The GTX 950 will be based on a cut down version of the company’s GM206 GPU, code named GM206-250. The fully unlocked version of the chip bears the code name GM206-300 and has 1024 Maxwell CUDA cores and a 128bit GDDR5 memory interface, exactly half that of the GTX 980.
The GTX 950 will have a few SMM ( Streaming Maxwell Multiprocessor ) units disabled to create this new SKU. This means that the GTX 950 will have fewer CUDA cores and texture units than its bigger brother the GTX 960. The exact specifications for the card are yet to be confirmed. But we’ll likely see the ROP count and the shared L2 cache stay the same, to avoid running into the same two segment memory implementation found in the GM204-200 GPU powering the GTX 970 graphics card.
Apart from that, videocardz indicates that the card will be launching with both 4GB and 2GB variants. Although a 4GB memory buffer on such a modest card is much less likely to provide any substantial benefit. Except perhaps in two-way SLI configurations where there’s enough performance to push the settings up and take advantage of the additional VRAM. We’ve yet to confirm if the GTX 950 does support SLI at this time, but the existence of 4GB variants could indicate that card may indeed support SLI.
The GTX 950 graphics card is set to launch within the next couple of weeks to compete against AMD’s $150 graphics card offering, the R7 370 based on the Trinidad GPU.
Below the $150 price point, in the $100-$120 entry level, we’re likely to see Nvidia’s first Maxwell GM107 based graphics cards continue the fight against AMD’s new Tobago based arrival, the R7 360. The GTX 750 Ti and GTX 750 might see some modest price cuts to improve their competitive edge in this hyper-competitive market segment. But there’s really no reason that warrants their replacement. The GM107 GPU is still very much relevant today. In fact very little separates it from Nvidia’s updated Maxwell 2.0 architecture in terms of performance and power efficiency.
Most importantly there’s a strong financial incentive not to replace it as the cost of designing a new MX 2.0 GM107-class GPU would greatly outweigh any benefits, hence there’s no clear business sense in pursuing it. Which is why we’re not likely to see any new GPU cores from Nvidia prior to Pascal in 2016.
Source : http://wccftech.com/
