter is of the greatest interest.

The first results of traditional synthetics from Futuremark demonstrate the bid for the leadership of the Radeon HD 7970. 3D Mark Vantage The new product has a 20% advantage over the GeForce GTX 580 in the standard Performance mode and almost twice as much advantage in the Extreme mode - very worthy results.

A similar picture can be observed in a more technologically advanced 3D Mark 11. With the only difference that the advantage in the mode with high resolution and graphics quality is a more modest 28%.

But, this is not yet a victory. Products from AMD have always shown themselves very well in synthetic applications - computing power allows, and possible software optimization for benchmark tests should not be discounted. However, the data obtained give some cause for optimism.

In the test Heaven Benchmark 2.5 the mode with the maximum tessellation level was used. These are very uncomfortable conditions for previous generation AMD chips. AT similar situations solutions from NVIDIA have a serious advantage. Actually, this can be seen by looking at the corresponding diagram. The GeForce GTX 580 is 37% ahead of the top-end Radeon HD 6970 until recently. However, in the case of the Radeon HD 7970, the situation is reversed. The Tahiti adapter turned out to be more agile and faster than the flagship Californians, counting millions of polygons faster than its predecessor by an impressive 67%. AMD has kept its word on tessellation processing speed. The corresponding blocks of the new generation really work very quickly. And although a 4-fold advantage was not obtained in this case, an increase of 1.7 times turned out to be quite real.

In Game Metro 2033 We see a confident advantage of the Radeon HD 7970 over the main opponent from NVIDIA. The performance difference is a convincing 21-26%. Let's pay attention to the fact that the Radeon HD 6970 also looks very good here, losing only 2-6% to the more expensive GeForce GTX 580. However, it is worth considering that during the test, the PhysX option was disabled in the game settings. Its activation adds visual effects, but puts the participants in unequal conditions. In the case of AMD solutions, the calculation of particle physics will fall on the shoulders of the CPU, while with the NVIDIA adapter it will be calculated by GPU computing units using the CUDA API.

A game Far Cry 2 in DirectX 10 mode it also runs faster on the Radeon HD 7970. It is unlikely that anyone is able to feel the difference in rendering speed between 153 and 142 frames / s, but the 8% advantage is still an undeniable fact. And with increasing resolution, it increases to a more significant 22%. In comparison with its predecessor, the decision on Tahiti is 35% faster.

A similar alignment of forces is observed in Colin McRae. The GeForce GTX 580 is 13-22% behind the Radeon HD 7970. And, as has already become a pattern, AMD's newcomer's lead increases with increasing resolution. Well, the fastest Cayman in this test is satisfied with the last place, lagging behind the leader by 42-47%.

Assessing the performance of the novelty, it should be said about its hidden potential, because Radeon HD 7970 has not just improved, but a fundamentally new architecture. As is often the case in such cases, subsequent software optimization can bring very good results. Therefore, it is quite realistic to count on additional performance of solutions with GCN as drivers improve. Developers will probably now focus on adapters of the Radeon HD 7000 line. Therefore, the next improvements during software updates will be felt first of all by owners of video cards with a new graphics architecture.

At the stage of measuring energy consumption, we have two participants left. The fact is that as a test Radeon HD 6970 was used MSI R6970 Lightning, which, in addition to the GPU used, has nothing to do with the reference video card. Seriously enhanced element base and original printed circuit board do not allow the obtained data to be used for direct comparison. But the GeForce GTX 580 we used the adapter ASUS ENGTX580/2DI/1536MD5, which is based on the reference design and has NVIDIA recommended frequency and electrical characteristics.

Frankly, the values ​​​​of 70 W on the dial of the wattmeter are very impressive. Especially considering that test system is based on LGA2011 platform, which is by no means the most economical one, with a six-core processor, four memory modules and a two-disk 7200 rpm HDD.

AMD has done a good job of keeping the power consumption of the Radeon HD 7970 as low as possible. In idle mode, a system with such a graphics card consumes 18 watts less than a PC with a GeForce GTX 580. When it comes to real gaming load, the difference increases even more - up to 50 -60 W. And this despite the fact that the novelty on Tahiti provides greater speed, and therefore has a better ratio of performance per watt.

Results

AMD has certainly caught on to the heterogeneous computing trend. Moreover, in the current situation, we can say that the company is contributing in every possible way to its development. At least in terms of hardware. It is likely that on a global scale and in the near future, Graphics Core Next will help AMD to assert itself more strongly as a solution provider for a wide variety of tasks and applications: from ultimate gaming systems to HPC computing.

As regards directly Radeon HD 7970, then the video card deserves praise. At the moment it is the most productive single-chip graphics adapter. Without any discounts, reservations and clarifications. The new product has a total advantage over the flagship of the previous generation, providing 30-70% higher performance than the Radeon HD 6970. It also managed to get ahead of its main competitor - the GeForce GTX 580 - although the difference here is already more modest.

However, for those who are willing to spend about $500 on a video card, the choice is not so clear. The price of a GeForce GTX 580 with a reference cooling system and recommended clock speeds has dropped significantly over the past two to three months. This model can already be purchased for $460-470. While the retail price of Radeon HD 7970 3GB in Ukraine is expected to be $600. AMD did not overestimate its device, but rather set the bar with which the Radeon HD 7970 offers approximately the same price / performance ratio as the top Californian solution. The latter, although it turns out to be slower, is also cheaper. Therefore, dumping did not work, although it was very desirable. However, we know that AMD is able to quickly respond to the market situation, skillfully adjusting its pricing policy. Fortunately for users, in the segment discrete graphics there is fierce competition, and there is no clear leader, which often makes it possible to dot the “i” in the price issue using purely market methods.

Radeon HD 7970 is a solid solution with excellent performance, enhanced functionality, technological innovation and good potential. Except for the price, in fact, this adapter has no other weaknesses. One way or another, the leadership jersey deservedly goes to AMD. At least for the next couple of months, after which NVIDIA will also introduce own developments with 28nm GPUs and the new Kepler architecture. So the next graphic party also promises to be very interesting.