Review: Great performance spells $$$ for Nvidia
Today we'll show you Nvidia's GTX 295, the company's „new“ single-PCB card which Gainward included in their offer. The card's full name is Gainward GTX 295 1792MB Dual DVI and its name makes it a bit hard to differentiate between this card and the dual-PCB GTX 295.
As you may already know, the original GTX 295 came out as a dual-PCB card, where each PCB housed one GT200b graphics core together with the accompanying components. Communicating via the internal SLI connection, the two GPUs render a picture as two standard Geforce cards would’ve done following predefined modes from within the driver. A good thing about internal SLI however, is that the end user sees the dual-PCB GTX 295 as one piece of hardware, rather than two chained in SLI.
Despite being fast and strong, Nvidia was looking for a way to replace this expensive and unprofitable product with a new, single-PCB based one, which would bring more cash as the production costs drop significantly.
The new GTX 295 single-PCB graphics card has the same specs as the dual-PCB GTX 295 when it comes to GPU power and speed. So we, the end users, shouldn’t notice any performance difference between the single and dual-PCB versions of the GTX 295.
As far as speed goes, the GTX 295’s chips are somewhere between the GTX 285’s and the GTX 260’s, where the Geforce GTX 295 comes with the GTX 260’s frequencies and memory quantity, and the shader processors from the GTX 285.
The GPU is built in 55nm, whereas the clocks (both chips have the same specs) are: 576MHz for the core, 1242MHz for the shaders and 1998MHz for the memory. The card comes with 1792MB of memory, so each chip addresses 896MB of memory. It’s easy to see that the card has a 448-bit memory interface and that one ROP partition has been disabled. Each of the 7 ROP partition is connected to the 64-bit memory controller, totaling to a 448-bit interface. Each framebuffer partition is connected to 128MB of memory, resulting in 896MB of memory per GPU.
The full GT200 chip (found on GTX 280/285) supports the 512bit interface for GDDR3 and come with 8 ROPs and 8 frame buffer partitions. One ROP partition contains 4 ROP units, resulting in total of 28 ROPs. In order to maximize the card’s power, Nvidia used all available shaders on the GTX 295, all 240 of them. Since the GTX 295 is a dual-GPU card, we can multiply all the numbers by two and say that the GTX 295 has 480 shader processors, 56 ROPs and 160 texture filtering units. The card’s bandwidth is 223.8GB/s.
Gainward didn’t alter the reference frequencies much, and you can see a slight difference in memory clocks, as it runs at 1008MHz (2016 effectively).
The following photo shows Gainward’s GTX 295 (single PCB) card.
Here’s how it looks under the hood.
Separate aluminum blocks with a large copper based and heatpipe technology were used to keep core temperatures in check, and it’s quite effective. The fan-driven air courses through aluminum fins and cools the GPUs on both sides by a couple of degrees Celsius lower than on the dual PCB card.
Just like its dual PCB version, our today’s GTX 295 requires an 8 pin and a 6 pin connector, and the card will draw up to 289W in toughest scenarios. It’s important to use a quality PSU with recommended 20A per rail, whereas the high-end systems should run on at least 650~700W PSUs. If by any chance the card is not enough of a performer by anyone’s standards, the ever power-hungry can always opt on Quad SLI. Note that an addition of another GTX 295, whether it’s single or dual PCB version, will require up to 289W more.
The I/O panel features two dual-link DVIs with HDCP support, whereas the HDMI can be done via the DVI-to-HDMI adapter which Gainward included in their package.
The package contents:
- Graphics card
- Documentation + Driver CD with ExpertTool
- DVI Adapter + HDMI Adapter
- PCI-Express power cable
- SPDIF audio cable