Radeon hd 7970 3gb review: 28nm And Graphics Core Next, Together As One

Recall that starting from the Radeon HD 2000 line, ATI (and later AMD) engineers used the VLIW (very long instruction word) architecture to develop Radeon graphics processors. Of course, since the introduction of the R600 chip to the Northern Island family of graphics adapters, the performance of solutions has increased several times, but there have been no fundamentally new changes during this time.

Nevertheless, the quality of graphics in PC games is constantly improving, and developers are finding new ways to increase the realism of the image on the monitor screen. Of course, the requirements for the speed and memory of video cards are constantly growing, but the nature of the tasks that they face is gradually changing. So, for example, recently discrete graphics adapters are increasingly being used in “non-graphical” applications and often the performance of a certain task by a video card requires less time and resources. This is one of the highest priority development trends for graphics chip manufacturers, and AMD and NVIDIA are well aware of this. Note that the developers from the «green camp» realized the need to create a fundamentally new unified architecture and a general-purpose graphics chip (GPGPU or General Purpose GPU) much earlier. As a result, NVIDIA CUDA technology is now supported by a large number of programs, including professional solutions. However, this technology has received significant distribution relatively recently, so AMD still has every chance to compete in this market segment.

So, the basis of the new Tahiti XT chip based on the Graphics Core Next architecture is super-scalar processors called Compute Units, which include 64 ALUs each. It is important to note that they consist of vector and scalar blocks, and also have a built-in scheduler and 16 kilobytes of L1 cache. The applied scheme of their structure allows to obtain higher performance in comparison with the VLIW-architecture, as well as high efficiency, both for solving graphic tasks and general-purpose tasks.

In total, the Tahiti XT chip includes 32 Compute Units (2048 stream processors), which are divided into groups of 4 pieces. Each such group has at its disposal 32 KB of cache memory for storing data and 16 KB for instructions. In addition, as we have already noted, each CU is equipped with its own 16 KB cache. The total amount of cache in the first level (L1) is 512 KB, and the second — 768 KB, which is twice as much as the previous flagship of the Radeon HD line.

The memory bus width was increased to 384 bits, but the number of ROPs did not increase proportionally to 48, but remained the same — 32. At the same time, the principle of their interaction with the rest of the GPU, according to AMD engineers, was significantly redesigned . Note that the new Tahiti XT chip includes 4.31 billion transistors in an area of ​​365 mm².

Particular attention in the new architecture, according to AMD representatives, was given to the tessellation block. Since the days of the Radeon HD 5000 line, in tasks with the active use of tessellation, solutions from NVIDIA have looked much more preferable, however, with the introduction of the tessellation unit of the tenth version in the Radeon HD 7000 line, the situation may change significantly.

Based on graphics from AMD’s official presentation, compared to the previous flagship, Radeon HD 7970’s tessellation processing performance has increased by up to 140% in certain applications.

An important innovation in the new Graphics Core Next architecture is the separate Video Codec Engine block. When software appears on the market that allows you to take full advantage of its capabilities, users will be able to use the processing power of video accelerators not only for decoding, but also for encoding video content.

A lot of attention in the new architecture has traditionally been paid to energy consumption. AMD Power Tune technology is designed to dynamically monitor the load on the GPU and, if the maximum TDP level is exceeded (for example, in FurMark, OCCT stress tests), it automatically reduces the GPU clock speed and its voltage. Thus, the manufacturer does not artificially reduce the performance of the video card, looking only at synthetic tests, which practically do not run in real operation.

Another nice feature of the Graphics Core Next is the ZeroCore feature. Now in idle mode, after the monitor screen automatically turns off, the video card can automatically switch to economy mode, even turning off the cooling fan. The power consumption of the graphics adapter in this mode does not exceed 4 watts, according to AMD engineers.

In CrossFireX mode, when switching to 2D mode, all adapters except one will similarly «sleep» and turn off the CO fans.

A lot of attention in the new architecture was also paid to Eyefinity technology, which, we recall, allows you to connect multiple displays to one video card in different modes at the same time. With the transition to Graphics Core Next, Eyefinity technology has been updated to the second version and now allows you not only to connect up to 6 monitors to one video card (using active adapters, purchased separately), but also to activate 3D mode when connecting multiple displays. Here again, AMD sets a new standard in the industry, because the similar NVIDIA 3D Vision Surround mode requires at least 2 graphics cards in SLI mode.

In addition, AMD Radeon HD 7970 is announced to support monitors and LCD panels of ultra-high resolution (7680 × 4320 pixels), which at the moment is already a good start for the future. Also, it should be noted that the video cards of the Southern Islands family support the PCI Express 3.0 interface. There is still little practical benefit from this innovation, however, the fact that new standards are being introduced is certainly pleasant. Note that the AMD Radeon HD 7970 is the first video card on the market to support PCI Express 3.0. At the end of the story about the architecture, here is a summary table in which we compared the specifications of the new flagship of the Radeon HD line with its predecessors and current competitors:

This list of innovations in the Graphics Core Next architecture looks really interesting and promising. Well, we are moving on to practical testing in order to practically study all the features of AMD’s advanced technologies on the example of the «reference» design Radeon HD 7970 video card.

It’s worth noting that our test lab did not include a retail version of the Radeon HD 7970, and not even the final revision of this video card. Today we are dealing with an engineering sample that probably has a number of minor differences from those graphics cards based on the Radeon HD 7970, which have already appeared in Ukrainian retail chains. Nevertheless, its main characteristics completely coincide with those declared for the AMD Radeon HD 7970, therefore, the performance level should be similar. Until the time when AMD partners present their versions of the Radeon HD 7970 based on a unique design, all video accelerators based on the Radeon HD 7970 will differ only in a sticker on the cooling system casing, delivery kit, warranty service conditions and, of course, cost.

The appearance of the video card is very similar to the design of the previous flagships of the Radeon line, however, the differences between them are clearly visible, and the Radeon HD 7970 has its own «face». In our opinion, the smoothed edges of the casing of the cooling system and the use of glossy plastic look more attractive in comparison with the strict and somewhat «brick» look of the Radeon HD 6970/6950. The length of the video accelerator is 285 mm, which is typical for hi-end solutions. In most modern gaming cases in the Middle- and Full-Tower format, its location will not cause problems.

The following interfaces are provided for displaying an image on the AMD Radeon HD 7970 video card:

Note that the developer refused the second DVI connector, motivating this by the desire to increase the efficiency of the turbine-type cooling system. It is likely that retail versions of the Radeon HD 7970 will be equipped with additional adapters and the end user should not have any problems connecting a monitor, plasma panel or projector to the video card. Recall that thanks to Eyefinity 2.0 technology to graphics accelerators on AMD Radeon HD 7970, you can connect up to 6 monitors with two additional DisplayPort hubs (sold separately). An HDMI 1.4a port with frame packing support delivers a 3D image at 60Hz to each eye.

The printed circuit board of the AMD Radeon HD 7970 video card that came to our laboratory is made on a red textolite, however, cards on a later revision of the black PCB are supplied to retail channels. Of course, it is worth noting the use of high-quality element base: field-effect transistors in a DirectFET metal case, chokes with ferrite cores and solid-state capacitors. For Radeon HD 79 level solution70 high-quality element base will not only ensure long and stable operation in normal mode, but also allow you to achieve better results during overclocking.

The power stabilization module is made according to a 6 + 1-phase scheme, and the CHiL CHL8228G chip is used as a PWM controller.

To power the video card, in addition to the PCI-E connector, one 6-pin and one 8-pin power connectors are used, which are located on the side at the end of the board.

Two CrossFireX bridge connectors allow multiple AMD GPU-based graphics cards to be connected together to calculate visual effects. Next to them is a small BIOS switch, which first appeared in the Radeon HD 6900 line. Here we found only the mounting cross of the cooling system and a few microcircuits of the VRM module. In addition, on the engineering sample that got to us in the test lab, there were several small switches and connectors for an incomprehensible purpose on the back of the printed circuit board. In the final version, obviously, there is nothing like that.

This is what the Tahiti XT graphics chip looks like, made according to the standards of the 28 nm process, which is the central part of the video card in question. As we already mentioned, it includes 2048 unified processors and 32 ROPs, and data exchange between the graphics core and memory is carried out via a 384-bit bus.

GDDR5 video memory with a total volume of 3 GB is assembled using twelve 256 MB chips manufactured by Hynix. The chips are marked H5GQ2h34MFR T2C and, according to the documentation, can operate at an effective frequency of up to 6 GHz. Considering that the memory in the tested video card initially operates at a frequency of 5500 MHz, it makes sense to try to speed it up a bit during the manual overclocking process.

The cooling system of AMD Radeon HD 7970 «reference» design includes a turbine and a rather massive aluminum radiator at the base of which there is an evaporation chamber, which contributes to an even distribution of heat over the entire heat exchanger area.

For effective air intake, there are special holes in the back of the plastic casing that covers the entire front of the printed circuit board.

The graphics chip is in direct contact with the main heatsink, while the memory chips and power transistors are in contact through a special thermal interface with a metal base plate on which the heatsink and turbine are mounted. On the whole, since the time of the Radeon HD 6970, the standard cooler has not undergone significant changes, and in order to test its effectiveness in practice, we conducted our standard set of tests. Traditionally, tests were carried out in two cooler operating modes: automatic control and maximum turbine speed.

In automatic mode, at maximum load, the temperature of the graphics core did not exceed 74 degrees, which is a good result for an adapter of this level. At the same time, the fan worked at 50% of its power, producing an average noise level that was noticeable against the general background of the system, but not annoying.

After we set the fan speed to maximum and tested again, we recorded a decrease in the maximum temperature of the GPU by 12 degrees, and the noise level emitted by the cooler, we would describe as very high.

When there is no load, the graphics core and memory frequencies are automatically lowered, resulting in lower power consumption and heat dissipation. The cooler in this mode works really quietly, and the temperature of the graphics core does not exceed 36 degrees.

Overall, the evaporative chamber cooling system used for the AMD Radeon HD 7970, , showed good cooling performance at an average noise level. Against the background of other components of a modern system unit, the noise from the operation of a video card under a typical load stands out insignificantly.

Test & Review: AMD Radeon HD 7970

AMD has decided to start a new battle for dominance in the GPU market. But today, both manufacturers, AMD and Nvidia, have to face new challenges and work in new conditions. In particular, AMD has to switch to a new 28 nm production technology, and a completely new GPU architecture, as it turned out. NVIDIA also plans to move to 28 nm, but only in a few months, and with a new architecture. But AMD was the first, and in our article we will talk about the new generation of GPUs in the form of AMD Radeon HD 7970.

» Photostrecke

AMD believes that PC gaming is in for a boom, and in the short term — especially given that consoles are updated for quite a long time. And since modern graphics engines benefit from the capabilities of advanced graphics cards, this development will only intensify. The PC games market was worth $15 billion last year and is expected to grow to $20 billion by 2013. And don’t forget that gamers today prefer to play at ever higher resolutions. 1080p resolution has already become the de facto standard, which is supported by rapidly becoming cheaper displays with a large diagonal. In addition, AMD is focusing on higher GPU efficiency and GPU computing capabilities. The latter area is very important for AMD today as the company wants to work around the limitations found in Cayman architecture GPUs.

AMD has only introduced the Radeon HD 7970 so far, as you can see on the slide, but new cards should be coming to the Radeon HD 7900 line soon.

	NVIDIA GeForce GTX 570	NVIDIA GeForce GTX 580	AMD Radeon HD 6950	AMD Radeon HD 6970	AMD Radeon HD 7970
GPU	GF110	GF110	Cayman PRO	Cayman XT	Tahiti XT
Process	40 nm	40 nm	40 nm	40 nm	28 nm
Number of transistors	3 billion	3 billion	2. 6 billion	2.6 billion	4.3 billion
Crystal area	530 mm²	530 mm²	389 mm²	389 mm²	365 mm²
GPU clock speed	732 MHz	772 MHz	800 MHz	880 MHz	925 MHz
Memory clock	950 MHz	1000 MHz	1250 MHz	1375 MHz	1375 MHz
Memory type	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5
Memory capacity	1280 MB	1536 MB	2048 MB	2048 MB	3072 MB
Memory bus width	320 bit	384 bit	256 bit	256 bit	384 bit
Memory bandwidth	152 GB/s	192 GB/s	160 GB/s	176 GB/s	264 GB/s
Shader model	5. 0	5.0	5.0	5.0	5.0
DirectX	11	11	11	11	11.1
Number of stream processors	480 (1D)	512 (1D)	1408 (352 4D)	1536 (384 4D)	2048 (1D)
Stream processor clock speed	1464 MHz	1544 MHz	800 MHz	880 MHz	925 MHz
Number of texture units	60	64	88	96	128
Number ROP	40	48	32	32	32
Maximum power consumption	219 W	244 W	200 W	250 W	250W
Minimum power consumption	—	30-32W	20W	20W	2. 6 W
CrossFire/SLI	SLI	SLI	CrossFireX	CrossFireX	CrossFireX

Radeon HD 79 video card70 is based on the «Tahiti XT» GPU, which is manufactured using a 28nm process. In total, the GPU has 4.3 billion transistors. By comparison, Intel’s Sandy Bridge-E processors (excluding quad-core models) have 2.27 billion transistors. And the predecessor from the Cayman Radeon HD 6900 family worked with 2.6 billion transistors. The crystal area is 365 mm². As you can see, the area is slightly less than 389 mm² for the «Cayman» GPUs, which are manufactured using the 40-nm process technology. NVIDIA’s GF110 GPU contains 3 billion transistors in an area of ​​530 mm². Most of the GPU transistor budget was spent on 2048 stream processors. GPUs and stream processors run at clock speed 925 MHz. AMD decided to keep the same memory as the Radeon HD 6970, i. e. GDDR5 at 1375 MHz. But the memory interface has been expanded from 256 bits to 384 bits, increasing the memory bandwidth to 264 GB/s. In addition, the capacity has increased from 2048 MB to 3072 MB. The Radeon HD 7970 has 128 texture units and 32 raster operations pipelines (ROPs) — we get an increase in texture units compared to the Radeon HD 6970, but the number of ROPs remains the same. AMD has indicated the maximum power consumption for the Radeon HD 7970 at 250W, which is also the limit for PowerTune. Typical graphics card power consumption is 210W. Let’s recall that the Radeon HD 6970 had a maximum power consumption of 250 watts, and a typical one under load — 190 watts. Thanks to ZeroCore Power technology (more on that below), power consumption in idle mode does not exceed three watts.

GPU-Z 0.5.7, as you can see in the screenshot, does not correctly display all AMD Radeon HD 7970 data. On our test system, the Socket 1366 interface was specified as PCI Express 3.