Architecture and features
We cannot tell you anything interesting here, GT200b GPUs do not differ from GT200 ones, except for smaller surface and lower power consumption. GT200 architecture was announced last summer. And if we take into account that it's actually the improved G8x/G9x architecture, it appeared even earlier -- in 2006. The main difference between G92 and G80 was the 65nm process technology, GT200 featured mostly quantitative changes, and GT200b is no different from GT200. You can read about it in our previous articles.
Prior to the rollout there appeared information in the Internet that the dual-GT200b card would consist of two GT200 GPUs with just as many execution units as in a couple of GeForce GTX 260 cards. But NVIDIA decided to equip the GTX 295 with sterling GT200 GPUs (240 ALUs and 80 texture units each). However, they preserved memory configuration from GTX 260 -- 448-bit bus and 896MB of GDDR3 memory per GPU. Just like the GTX 260, each GPU in the GTX 295 uses only seven wide ROPs out of eight available in the chip, which gives us 56 ROPs.
Clock rates of the GPU match those of the single GeForce GTX 260 completely. Frequency of the GPU, TMUs and ROPs equals 576MHz, and stream processors operate at 1242MHz. GDDR3 memory works at 1000(2000) MHz.
Like a previous NVIDIA's dual-GPU solution (GeForce 9800 GX2), the new GeForce GTX 295 uses two PCBs. This design offers better characteristics than others. That is each GPU is mounted on its own PCB. This design has the following advantages: each GPU heats only one PCB; the cooler chills both GPUs simultaneously, unlike GeForce 7950 GX2, which uses two cooling systems. The cooler has been improved since the 9800 GX2 times, it has acquired new characteristics to dissipate almost 1.5 times as much heat as the previous dual-GPU model.
Maximum power consumption of GeForce GTX 295 amounts to 289W, which is similar to that of RADEON HD 4870 X2 (286W). The card requires two power connectors: 6-pin and 8-pin ones. And a power supply unit for a single GeForce GTX 295 should be at least 680W.
A few words about peculiarities of the multi-GPU operation. As in all previous dual-GPU solutions, a special chip is installed between the GPUs to connect them. In this case we have nForce 200 (BR-04) that supports a necessary number of lanes for three PCI-E 2.0 ports. Sixteen PCI-E 2.0 lanes are assigned to each GPU, and as many lanes are used to exchange data between a motherboard and this graphics card.
This PCI-E bridge was installed on the previous dual-GPU product. Besides, nForce 200 was also offered as an alternative SLI solution for motherboards with the Intel X58 chipset.
GeForce GTX 295 operates as a dual-GPU system, but the SLI technology allows to join two such cards. This configuration is called Quad SLI. The modern implementation of Quad SLI uses pure AFR for all four GPUs, when four frames are processed simultaneously. This mode significantly accelerates the frame rate, up to 80-90% per each GPU doubling.
We are preparing a separate article about drawbacks of multi-GPU rendering, where we'll cover all peculiarities of the AFR mode. It has to do with input latencies, which are not reduced that much, when the average FPS grows, low minimal FPS and uneven frame rate. All these trifles keep low profile on a dual-GPU system, but a quad-GPU configuration is more prone to these problems -- the average frame rate grows, but it does not become more comfortable to play.
Support for external interfaces in GeForce GTX 295 does not differ much from what we saw in previous solutions with the same GPU. The new card can only boast of two NVIO2 chips, responsible for external interfaces.
GeForce GTX 295 has two Dual Link DVIs with HDCP support and one HDMI output. These DVIs output video from the first GPU, and the second GPU is responsible for a single HDMI. The latest versions of the drivers support video output to two DVIs in SLI mode. If you want to use all three video outputs simultaneously, you should disable this mode. In order to output audio via HDMI, traditional NVIDIA feature, the board has an SPDIF input, to which you should plug an audio cable.
It goes without saying that GeForce GTX 295 supports NVIDIA PhysX that is actively promoted to games now, e.g. Cryostasis and Mirror's Edge (to be released in future). In fact, GeForce GTX 295 can work in SLI mode, process physics effects and render a frame simultaneously even when one GPU is busy with 3D rendering, while the other is doing PhysX only.
At the end of the theoretical part we remind our readers of the question we asked in the baseline article about RADEON HD 4870 X2. We wrote that AMD mentioned problems in large High-End chips (apparently hinting at GT200 from NVIDIA), one of them being very high power consumption of such GPUs. So why were we astonished at the lack of this advantage in dual-GPU solutions from AMD (for example, HD 4850 X2) versus the single-GPU GTX 280 card from NVIDIA, now that we are still surprised where the imaginary advantage in power consumption has gone? Especially in case of the HD 4870 X2, after the rollout of the dual-GT200b card manufactured by the 55nm process technology. NVIDIA GPUs are 1.5 times as complex, and they do not consume more than the RV770.
That's the end of the theoretical part, we know all details about the GT200 architecture and SLI. And now we proceed to the practical part of the article -- synthetic tests of the new card based on two GT200b GPUs and comparison of its performance with other cards from NVIDIA and the rival from AMD.
Reference cards we receive for tests in most cases have no bundle whatsoever. However, that's what you may expect from a retail version:
- User manual
- Software CD
- DVI-to-VGA adapter
- DVI-to-HMDI adapter
- TV-out adapter
- External power splitters
- HDMI audio cable
Write a comment below. No registration needed!