Articles: Graphics
 

Bookmark and Share

(39) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 ]

NVIDIA GeForce 7800 GTX 512: PCB Design

It is not so easy to spot any difference between the PCBs of the GeForce 7800 GTX and the GeForce 7800 GTX 512 because the huge dual-slot cooling system borrowed from the NVIDIA Quadro FX 4500 leaves very little to the eye:

 

Yet there are differences, and quite serious ones. The PCB of the GeForce 7800 GTX 512 is actually a thoroughly modernized PCB of the GeForce 7800 GTX with a redesigned and reinforced power circuit. Some of its components can be seen in the card’s top right corner – they are missing on the ordinary GeForce 7800 GTX as well as on the professional Quadro FX 4500 card.

More differences become evident if you take a look at the reverse sides of the PCBs. The GeForce 7800 GTX has a scattering of small components around its power circuit. The same area on the GeForce 7800 GTX is populated much less densely. Another important feature is that there are no memory chips and no seats for them on the reverse side of the PCB. All memory is located on the face side of NVIDIA’s new graphics card. Since there are no chips, the cooling bar is not necessary, either. So, it is missing here, just like the back-plate of the cooling system is. The cooler is now simply secured with nine spring-loaded screws. This is a rather strange solution, considering the large dimensions of the cooler. We removed it to see the following:

The left parts of the PCBs of the two versions of GeForce 7800 GTX are nearly identical, but the right part of the GeForce 7800 GTX 512 card carries a more complex power circuit. The main power components of the circuit are still covered with a small rectangular heatsink. The ordinary version of the GeForce 7800 GTX permitted to install a total of 512MB of graphics memory by adding another eight 256Mb GDDR3 chips in 144-pin FBGA packaging, but such memory would not have allowed to be overclocked from 600 (1200) MHz to 850 (1700) MHz. The min access time of K4J55323QF series chips is 1.25 nanoseconds which corresponds to 800 (1600) MHz frequency. Moreover, Samsung doesn’t supply such chips in mass quantities, but only offers samples.

NVIDIA took another approach in order to reach the record-breaking memory frequency. The company redesigned the PCB of the new card in such a way as to make possible to use more advanced memory chips from the K4J52324QC series, in 136-pin FBGA packaging and with a double capacity of 512Mb. These memory chips with an access time of 1.1 nanoseconds are really unique. Even the manufacturer’s website has no info about their existence. According to official info, the GDDR3 chip series from Samsung currently stops at 1.25ns products. It is possible Samsung doesn’t yet supply such fast memory in mass quantities and this partially explains the exclusiveness of the NVIDIA GeForce 7800 GTX 512. The memory frequency is reduced a little below the chips’ rating, to 850 (1700) MHz, probably to ensure stable operation of the device.

Samsung produces its K4J52324QC series in two versions, marked as BC and BJ. Here we have the second version that works at 1.8V voltage. This should reduce the total power consumption of the GeForce 7800 GTX 512 a little. Since these chips have 512Mb capacity and 16Mx32 design, eight such chips suffice to yield a total of 512 megabytes of graphics memory accessed across a 256-bit bus. The peak bandwidth of the memory subsystem of NVIDIA’s new card is really impressive, being as high as 54.4GB/s (compare this to the 48GB/s memory bandwidth of the ATI RADEON X1800 XT 512MB). Of course, the efficiency of the memory controller affects the real performance of the card, yet NVIDIA’s new solution enjoys a 6.4GB/s advantage over its opponent in the “pure” memory bandwidth.

As for thermal interface, NVIDIA seems to have found an optimal solution and uses it here. We mean the cloth pads soaked in white thermal paste. They have already proved their efficiency, providing good thermal conductivity and proper contact between the memory chips and the cooler’s heatsink. In this case, the memory must be cooled properly. Clocked at 850 (1700) MHz, it has to consume a lot of power and, as a result, to dissipate a lot of heat, even despite the reduced voltage it works at.

 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 ]

Discussion

Comments currently: 39
Discussion started: 11/15/05 10:07:52 AM
Latest comment: 01/09/07 05:03:13 AM

View comments

Add your Comment