Last month, NVIDIA made an impactful product launch, the GeForce GTX 580, which shook the GPU world and tilted performance leadership and performance per Watt figures back in favor of NVIDIA. Such was its impact that it may have caused its competitor to delay upcoming product launches to probably rework them up. Not very far from the launch of those products, NVIDIA launched its second GPU in the GeForce 500 series, the GeForce GTX 570. As the name would suggest, the GTX 570 is the "value" SKU based on the GF110 GPU, while the GTX 580 leads with all components of the GPU enabled.
Reviewed today, the GeForce GTX 570 is GF110 with one streaming multiprocessor (SM) disabled, yielding 480 CUDA cores (same number as that of the GTX 480 from the previous generation), but with a 320-bit GDDR5 memory interface, connecting to 1280 MB of memory. The ROP as a result is 40. Thanks to the improved power efficiency, NVIDIA was able to up clock speeds on the GeForce GTX 570 over the previous generation without having to worry about rising TDP. GTX 570's core is clocked at 732 MHz (higher than that of GTX 480), CUDA cores at 1464 MHz (again, higher than that of GTX 480, translating into higher shader compute power), and 950 MHz (3800 MHz GDDR5 effective) memory, yielding 152 GB/s of memory bandwidth. Palit's GTX 570 Sonic Platinum reviewed today, is based on an in-house design for the PCB and cooler (one of the first for the whole GeForce 500 series). It makes use of a refreshingly silver dual-fan GPU cooler, and a non-reference design PCB, geared for some overclocking.