NEXT-GEN GRAPHICS CARDS
AMD and Nvidia are at it again, but this time they’re aiming at opposite ends of the market!
JUST WHEN WE THOUGHT MOORE’S LAW was history, that gaming graphics was grinding to a halt, it’s happened. Finally, belatedly, both AMD and Nvidia are rolling out a hot new generation of die-shrunk graphics chips. Cue a collective sigh on behalf of PC gamers across the globe.
Among the new chips are the bases of indisputably the fastest and most powerful graphics cards we’ve ever seen. Some of the numbers involved are spectacular, going on preposterous. Billions of transistors per chip, trillions of floating point operations per second, countless pixels pumped—it’s all very exciting.
But along with a new performance paradigm has come a new pricing paradigm, from Nvidia at least. The Mean Green Graphics Machine has set a distinctly unwelcome standard in that regard with the $1,200 Titan X. That’s one hell of a price for a 3D card. Even some of Nvidia’s latest mid-range boards seem to be realigning the market at higher prices than before. Happily, that’s just one half of the story. At the same time, AMD is making its play in exactly the opposite direction, with Polaris, an unambiguously and unashamedly value-oriented family of GPUs. The most expensive of the new Polaris-based Radeon cards rocks in at just $260. We won’t see a new big-money hitter from AMD until at least the end of this year or, more likely, in early 2017.
In that sense, then, this new generation of graphics cards ought to have something for everyone. At the top end, there are new uber GPUs for those with money to burn. Lower down, there’s the promise of more performance at lower prices than ever before.
The real test will be what all that means in practice. Has Nvidia, for instance, finally created a GPU capable of cranking out smooth 4K frame rates, no matter what you throw at it? Likewise, has AMD delivered on its promise of VR gaming at mainstream prices? Put simply, are these new graphics cards all they’re cracked up to be? Has the wait been worth it? There’s only one way to find out….
“STOP THAT, IT’S GETTING SILLY.” That was our first reaction when Nvidia announced its new Titan X graphics board for $1,200. It’s not that we’re against high-end hardware—high-end hardware is what makes Maximum PC tick—but there’s a limit to everything, and 1,200 bucks for a single rendering stick? Puh-lease. That’s especially true when you take a closer look at what you’re getting for your money with a Titan X. Indeed, our objections to Titan X’s pricing arguably apply to other members of Nvidia’s new Pascal graphics family. But that’s getting ahead of ourselves. First, let’s consider the context behind this new generation of boards.
More than any other generation, this one is all about those tiny nanoscopic suckers known as transistors. Or, more specifically, it’s about the size of those transistors. Four and a half, long hard years, we’ve waited for the production process by which graphics chips are made to move on from 28nm. Yes, it was late 2011 when AMD launched the first 28nm graphics chip, the Radeon HD 7970.
In the meantime, Intel CPUs made the leap to 22nm, then 14nm transistors, almost in step with Moore’s Law, which has long predicted that transistor density in computer chips doubles every couple of years. Things definitely didn’t work out like that for a while in the PC gaming graphics industry. Part of that is down to the fact that just one Taiwanese outfit, TSMC, was responsible for manufacturing all of Nvidia’s and AMD’s high-performance GPUs. If TSMC stumbled, and it did after 28nm, the whole of PC gaming would be tripped up.
MOORE OR LESS
Thankfully, it seems like TSMC has got its act together again. It’s basically skipping the 22nm node altogether, and jumping straight to 16nm, which almost puts graphics back on that Moore’s Law track. Just for an added twist, however, Nvidia is sticking with TSMC and its new 16nm node for the new Pascal family of graphics, but AMD has made a bold leap into the relative unknown, and jumped ship to Global Foundries, and a brand new 14nm node.
We say “relative unknown” because Global Foundries was once part of AMD, and makes its CPUs and APUs. However, manufacturing complicated graphics chips is a whole different ballgame, and there Global Foundries has little to no track record.
That very likely explains why AMD has chosen to start its graphics production partnership with a family of compact, value-oriented graphics chips, rather than roll the dice on something bigger, beefier, and more likely to throw up manufacturing glitches. Suffice to say that this all adds up to an awful lot more than just a regular graphics refresh. There’s some major action going on this year.
So far, Nvidia has been busiest. First, it cranked out its new performance boards, the GeForce GTX 1080 and 1070. They’re a major step forward over Nvidia’s Maxwell family. What’s interesting is that it’s arguably clock speeds rather than transistor count where Nvidia’s new chips are doing the damage.
GIANT LEAPS
The GP104 chip inside the GTX 1080 and 1070 rocks in at just 7.2 billion transistors, which is modest when you consider the double node jump from 28nm to 16nm, and the fact that the old GTX 980 packed 5.2 billion transistors in its GM204 chip. It’s not a huge surprise to find that the 1080 only has around 25 percent more CUDA cores than a 980. Instead, it’s the big bump from 1,216MHz to 1,733MHz for the 1080’s Boost clock speed that stands out. Imagine if Intel suddenly launched a CPU running at over 6GHz. That’s the kind of leap Nvidia has delivered.
Nvidia has also wheeled out its mainstream gaming chipset, the GTX 1060. That’s based on another new GPU: the GP106. With 1,280 CUDA cores, it’s also a modest step over its progenitor, the GTX 960, with 1,024 CUDA cores. Again, Nvidia has cranked the Boost clock from 1,178MHz to 1,709MHz, which is where the performance gain lies.
Finally, we have the daddy of the new range, the Titan X. Nvidia’s mega-chip follows a similar pattern—
All are more expensive than the boards they replace, despite being
based on smaller, cheaper GPUs.
the new Titan X gets a modest uptick in functional units, but a major boost in operating frequencies.
The thing about all these new Nvidia GPUs is they’re not very big. And size matters. Consider the new Titan X: At 471mm2, its GP102 chip is much smaller than the 601mm2 of the old Titan X. In fact, it’s nearer to the GM104 chip in the GeForce GTX 980, which comes in at 398mm2. The Pascal Titan X also has the same 12GB memory buffer as the old Maxwell edition. So why is it so expensive?
You could ask the same question about the new $600 GTX 1080 and, indeed, the $249 GTX 1060. All are more expensive in terms of launch price than the boards they replace, despite being based on smaller, cheaper GPUs. Enter, therefore, the alternative approach from AMD. Its new Polaris GPUs don’t break any outright performance records, but they do claim to set new standards for gaming grunt at an affordable price.
The most expensive of AMD’s new boards, the Radeon RX 480, is cheaper at $260 for the 8GB version than Nvidia’s GTX 1060. The 4GB version is just $200. For that you get what AMD claims is a “premium VR gaming experience,” the inference being that this new $200 board is good to go with the Oculus Rift or the HTC Vive. From there, the other members of the Polaris family that AMD has revealed, the RX 470 and RX 460, only get cheaper.
PERFORMANCE ART
In fact, the peculiar truth is that there is no overlap between Nvidia and AMD thus far with this new generation. It’s not a straight fight at any price point. Instead, it’s about measuring up the gaming reality against the claimed proposition. AMD’s new boards need to deliver a level of performance never seen before at $200, and make cards from the previous generation look old, expensive, and pointless. As for Nvidia’s new GPUs, they need to deliver a level of performance never seen before, period.