Gainward Bliss GTX 275 review
Author: Luka Rakamaric
Date: 14 Oct 2009

For quite some time, the only competition to ATI’s aggressively priced 4800 cards was the NVIDIA 9800 series. The GTX 260 was too expensive, and its performance was just equal to the cheaper 4870. The GTX 280, although with a lot more power, was in the price league of it’s own. Prices kept dropping, so NVIDIA had to react. Their first move was the transition to 55 nm production process. The GTX 295 and 285 were the first models, but they carried an even heftier price tag. What NVIDIA needed was a competitor to the highly successful, GDDR5 carrying 4870 and it’s upgraded version, the 4890 which launched about the same time as GTX 275 did.

NVIDIA GeForce GTX 275 is actually a mix of two cards. The GTX 260 and the GTX 285. From its bigger brother it got the GT200b 55 nm GPU, with 240 cores. To keep it distinguished, and cheaper, from the 285, it inherits the 448 -bit memory bus and the 896 MB of memory from the GTX 260. It still has more memory throughput than 4890 (130:124 GB/s), but less than 285 which is at a much more impressive 158 GB/s. Also, there’s a drop in render back ends compared to the 285 (28:32). Basically, one eight of that part of the GPU was disabled.

Clockwise, the card is much faster than 260, as you only lose 15 MHz for the core, 72 MHz for the shaders and 154 MHz for the memory compared to the GTX 285. The performance difference should be greater, since there’s also the memory bandwidth loss of 30 GB/s. However, since the 130 GB/s throughput is quite enough for lower resolutions, the real difference could only be visible at 19x12 or 25x16 resolutions, which are present on very few displays today. With that in mind, we feel that it could be a much closer call than the price difference would indicate.

The whole GT200 family of GPUs is the latest of what NVIDIA has to offer on the market. Its full version uses 240 parallel processor cores, which are connected to 32 ROP units, and has a 512 bit memory bus that connects to GDDR3 memory. You could ask why only GDDR3 and not GDDR5. The truth is that GT200 cannot use GDDR5, nor would GDDR5 be valid from the price/performance point of view, as the 512 bit bus costs a lot more to make than 256, and then the GDDR5 would just be a further increase with not so great improvements. ATI on the other had uses GDDR5 because the GPUs use a 256 bit memory bus, which is then compensated with the faster memory. A jump from 60 to 120 GB/s has a lot more influence on performance than a jump from 158 to 250 GB/s.

The scalar architecture of GT200 GPUs enables all of the ALUs to be used simultaneously, where ATI uses a superscalar architecture where the ALUs are grouped in small pipelines of five, greatly reducing their effective number in 3D rendering, which is not the best environment for pipelined design. That is why, while ATI’s competing cards have over three times as much processors, they are still not faster than GT200 GPUs, because ATI in the worst case scenario has only 160 processors. The real number is somewhat higher, but nowhere near 800. Ofcourse, since general purpose GPU usage is becoming more widespread, there are certin scenarios where ATI’s cards will be able to breathe fully, and we believe that they will greatly outperform NVIDIA with its more conservative design. However, since these cards are 99% intended for 3D rendering, it can’t be taken as a real advantage.