QUICKSTART
THE NEWS
First 7nm gaming GPU; Google targets adblocking; the 7nm Zen 2 chiplet; Facebook platforms to merge; more….
AFTER MONTHS of focus on Nvidia and its groundbreaking RTX cards, we have a solid response from AMD in the form of the Radeon VII. Unlike the RTX cards, we don’t have a fancy new GPU architecture or extra features, but we do get the first consumer outing for the 7nm Vega 20 GPU, which first appeared in November in the Instinct enterprise cards.
The Vega 20 in the Radeon VII has 3,840 stream processors—256 fewer than the Vega 10-powered Vega 64. It carries 60 compute units (four fewer than the enterprise incarnation, and four fewer than the Vega 64). Clock speeds have jumped significantly; the base clock is now 1,400MHz, with a peak boost of 1,750MHz. The die shrink may have lost some cores, but the 200MHz speed bump compensates for that. On paper, the computational power is up by 10 percent.
On the memory side, AMD has packed the Radeon VII with 16GB of high bandwidth memory, HBM2, on a 4,096-bit bus. It has twice the memory on a bus twice as wide as its predecessor, which helps push the headline memory transfer rate from 484GB/s to 1TB/s, faster than is possible with GDDR6. While the GPU might be small—just 331mm2—it’s on a hefty double-width, fulllength card, with triple fans. Judging by the two eight-pin power connectors, it’ll be burning near 300W at full tilt.
AMD’s presentation at CES claimed about a 25 percent performance gain over the Vega 64 on apps such as Adobe Premiere and Blender, helped considerably by that fat memory pipe. On games, it boasted a 35 percent jump for
BattlefieldV, and a 25 percent jump for Fortnite. AMD also matched it against the RTX 2080, and claimed almost identical frame rates on
BattlefieldV and FarCry5. The Radeon VII will launch at $699, leaving quite a gap between it and the Vega 64, at $400 or so. It’s aimed squarely at the RTX 2080, which also starts at $699. The comparison is unavoidable, and all the first benchmarks pit the two against each other. It would appear to be a close-run thing for the most part, too. Next month, we’ll bring all the gritty details you could want.
Is it enough? If games are your thing, similar performance for the same money as an 2080 is hardly compelling, as you don’t get the RTX ray tracing or AI features (especially that juicy DLSS). That extra 8GB sounds useful, but few games need that much headroom. If you’re looking at content creation, things look much more desirable—oodles of fast memory is just what you need. It does have some great tricks, such as strong FP64 performance, which shows its enterprise roots.
AMD has more to come, of course—the all-new Navi architecture is anticipated to surface before the end of the year. It’s expected to be aimed at the mass market, including consoles and mobile devices, leaving Vega-powered cards at the top end for now. It’s no secret that AMD has been working closely with Sony to get Navi into the PlayStation 5. The next-gen Xbox also looks a likely customer.
Back at Nvidia, we have confirmation of new midrange Turing-powered cards using TU116 cores. There’s the GeForce GTX 1660 Ti with a 1,500MHz base clock, 1,536 CUDA cores, and 6GB of RAM. It’s launching at $279. Below this is the GeForce GTX 1660 with 1,280 cores and a slower memory clock, due in March for $230 or so. A 1650 will follow at $180. This fills out the Nvidia range nicely. Being GTX cards, you won’t get the fancy extras of the RTX series, but the Turing core will give a useful performance hit. A 1660 Ti is about 20 percent faster than a 1060 in first tests.
Meanwhile, Nvidia CEO JenHsun Huang’s response to the Radeon VII has been somewhat tactless. He reportedly called its performance “lousy, and there’s nothing new.” Harsh, sir. The Nvidia and AMD rivalry over graphic cards is analogous of Intel and AMD’s sparring over processors. One side produces the expensive and ultimately faster gear, while the other produces something that can still do most of the heavy lifting you need, but at a more attractive price. Or that used to be the case; AMD has pitched the Radeon VII pretty high. At least we have some competition near the top now, and that is always a good thing.
While the GPU is small, it’s on a hefty double-width, fulllength card, with triple fans.