A high-end GPU is typically released over the course of several months. First, one or two single-GPU cards will be available. Then, a dual-GPU beast will be available over the next few monthsThis is a standard procedure because manufacturers and board partners need to ramp up production, stockpile chips, and work on custom designs.
It appears that this year will be different. Even the release of the GK104, which was not your typical NVIDIA high-end GPU, was out of the ordinary.
NVIDIA’s entire GK104-based product line was released quickly. NVIDIA released the GeForce GTX 690, a dual-GK104 monster, six weeks after the GeForce GTX 680. NVIDIA returned one week later with the release of the GK104-based GeForce GTX 670.
The GTX 670, like its predecessors, will be less expensive, and slower. However, it will be a more power-efficient alternative to NVIDIA’s flagship graphics card. This method not only enables NVIDIA to use GPUs with limited performance. But also enables the company to meet the needs of customers at lower prices. The key to releasing a competitive mid-range graphics card has been striking a delicate balance between performance, price, and yield; as we’ll see, NVIDIA got it just right, releasing a product that’s both competitive and profitable.
Speed
The nvidia geforce gtx 670mx and GeForce GTX 680 are powered by the NVIDIA GK104 GPU. Yes, we’re looking at the same Kepler design and features, albeit at a slower performance level. The main difference is that NVIDIA can achieve this with a mid-tier card by reusing GPUs.
The GTX 670 is notable for being the first NVIDIA GPU to disable one of GK104’s eight SMXes and lower the core clock. On the nvidia geforce gtx 670mx, all 32 ROPs on the GK104 and all four GDDR5 memory controllers. It will be active. Normally, NVIDIA would aim for a larger performance gap by focusing on all aspects of the GPU at once. But as they did with the GTX 570 and its 15 SMs and 40 ROPs, but not this time.
Meanwhile, time rates are equally fascinating. The official figures show a significant decrease in base and boost clocks when compared to the GTX 680. The GTX 670’s base clock of 915 MHz and boost clock of 980 MHz are 9%, respectively than the GTX 680. However, the GK104, like the GTX 680, will spend the majority of its time boosting. And not always at the official boost clock. Depending on the game and the GPU, the GTX 670 can boost to within 40MHz of the GTX 680, or about 3.5% of the more powerful sibling’s clock speed.
Memory
Aside from ROPs, the memory infrastructure has gone unnoticed. The GTX 670 will have a memory clock speed of 6.008GHz, a 256-bit memory bus, and a memory bandwidth of 192GB/sec, which is identical to the nvidia geforce gtx 670mx. This is especially intriguing because NVIDIA has previously routinely removed memory controllers and ROPs, lowering memory clocks. Because the GK104 is an xx4 GPU rather than a full successor to the GF110 with its 48 ROPs, it appears that NVIDIA is concerned about ROP and memory performance and will not sacrifice performance for the GTX 670.
Performance
When both cards are running at their base clocks, the GTX 670 has 100% of the memory bandwidth, 91% of the ROP performance, and 80% of the shader performance of the GTX 680. In terms of specifications, the GTX 670 is much more comparable to the GTX 680 than the GTX 570 was to the GTX 580 or the GTX 470 was to the GTX 480. That is, the difference between the GTX 680 and the GTX 670 will not be as great as it was between the GTX 570 and GTX 580.
The power consumption difference between the GTX 580 and GTX 570 will be about the same. The GT 670 has a TDP of 170W, which is 25W lower than the GTX 680. The unofficial GPU Boost power target of the GTX 670 is 141W, which is 29W lower than the GTX 680. Because the GTX 680 had the lowest TDP for a GPU in its class, the GTX 670 has the lowest TDP we’ve seen from NVIDIA in a long time.