PCB Design and Specifications
This version of the GeForce GTS 250 uses a nonstandard PCB design that has nothing to do with the PCB developed by Nvidia specifically for its new mainstream product. The PCB of the Palit GeForce GTS 250 1GB resembles the one we described in our report Brothers in Arms: Two Palit GeForce 9800 GTX+ Graphics Cards Reviewed, but there are a number of small discrepancies, especially in the left part of the PCB due to the configuration of the card’s connectors.
Palit’s design has one advantage over the reference GeForce 9800 GTX+. Its PCB is shorter. But it has one drawback in comparison with the reference GeForce GTS 250 because it has two power connectors rather than one. Considering that the power consumption of the G92b chip is about 80 watts, it seems unnecessary for the card to have two 6-pin PCIe 1.0 connectors.
The GPU power circuit is identical to that of one version of Palit GeForce 9800 GTX+. It is a four-phase regulator with three power transistors in each phase controlled by ON Semiconductor NCP3488 drivers. The regulator is based on an NCP5388 controller.
The memory voltage regulator is based on an Anpec APW7068 PWM-controller operating at 300kHz.
The card carries 16 chips of GDDR3 memory (Samsung K4J52324QH-HJ08, 512Mb, 16MB x 32, 2.05V). Half of the chips are located on the reverse side of the PCB and covered by an aluminum heat-spreader. The total amount of memory is 1024 megabytes and it is access across a 256-bit bus.
The suffix of the chips’ marking indicates an access time of 0.83 nanoseconds which means that they are capable of working at a clock rate of 1200 (2400) MHz. According to Nvidia’s specs, the GeForce GTS 250 has a memory frequency of 1100 (2200) MHz, but CPU-Z reports that Palit’s version has a memory frequency of only 1000 (2000) MHz. Therefore we can expect the Palit card to be somewhat slower than the reference sample, even though the latter is equipped with only half the memory, in some tests.
The GPU is marked as G92-428-B1, which differs from the GPU of the Palit GeForce 9800 GTX+ (G92-420-B1), but the revision number is the same, indicating that it is a 55nm G92b chip. The die size indicates the same thing. This GPU was manufactured on the second week of this year, i.e. in early January. Interestingly, the marking itself is made in larger print than on older samples of G92/G92b processors.
Like the memory frequency, the GPU clock rates – 745 and 1848MHz for the main and shader domains, respectively – differ from the reference card’s 738/1826MHz. This may make up for the lower memory frequency. The GPU has a standard configuration with 128 universal shader processors, 64 texture processors and 16 raster back-ends.
The Palit GeForce GTS 250 1GB also has a nonstandard configuration of the interface connectors. Instead of two DVI-I ports with a universal analog video output, it has one DVI-I port, one analog D-Sub connector, and one HDMI. That’s quite an odd solution because there is no particular demand for an individual D-Sub connector. An appropriate connection may be established with a cheap DVI-I-to-D-Sub adapter. With the available connectors, you have to attach a second monitor with DVI interface via the HDMI-to-DVI-I adapter whose reliability raises our apprehensions. We guess two DVI-I and one HDMI would be an optimal combination since HDMI is more required today than D-Sub.
Then, the card has two MIO connectors on board for joining two or three such cards into SLI or Triple-SLI subsystems. The onboard S/PDIF header is necessary for audio-over-HDMI.
Frankly speaking, we are suspicious about Palit’s nonstandard PCB design for GeForce GTS 250. Perhaps it is no costlier to make than Nvidia’s reference design, but its two power connectors instead of one are not handy.