PCB Design and Specifications

Since all GeForce GTX 295 cards share the same reference PCB design, everything we say about Inno3D card refers to other manufacturers’ versions, too. The updated GTX 295 resembles the GeForce 7900 as it has a large fan placed in the center. The PCB dimensions are the only thing the updated version has in common with the older one.

GeForce GTX 295: new revision

GeForce GTX 295: old revision

The most exciting things are hidden inside, of course. In order to see them, we had to take the card apart and the new GeForce GTX 295 proved to have nothing in common with the old version even from this aspect: it is far easier to disassemble. The external plastic casing embellished with Inno3D stickers is fastened with latches and can be easily removed, revealing the cooling system.

The heatsinks can also be removed by unfastening the screws that connect the cooler’s base with the metallic back-plates. Then you will see the following:

GeForce GTX 295 PCB: new revision

GeForce GTX 295 PCB: old revision

So, the two versions of GeForce GTX 295 are very different indeed. The newer one is much simpler and has one PCB although we had suspected that would be technically impossible. We can’t but applaud Nvidia’s engineers who have managed to fit a couple of huge G200b processors, two sets of GDDR3 chips with 448-bit memory buses, two NVIO2 chips, an nForce 200 PCI Express switch and voltage regulators into such a limited space. The component density of the GeForce GTX 295 is far higher than that of the Radeon HD 4870 X2. There are almost no unpopulated areas on it.

Each GPU is powered by an individual 3-phase regulator based on an ADP3192A controller from Analog Devices. Nvidia advertises an enhanced pulse-width modulation technology that can react to changes in load faster and more flexibly, but its real benefits are unclear. It is also unclear if the GPU voltage can be controlled by software.

The memory sets are powered by single-phase regulators that seem to be controlled by a tiny chip marked as “N12 VGF936”. The GPU and memory voltage regulators are based on Renesas R2J20651NP integrated packs that combine power MOSFETs and drivers, making the power circuit compact.

The additional voltage regulators based on uP6161 and APW712 chips seem to power the other components of the card such as the PCI Express switch and the NVIO2 chips.

External power is provided to the card via two connectors, a 6-pin and an 8-pin one. As before, the 8-pin power cable must be plugged in. Otherwise the card won’t start, reporting a power problem with the LED on its mounting bracket. This protection can still be easily bypassed by closing the two outermost pins in the connector, so the GeForce GTX 295 can be powered by two 6-pin PCIe 1.0 connectors if these PSU outputs have a sufficient load capacity.

For some reason the card is still equipped with two NVIO2 chips. In the older, dual-PCB version of GeForce GTX 295, the second chip was necessary for a dedicated HDMI port but the new version has two DVI ports that can be supported by one NVIO2 chip. We suppose that the two chips may be needed to support multi-monitor configurations in multi-GPU mode. Nvidia used to have problems with that function and has only solved them recently.

The nForce 200 chip is an intelligent PCI Express 2.0 switch that supports 48 PCIe lanes and allows the connected devices to communicate directly in peer-to-peer mode. In fact, the nForce 200 moves the SLI logic from the mainboard to the graphics card so that the GeForce GTX 295 could work in SLI mode even on SLI-incompatible mainboards.

The PCB is equipped with a couple of revision B3 G200b processors. One of them was manufactured on the 23rd week of 2009 and is marked as “G200-401-B3”. The other is 2 weeks younger and is marked as “G200-400-B3”.

GPU-Z 0.3.4 does not recognize the new version of GeForce GTX 295 but reports a number of its GPU parameters correctly. Despite the word “Platinum” in the name of the card, its basic specs are standard: the GPUs are clocked at 576MHz and 1242MHz (the main and shader domains, respectively) and have 240 ALUs, 80 texture processors, and 28 RBEs each. For all their efforts, Nvidia’s engineers were unable to place two 512-bit memory banks on the PCB, so the memory chips are still connected to the GPUs with 448-bit buses, each GPU having only 7 active memory controller sections out of 8.

Each memory bank consists of 14 GDDR3 chips from Hynix (H5RS5223CFR-N0C). Half the chips are located on the reverse side of the PCB. Each chip has a capacity of 512 Mb (16Mb x 32), a voltage of 2.05V, and a rated frequency of 1000 (2000) MHz. The card’s actual memory frequency is 1008 (2016) MHz, which leaves almost no room for overclocking.

Each memory bank is 896MB large, but it would be wrong to say that the card has 1792MB of memory. 3D applications can only use half the total amount due to the specifics of today’s multi-GPU technologies. This should not be a problem since 896MB is more than enough for most modern games. And even if there is any shortage in memory, the developer can increase its amount by installing 1Gb GDDR3 chips.

The new GeForce GTX 295 comes with two DVI-I ports. As opposed to the older version, it does not support HDMI or analog video outputs natively. The developer has a reason for that since HDMI connection can be easily established by means of an appropriate adapter while outdated analog interfaces can hardly be necessary for people who buy graphics cards like GeForce GTX 295 in the year of 2009. With fewer connectors, the “second storey” of the card’s mounting bracket is all vent holes, which improves cooling efficiency. Besides that, the card has onboard S/PDIF and MIO headers. Like the older GeForce GTX 295, the new version has only one MIO connector because the second MIO line is wired on the PCB and connects the two GPUs. Thus, the maximum amount of GPUs in a multi-GPU subsystem is 4.