Slashdot Mirror
NVIDIA Lifts Veil On GeForce GTX 1080 Ti Performance Reviews, Which Show Faster Speeds Than Titan X (hothardware.com)
MojoKid writes from a report via HotHardware: NVIDIA is officially launching its most powerful gaming graphics card today, the GeForce GTX 1080 Ti. It was announced last week at the Game Developers Conference and pre-orders began shortly thereafter. However, the cards will begin shipping today and NVIDIA has lifted the veil on performance reviews. Though its memory complement and a few blocks within the GPU are reduced versus NVIDIA's previous top-end card, the Titan X, the GeForce GTX 1080 Ti makes up for its shortcomings with a combination of refinement and the brute force of higher memory clocks, based on new and improved Micron GDDR5X memory, faster core clocks and an improved cooler. For gamers, the good news is, the 1080 Ti retails for $699, versus $1200 for the Titan X, and it is in fact faster, for the most part. Throughout a battery of game tests and benchmarks, regardless of the resolution or settings used, the GeForce GTX 1080 Ti performed on par with or slightly faster than the NVIDIA Titan X and roughly 30-35% better than the standard GeForce GTX 1080 Founders Edition. Versus AMD's current flagship GPU, the Radeon R9 Fury X, there is no competition; the GeForce GTX 1080 Ti was nearly 2x faster than the Fury X in some cases.
How far we've come since the SGI Indigo 2 Max Impact is just phenomenal.
In a world of the blind, the one-eyed man is king--and the two-eyed man is a heretic.
It's funny how we say Moore's law is dead when it's about the only thing still kicking.
Clock speeds, the x86 architecture, and software design are all more or less stagnant, which means your typical single-threaded business logic is barely running faster year upon year and CPU benchmarks are pretty flat.
But anything parallel and transistor-hungry is improving by leaps and bounds: the 1080Ti is ~70% faster than its predecessor (with 50% more transistors), AMD is offering 8 cores for the price of 4, 32 for their server models, Intel's Phi is at 72... even smartphones are at 8-10 cores. As Moore predicted, dense ICs are packing more transistors every year, and it looks set to continue for the next several years at least.
How can I believe you when you tell me what I don't want to hear?
Are there great Linux drivers? Are they removing features from the Windows version so you have feature parity with Linux? Oh, well, fuck you Nvidia.
Anons need not reply. Questions end with a question mark.
What do they mean they're launching it today ? I've had a 1080Ti since November - and I had a month-long import delay after ordering in October last year !
Unicode killed the ASCII-art *
The "faster than a Titan" thing has been causing a bit of angst. The early reviews and benchmarks do indeed show that the 1080 Ti outperforms the Titan X (Pascal) in many cases. It's not universal; some games and benchmarks still favour the Titan by a tiny margin, but those are a minority.
But the sheer price of the Titan X (which was unprecedented in the Pascal series) has driven a lot of extra discontent this time around, especially as the 1080 Ti came out with a lower price than a lot of observers had been expecting (there were confident predictions from usually-reliable sources that it would be $200 north of where it actually landed). If you need a bit more salt in your diet, take a look at some of the threads over on the Nvidia forums today from disgruntled Titan X owners.
This is, however, pretty much par for the course in the high-end PC game and it's not as though Nvidia haven't slipped into a predictable cycle over their last few generations (at least since the 700-series) that makes clear how things work. If you want to buy a card that is "top of the range", you've basically got three options:
1) Buy the *80 card that arrives with the first wave of consumer cards in each generation. You will get a few months at the top of the tree, until the release of the (massively more expensive) Titan. This is always the cheapest of the three options, but also the most time-limited.
2) Buy the Titan that comes out a few months after the *80. This will have an absurd price tag - often twice that of the *80. It will be the fastest thing around for, in general, 6-9 months, and even then, the next card may only match it rather than beating it.
3) Buy the *80 Ti that comes out 6-9 months after the Titan. This will generally give you framerates in most games in the +/- 3% range of the Titan, but for a price much closer to the *80. This will hold its place at the top for anywhere from 9 to 15 months, until the release of the next generation of cards. In the next generation, the *80 will outperform the last generation *80 Ti and the *70 will offer broadly comparable (maybe slightly better) performance for around half the price.
I've been going for the *80 Ti route for a while now, on the grounds that the price/performance ratio tends to hold up better over time. I'm seeing complaints at the moment from people who bought a Titan within the last few weeks, which is just bizarre. The 1080 Ti has been known to be close to release since January, so why anybody would take the plunge on a Titan at $1200 under those circumstances is beyond me.
I'm working from home today and waiting for my 1080 Ti to be delivered. I wish I could say I'm not bouncing up and down in my chair going "SQUEEEEEE!!!" like a 12 year old girl at a One Direction concert, but I'm not sure how convincingly I could make that case.
The PC exclusive Ashes of the Singularity is a standard in graphics benchmarks, including the one from the article. Even clearly console-oriented games like Dark Souls 3 are significantly better in graphical quality, framerate and load times with expensive hardware like this.
Maybe I'm out of the loop on something but I can't help but feel this review is inherently biased because they're comparing the 16nm 1080 Ti versus the much older 28nm dual-GPU Fury X (and it should be noted most games can only run on a single GPU without serious problems). With that said wouldn't it be more appropriate to measure it against the 16nm Polaris-based RX 480?