Using Boost, the standard GTX 980 and 970 can reach core clock speeds of 1216MHz and 1178MHz respectively. One of the beauties of Maxwell’s smaller and more efficient Streaming Multiprocessors is that they can be pushed to higher clock speeds. NVidia GTX 980 vs NVidia GTX 970 comparison: Clock Speeds and Textures It’s interesting, though, that nVidia have made so little attempt to differentiate the 980 and 970 here, with both having essentially identical bandwith figures. So we’ll have to reserve judgement for now. Of course, as we’ve already noted, the 980 and 970 use special compression to optimise the memory bandwidth, and those effects won’t be reflected in the raw figures. In comparison, the 780 Ti comes with a gargantuan bandwidth figure of 336GB/sec. However, the 256bit memory buses of the 980 and 970 push them back on paper, leaving them both with memory bandwidth figures of 224GB/sec. Those 7GHz speeds place them on a par with the 780 Ti.
The standard GTX 980 and GTX 970 both come with memory clocks of 1.75GHz (effectively 7GHz when you take into account the quadrupling properties of the RAM). That’s particularly so when we look at memory bandwidth calculations. As such, it doesn’t perhaps fare well in terms of raw figures. The main idea behind Maxwell is that it does rather more with considerably less. NVidia GTX 970 vs NVidia GTX 780 Ti vs AMD Radeon R9 290X comparison.) The end-result is that the new cards’ 256bit memory interface proves not to be the bar to high-performance that you’d think it to be. This concept has been seen in AMD products before, but is relatively new to the mainstream nVidia products, and uses compression to allow the memory bus to process more data without causing a bottleneck. However, Maxwell does have one slick trick up its pipline – the Third Generation Delta Color Compression.
Unfortunately, the 384bit memory bus offered by the earlier GTX 780 Ti has been reduced to a 256bit version in the case of both the 980 and 970. You might expect the 980 and 970 to offer strong memory bandwidth. NVidia GTX 980 vs NVidia GTX 970 comparison: Memory Bandwidth There’s a knock-on effect on texture units too, with each SMM offering a mere eight of these, against the 16 of the SMX. So whereas a standard SMX, for instance, contains 192 shader cores, the humble SMM contains just 128. Inevitably, though, this means that single SMMs are armed with fewer weapons than their SMX counterparts. Each individual SMM is much smaller, meaning that instructions can be handled with greater efficiency. The older Kepler GPUs already made use of the cleverly tweaked SMX (Streaming Multiprocessor neXt generation), but SMMs take this to a whole new level. The key to this reorganisation is the enhanced SMM (Streaming Multiprocessor Maxwell). If the town temporarily runs out of expansion space, the planners have to work instead on making the town itself work better, redirecting traffic more efficiently, and allowing the flow to continue unimpeded. It’s rather like a town that has consistently expanded in size every two years for over a decade. Unfortunately, the two-year cycle has been held up this time round, which is partly why nVidia has worked so hard on streamlining the way the technology itself works.
0 kommentar(er)
