Articles: Graphics

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 ]

PCB Design and Functionality

The new AMD budget graphics accelerator looks very similar to Radeon X1950 XTX, primarily because of the cooling system casing. The card is very compact, it is only 20.7cm long, which means it will fit in almost any system case, including barebone solutions like Antec NSK1380.

Although they used 128-bit memory bus that made the PCB circuitry quite simple, the board is “packed” with electronic components, mostly because they used 8 memory chips and pretty powerful voltage regulator circuitry. Although they use new 40nm production technology to make their new RV740, which should demonstrate pretty modest power consumption, there is a three-phase voltage regulator with three power transistors in each phase.

The heart of this VRM is the L6788A controller from ST Microelectronics that we haven’t seen on graphics cards yet. This chip supports software control over the output voltage with 1.35V being the maximum. Therefore, real overclocking enthusiasts will most likely need to resort to hardware voltmodding. The voltage regulator works at only 200kHz frequency that is why they used regular polar capacitors with polymeric dielectric for the regulator circuitry. Compared with more powerful graphics cards’ VRM working at 1MHz+ frequencies, it is a definite step backwards. However, the use of low-frequency circuitry lowers the production costs, which is very important for a sub-$100 solution.

The memory voltage regulator uses a single-phase circuitry with uP6101 chip from uPI Semiconductor performing as a PWM-controller.

We would like to pay special attention to a very interesting Renesas R2J20602NP chip located beneath CrossFire connectors. It combines MOSFET and corresponding drivers within the same packaging and can work at up to 2MHz frequency returning about 40A current. These chips are generally known as DrMOS (Driver-MOSFET) and are often used in CPU voltage regulator circuitry of advanced mainboards. However, the role of R2J20602NP in the Radeon HD 4770 voltage regulator is still a mystery to us.

The card may have separate voltage regulators for memory VDD and VDDQ, but why would anyone want to use a super-powerful solution capable of handling 40A and theoretically powerful enough for the entire Radeon HD 4770?

We also expected Radeon HD 4770 to require no external power and to receive enough through the corresponding part of the PCI Express x16 slot. However, ATI’s reference design offers at least one 6-pin PCIe 1.0 connector with up to 75W capacity. We are going to find out later if this connector is really needed.

The card is equipped with 8 GDDR5 chips from Qimonda marked as IDGV51-05A1F1C-40X. According to the manufacturer specification, these 512Mbit chips can work at up to 1000 (4000) MHz frequency with 1.5V voltage as 16Mx32 or 32Mx16. Since RV740 uses a 128-bit external memory bus, they use the latter mode.

The current GPU-Z version reports incorrect memory frequency for Radeon HD 4770: 850 (3400) MHz. In fact, the correct frequency is displayed in Default Clock field: 800 (3200) MHz. The graphics memory works at this frequency providing about 51.2GB/s peak bandwidth. It is a remarkably high value for a graphics solution with a 128-bit memory bus. However, now that GDDR5 gets more and more widespread, these numbers may become more and more common. The use of memory chips capable of working at up to 1000 (4000) MHz frequencies suggests that the video memory on this graphics card may be seriously overclocked, however, let’s not jump ahead of our story here.

40nm manufacturing process allowed to significantly reduce the size of RV740 die. You can see right away that RV740 is much smaller than RV630, even though it has 312 million more transistors than RV730 (RV740 has 826 million transistors). However, there is also a similarity: the die inside the package sits at a 45-degree angle and the package itself has no protective frame around it. The GPU marking tells you only the production date: week 11 of 2009 that falls onto March 8-14, 2009.

As we know, Radeon HD 4770 boasts pretty advanced technical characteristics for a sub-$100 solution. It contains 128 super-scalar computational units each with 5 ALU (640 ALU total, twice as many as by RV670) that are arranged into 8 SIMD cores. 4 texture processors are responsible for each SIMD core (32 TMU total). Raster operations are performed by four large Render Back-Ends (RBE) that are equivalent to 16 regular ROPs. This way, the configuration of Radeon HD 4770 functional units makes it equivalent to Radeon HD 4830 with that only difference that the former works at much higher frequency: 750MHz vs. 575MHz. It gives us reasons to expect considerably higher gaming performance from the newcomer. It is a true leap forward compared with Radeon HD 4670.

The interface connectors on the new Radeon HD 4770 are quite standard. They include a pair of dual-channel DVI-I ports supporting resolutions up to 2560x1600 and sound output over HDMI, a seven-pin mini-DIN port delivering analogue video output in S-Video, Composite and YPbPr formats, and two CrossFireX interface connectors. This is the most optimal configuration today because you can easily implement VGA and HDMI connections with the corresponding DVI-I adapters. As for DisplayPort, it is not popular enough yet to justify the support of this standard in gaming graphics accelerators, even though RV740 does have this feature. I have to say that analogue video out support looks a little bit archaic for the year 2009, but it hardly affects the production cost and retail price of the product, so cannot be regarded as a drawback by all means. Especially, since quite a few users out there may actually need it.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 ]


Comments currently: 15
Discussion started: 04/27/09 11:07:55 PM
Latest comment: 05/11/09 02:41:55 AM

View comments

Add your Comment