BFG GeForce 7800GS OC 256Mb AGP: Review and Testing - Cooling System. Basic Components.

The radiator covers both the graphprocessor and a PCIe to AGP bridge. There are thick rubber pads between the radiator and 4 memory chips. Another set of pads is situated between the retention plate and 4 memory chips on the other side of the card. Hence, the retention plate also serves as another radiator to some extent. If to take into consideration that memory chips don't consume much power, this approach is a good idea in general. There are 45 ribs soldered onto the radiator's main plate which is 2 mm thick. The assumption made above is correct absolutely: both the main plate and the ribs are copper. Each rib is 70 mm long, 10 mm tall and ˜0.3 mm thick. No doubt that monolithical radiators are better than those ones soldered from parts, but advanced copper casting is a difficult process and it would make the device expensive unreasonably. The 70 mm fan with 15 wings and blue light (T&T 7070TH12C 0.44A 12V) supports pulse width modulated (PWM) power supply for dynamic speed control, that's why it comes with a 4-pin connector. At the same time, the whole cooling system is compact considerably because it doesn't occupy space of the neighborous PCI slot. Air is blown along the radiator's ribs in the right direction, i. e. inside a computer case.

BFG GeForce 7800GS OC 256Mb AGP (front view with the radiator removed)

(click to enlarge, 121Kb)

The other side of the radiator is flat enough, though it could be polished better. The radiator itself is mounted using four screws.

BFG GeForce 7800GS OC 256Mb AGP (the radiator's back)

(click to enlarge, 27Kb)

Overall, the cooling system produces a good impression. There are no heat pipes or something like that, but it's just well done. At least, it seems to be of a better effect than a respective reference system by NVIDIA which has been present on early revisions of BFG GeForce 7800GS OC.

Basic Components

As it has been mentioned before, the heart of this video card is a NVIDIA G70 graphprocessor of A2 revision operating at 398.3MHz. It consists of a whole lot of 302 mln. transistors and is manufactured using a 110nm technology process. Just to get the idea how many it is consider one of dual-core processors from AMD — Athlon 64 X2 (Toledo). It features 1Mb of S-cache per core and is manufactured under a more advanced 90nm process, but comprises only 233 mln. transistors. Back to G70, it's a known fact that recommended by NVIDIA clock frequency varies between 375MHz for 7800GS cards and 430MHz for 7800GTX boards with 400MHz for 7800GT ones in the middle. However, this G70 is clocked higher by default (˜400MHz), and it allows BFG to add a couple of letters to the card's name — "OC" stands for "OverClocked". That's good, but even the higher speed of the graphprocessor doesn't make it equal to 7800GT because of architectural differences. NVIDIA has designed G70 with 8 vertex pipelines and 6 sets of 4 pixel pipelines each. Every pipeline can process shader instructions of the 3rd generation (SM 3.0), though support for all previous generations is also available. In addition, pixel pipelines are able to process texture data, but not to rasterise. Unlike NV40 and its derivatives used for video cards of GeForce 6800 series, where every of 16 pixel pipelines has been able to rasterise texture data processed, G70 features 16 independent raster operators (ROPs) which communicate with the pixel pipelines through a high-speed switch called crossbar. Those ROPs also handle all tasks related to anti-aliasing. So, what is it all about? Big bosses at NVIDIA have decided that GeForce 7800GTX cards only deserve for fully-featured G70s while 7800GT and 7800GS boards should get used to chips with limited functionality. In other words, G70s for 7800GT have 1 vertex pipeline and 1 set of pixel pipelines disabled, and G70s for 7800GS — 2 vertex pipelines and 2 sets of pixel pipelines disabled. Finally, G70s for 7800GS have lost 8 ROPs what isn't a good thing at all. It needs to be noted that those pipelines and operators are disabled in hardware and there is no way to get them working. Software or BIOS hacks which have been possible with many GeForce 6800 series cards don't stand a chance for GeForce 7800 series cards. It's obvious why those people at NVIDIA have agreed upon these castrations: G70 is a big piece of silicon, so throwing away those cores with at least one defective pipeline would make the rest of chips which enter the market much more expensive, also a new design should be developed and manufactured to fill the void of non-born GeForce 7800GS and 7800GT. What to say, not good at all.

Since G70 has been developed with PCI Express in mind, a PCIe to AGP bridge chip has to be placed on the PCB. That's an old good HSI (High Speed Interconnect), known well from many video cards of GeForce 6600 and 6800 series. It supports AGP 8x with SBA (SideBand Addressing) and FW (Fast Writes).

There are eight 256Mbit Samsung GDDR3 SDRAM 1.4ns (K4J55323QG-BC14 — PDF datasheet, 1069Kb) memory chips, and each one contains 4 banks x 2097152 32-bit words. So, they produce a 256Mb array of video memory featuring a 256-bit data bus. That's very good because these days such a wide data bus is a must have for high-end video cards as opposed to 128-bit data bus for mainstream solutions. Of course, GeForce 7800GT and 7800GTX utilise a 256-bit data bus as well. About the clock speed, NVIDIA recommends 300MHz (1200MHz effective) for 7800GS and 7800GTX boards but 250MHz (1000MHz effective) for 7800GT ones. Our card is factory overclocked, so its memory speed is set at 313.9MHz (1255.5MHz effective). Nevertheless, let's keep in mind that the memory chips are labelled as 1.4ns, what equals roughly to 350MHz (1400MHz effective), and the fastest model mentioned in the datasheet is 1.2ns. So, there is plenty of room for overclocking, but let's leave it until later.