Maxwell on its own is the most impressive GPU technology NVIDIA has ever created, especially when it comes to power consumption. But with NVIDIA, there's never just one part of the architecture that stands out, but when it came to Maxwell, the benefits of the low power consumption were a snowball effect, but a very, very good one.
The GeForce GTX 960 continues this tradition, with a TDP of 165W, so a single card will only draw a maximum of 120W. That is, unless it's a third-party card with higher clock speeds and a more elaborate cooling system. This allows the GTX 960 to only require one 6-pin PCIe power connector from its users, which is a great achievement from NVIDIA considering it also features a decent punch in the performance department.
Inside of the GTX 960 we have the GM206 GPU, which is part of the Maxwell architecture. NVIDIA has made some changes to the cache hierarchy of the Maxwell architecture, with each of the GM206's SMM units featuring its own dedicated 96KB shared memory, while the L1/texture caching functions are joined together into a 24KB pool of memory per pair of processing blocks (48KB per SMM). Before Maxwell, the Kepler-based cards featured 64KB of shared memory function that was shared with the L1 cache.
This means that the GM206 core, or the GTX 960, is capable of pushing around 1.4x more performance per core than the GK106 Kepler CUDA cores, and 2x the performance per watt (which is just as, if not more important).
Keep in mind, NVIDIA is doing all of this magic on the ageing 28nm process. There were rumors that the company would be shifting over to the 16nm or 20nm process, but this has been delayed until at least late this year, or 2016 at the earliest. NVIDIA's Maxwell architecture couldn't have come at a better time, as it was able to provide a dramatic improvement to the performance per watt, something that the GTX 960 drives home, still on 28nm.
Memory Subsystem - 128-bit Isn't All That Bad
This is one of the bigger changes that mid-range GPUs face, the memory bus is usually cut short of the bigger brother. On the GTX 970 and GTX 980 cards, we have a full 256-bit memory bus, but NVIDIA has cut this down to a 128-bit bus. Normally, I would baulk at this, but the memory bus that NVIDIA has been working on with Maxwell is super-efficient.
Gone are the days when we would look at a 128-bit memory bus and instantly think "yep, this is a low-end card that is going to suck at performance" and gets thrown into the "low-end gamers" pile. NVIDIA does some incredible things in its memory pipeline, which we're going to explain now.
NVIDIA has baked 2GB of RAM onto the GeForce GTX 960, with the company also revamping the memory subsystem of the GM206 core, with their "third-generation delta color compression engine" offering "new modes for color compression, allowing the GPU to more effectively use its available memory bandwidth". The new Maxwell-based GM206 core also uses around 25% fewer bytes per frame compared to the previous architecture, Kepler.
This translates to the GM206 GPU core, that the Kepler-based memory system running at 9.3Gbps would provide the effective memory bandwidth similar to Maxwell's enhanced memory system. But with its 128-bit memory bus, the GTX 960 has memory bandwidth of 'just' 7Gbps, but it is able to provide a tad more bandwidth than its direct predecessor, the GK106.
Moving this into numbers we better understand, the GTX 960 has an effective memory bandwidth of 148.8GB/sec, while the GTX 660 has 144.2GB/sec. So you can see, that while we have a 128-bit memory bus, the Maxwell architecture is doing some tricky things inside of the GM206 core to boost these numbers up.
PRICING: You can find products similar to this one for sale below.
United States: Find other tech and computer products like this over at Amazon's website.
United Kingdom: Find other tech and computer products like this over at Amazon UK's website.
Canada: Find other tech and computer products like this over at Amazon Canada's website.
- We at TweakTown openly invite the companies who provide us with review samples / who are mentioned or discussed to express their opinion of our content. If any company representative wishes to respond, we will publish the response here.