Today, Nvidia launched the GTX 1080 Ti ($699), its new highest-end consumer GPU. This release follows a pattern Nvidia established with the GTX 700 series back in 2013 of releasing a new workstation / prosumer GPU at the highest end of the market (the original Titan back then), followed by a cheaper consumer variant some months later. The “Ti” cards have historically been much better values than the Titan family, and we expect to see that trend continue here as well.
The GTX 1080 Ti is a GP102-based design with 3,584 GPU cores, 224 texture units, and 88 ROPS (3584:224:88). Its base clock is 1480MHz with a 1582MHz boost clock, and it features 11GB of RAM as opposed to the 12GB buffer on the Titan X. Nvidia has stated they expect the GTX 1080 Ti to equal or surpass the Titan X’s performance, making this card a far more cost-efficient choice compared with its $1,200 predecessor.
Part of those gains come courtesy of the 1080 Ti’s increased memory speeds (courtesy of cleaner RAM signaling, as shown above). While it uses an unusual 352-bit memory bus compared with the Titan X’s 384-bit architecture, the 1080 Ti compensates for its narrower pipe with higher memory clocks. Where the Titan X used 10Gbps RAM, the 1080 Ti relies on 11 Gbps memory, for a very slight increase in overall memory bandwidth (480GB/s versus 484GB/s).
Nvidia has promised that the GTX 1080 Ti will be an average of 35% faster than the GTX 1080, and while we don’t have a 1080 to directly compare against, our comparisons against the 1070 suggest they largely hit that target.
Product positioning, review focus
The 1080 Ti arrives at an unusually quiet time in the GPU industry. When AMD announced that it would refresh its midrange GPUs with its 14nm Polaris architecture, it implied that its new high-end GPU family, codenamed Vega, would arrive by the end of the year. That schedule has since slipped by a full six months, with Vega now expected sometime in the second quarter.
By delaying, AMD has opened an unprecedented gap in its own high-end GPU lineup. By the time the RX Vega launches, it will have been nearly two years since AMD launched Fury X. It’s not unusual for AMD or Nvidia to enjoy a 3-6 month lead over the other, and longer periods are not unheard of, but two years is higher than anything we’ve seen from either company… ever.
Unfortunately, our ability to compare the GTX 1080 Ti against other Nvidia GPUs is limited as well. Nvidia was neither able to provide nor assist us with acquiring a GTX 1080 (the most logical point of comparison) and our own efforts to secure a sample in time for this review were unsuccessful. The GTX 980 Ti and older Maxwell Titan X were both available for testing, but we already know how those cards perform — they’re between 0 and 10% slower than the GTX 1070 we’ve already tested.
At its core, a review is designed to answer questions that a reader might have about a particular part. Rather than writing a 3,000-word exercise to answer a question we already know, we decided to tackle the 1080 Ti from a different angle. Last week, AMD launched Ryzen, its biggest CPU architectural refresh in 15 years. While the Ryzen 7 1800X is an amazing chip in most respects, its 1080p gaming performance was significantly lower than Intel’s Broadwell-E in our handful of gaming tests.
The launch of the GTX 1080 Ti gives us an interesting opportunity to examine whether AMD’s new Zen architecture can compete against Intel in multiple games and resolutions, including extremely demanding 4K gaming scenarios. Testing against the most powerful GPU on the planet helps ensure we’re getting the best look at the long term scaling prospects for Ryzen that we can. We’ve also tested the GTX 1070, to see how both CPU manufacturers’ fare with a lower-end, but still quite powerful, Pascal GPU.
There were two plausible CPUs to pick for this review — either the Broadwell-E 6900K, which compares directly with the Ryzen 7 1800X in core count and clock speed, or the Core i7-7700K, which is Intel’s top mainstream part (there wasn’t time to benchmark a third system). In this case, we opted for the 6900K, since that’s the Intel CPU closest to Ryzen’s overall value proposition. Chances are, if you buy a $500 CPU, you’re buying it for multi-threading, with gaming as an important option to serious hobby.
We had to compromise somewhat on memory configurations due to early Ryzen motherboard BIOSes offering limited compatibility with high-speed, high-capacity DDR4-3200. We tested Ryzen 7 1800X with 16GB of DDR4-3200 (two DIMMs), and our Broadwell-E system with 16GB (4 DIMMs) of DDR4-2667. The Broadwell-E still enjoys a substantial memory bandwidth advantage over Ryzen thanks to its quad-channel memory configuration, the while Ryzen 7 1800X has the best memory configuration we could give it.
Our GTX 1070 tests were run using the Nvidia Forceware 376.88 driver, while our GTX 1080 Ti used the 378.78 release driver. We attempted to use the 378.78 driver for our GTX 1070 tests as well, but the card proved unstable and we fell back to the earlier release for both our Intel and AMD benchmarks. Speaking of the GTX 1070, we’re tapping the Gigabyte G1 Gaming GPU we reviewed last fall for this review.
Click here to read more.
Source: Extreme Tech