GeForce GTX 460 and GTX 460 SLI - A Beloved Fermi!
Fermi 2.0, or NVIDIA, is back
It’s all hard to begin with, the saying goes, and that’s exponentially true for the NVIDIA GTX 400 series. Despite the fact that the architecture is brutally fast, extremely technologically advanced, it cannot be called successful. There is a simple reason for this. The GF100 chip is not designed for the desktop segment, therefore video cards based on it (GTX 465/470/480) cannot be sold. The GF104 graphics processor, which is placed in the main characters of our test today, has a small task: to serve the goals and needs of the players, it must also meet the aspects of economy and production technology.
Attack on the middle class
From NVIDIA’s point of view, it was extremely important that the new chip be significantly smaller than its predecessor. The GF104, of course, comes off TSMC’s 40-nanometer production lines and hides 1,95 billion transistors. Thanks to its updated microarchitecture, the chip has become 366 square millimeters according to the latest news, which is already a competitive value, although there is still room for improvement compared to the Cypress GPU (334 mm2). In the spirit of optimization, ECC support has been discontinued, double-precision operations are run on only 16 ALU units, and the core is no longer square but rectangular.
The first sign of the architecture upgrade is that the chip has two GPCs (Graphics Processing Cluster) with eight SM (Shader Multiprocessors) and a total of 384 CUDA cores. Each SM array hides 8 texturizers. To improve production yield, an SM has been disabled, allowing the GTX 460 to farm “only” 336 CUDA cores and 56 texturizers, but it’s not out of the question that a full-chip chip-based card will arrive soon. The two geometric units and the seven Polymorph Engines are a drastic setback compared to Fermi, but even so, it is unlikely that triangle setup and tessellation performance would become a limiting factor.
Shader Multiprocessor
The GF104 chip is NVIDIA's black lamb, as superscalar processing has now appeared for the first time. As mentioned above, each SM array contains 48 CUDA cores, which are arranged in three smaller groups. Group 3 is controlled by two schedulers, so the Greens have been using the ILP (instruction level parallelization) procedure, just like AMD has been since the R600. The beauty of the thing is that if a dependency develops between each instruction to be executed, we cannot use the full ALU capacity of the chip. Another change within the SM is the doubling of the number of SFUs (Special Function Units), which can cause a non-negligible increase in performance during interpolation.
GeForce GTX 460 1024MB
GeForce GTX 460 768MB
At this point, we would like to draw our readers ’attention to the fact that the two GeForce GTX 460 models do not only differ in the size of the video memory, because the backend part of the GF104 still holds interest. The chip contains 8 ROP blocks (32 units) to which a 256-bit crossbar memory controller is connected. The smaller rider (GTX 460 MB) has also been shortened here, which has 768 ROPs and therefore a 24-bit memory bus., and the size of its L512 cache has been reduced from 384 KB to XNUMX KB.
The GTX 460's graphics core has a clock speed of 675 MHz, stream processors tick at 1350 MHz, and GDDR5 memories have an effective clock speed of 3600 MHz. The controller, equipped with a PCI Express 2.0 interface, is powered by two 6-pin PCIe power connectors. and 1024 watts for the 160 MB card and 768 watts for the 150 MB version. The chip has significant clock reserves, which a Colorful iGame 460 testifies.
GeForce GTX 460: it's time!
We hope that we have not managed to lose a significant group of readers due to the detailed technological analysis, but we believe that there are some of our readers who are more seriously interested in the inner spiritual world of VGAs and GPUs, we wanted to favor them with the above few with a paragraph. Those who jumped here immediately after the introduction would perhaps take a short summary. The GF104 chip is a gamer GPU to the core, but at the same time it can also "perfectly" meet NVIDIA's needs, i.e. it can be produced and produced relatively cheaply. Analyzing it from the customer's point of view, we can only say good things about the chip, because it is extremely tuning-friendly and has more moderate consumption.
Summary specification table
