GeForce GTX 550 Ti

The new GeForce GTX 550 Ti model, based on the improved GF116 GPU, will not replace the GeForce GTS 450 graphics card in the market. At the beginning of its journey, these two models will coexist in the company's lineup together. The updated GPU is made according to the same 40 nm technological standards as other video chips of this generation, and the improvement of its characteristics became possible due to the debugging of the 40-nanometer process technology and deep processing of the GPU.

The principle of naming a video card has changed slightly again. The updated model no longer just belongs to the new fifth generation, and compared to the GTS 450, not only the Ti suffix has been added. For some reason, NVIDIA decided to move the released solution based on the GF116 chip to a higher GTX class, compared to the previous similar card belonging to the GTS class.

Like previous models of the GeForce GTX 500 series, the announced GTX 550 Ti uses an improved graphics processor, already familiar to us from the previous series, and only slightly modified in order to increase performance and improve energy efficiency.

The modifications to the GF116 are aimed at reducing leakage and improving chip yield at high frequencies compared to the GF106. As a result, the GTX 550 Ti performs better while running at nearly 120 MHz more than the GTS 450 GPU while only consuming 10% more power. The GeForce GTX 550 Ti uses the same multiprocessor configuration as the GTS 450. The GF116 contains one GPC cluster, four SM multiprocessors, and three 64-bit memory controllers interlocked with L2 cache and eight ROPs each. In total, the new GPU includes 192 stream processors, assembled into four multiprocessors of 48 pieces each.

But the GF116 memory subsystem has undergone some changes. Each multiprocessor still has the same 64 kilobytes of on-chip memory, and the GPU as a whole contains 384 KB of unified L2 cache, which serves all load and save data requests. But the GTX 550 Ti already has three active memory controllers, and not two, as it was in the GTS 450, which in total makes up a 192-bit bus.

But even that is not all. In the GF116, not only the frequencies and bus width have changed. As you may remember from the technical specifications of the solution, with a 192-bit memory bus, the GTX 550 Ti somehow contains exactly 1 GB of video memory, although usually in such cases it should be 768 or 1536 MB. If you use the same GDDR5 chips, of course.

So the memory controller of the new GPU supports the use of mixed-density memory chips. Which allows you to dial exactly 1024 megabytes using a 192-bit interface. Four chips of the same density (32M×32) with a total volume of 512 MB are used, and two more chips of twice the density (64M×32), which together also make up 512 MB. And in total it just turns out 1 GB.

The most important change in the GTX 550 Ti, in terms of performance, is a significant increase in GPU and memory clock speeds, as well as an expansion of its bus. Unlike the reference frequency of the GTS 450 chip, equal to 783/1566 MHz, in the GTX 550 Ti it was increased to 900/1800 MHz, which is 15% higher. It's even better with video memory, its frequency is increased from 3608 MHz to 4104 MHz, that is, by 14%, and the bus has also expanded by one and a half times. As a result, the overall throughput increase of the GTX 550 Ti over the GTS 450 is over 70%.

Such an increase in theoretical characteristics causes an increase in practical rendering performance in games, especially in cases where the speed is limited by the video memory bandwidth. Judging by the tests of NVIDIA itself, with typical settings for the GTX 550 Ti (1680x1050 and MSAA 4x), expanding the bus to 192-bit alone gives about 14% on average, and in the case of some games even higher (StarCraft II - 31% , Aliens vs Predator - 23%, Batman: Arkham Asylum - 21%).

The GF116 is a reworked and slightly improved GF106, just like the GTX 550 Ti is an accelerated GTS 450. The upgraded GPU from the upper end of the lower price range compares favorably with its predecessor with significantly higher clock speeds, more ROPs, and noticeably more video memory bandwidth at only slightly increased power consumption.