Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Nvidia’s stronger, faster Pascal architecture expected to debut Friday // GTX 1080 Benchmarked
![[Deleted User]](http://we.vanillicon.com/v2/e7c9305aa2cdf6069206388b4771c61d.svg)
[Deleted User]
Posts: 12,262
in Hardware
The user and all related content has been deleted.
Comments
In the political world, it's well-known that if bad news has to come out eventually, you announce it on a Friday evening because that's when it will get the least attention.
I can only think of one tech NDA ever that ended on a Friday evening: the GeForce GTX 480 and GTX 470, a.k.a., the biggest disaster of a GPU in the last eight years.
I'm not sure if this is ominous for Nvidia or just bizarre. One can't rule out bizarre; who would have expected Nvidia to start naming its top end GeForce cards Titan instead of a number? Still, AMD publicly showed off Polaris in early January, and Nvidia still hasn't publicly demonstrated working Pascal cards.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
거북이는 목을 내밀 때 안 움직입니다
When you don't want the truth, you will make up your own truth.
HB2? "Unlikely to make it? Yeah, if youre unsure about that....NO it wont make it in cards this year lol.
40% over Titan X? OCed 980ti reaches well over 25k. 980ti at 1500=1080 at 1860 apparently. And clock for clock 1080 (Pascal) is slower than Maxwell if those benches are actually true. At this time i would call that a fake.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
Ceratop, you know we're talking about info that release LAST Friday. As in 2 days ago. What I am doing is calling them out to admit they were wrong. Which they won't do. I will also call them out when they are proven wrong about the release dates of the 1080 and 1070.
I am one of the few people on these forums that will admit when I am wrong about something and take ownership of that. I'm getting tired of other people making grandiose claims based squarely in 9/11 conspiracy theory level garbage, and "Getting away" with it. Hell, not even getting away but getting lots of pats on the back from other fanboys.
This whole NVidia/AMD fanboying on both sides is getting ridiculous tedious and tiresome. I would be ecstatic if AMD actually released a product that was objectively better than Nvidia in every way, because this is always good for the consumer and causes innovation, competition, and lower prices for better products. The consumer are the people who win in these scenarios.
That being said, being intellectually dishonest and trying to smear the competition achieves nothing. It's not going to change the reality. All information points to the fact that reviewers already have review cards and the NDA will be lifted on 05/17 and the cards will be available on 05/27. Now, will there be supply issues? Probably, there almost always are. The R9 Fury was extremely difficult to get ahold of because of supply issues. So neither Nvidia nor AMD have the market cornered on supply issues on high end parts.
Instead we have Quiz claiming that Nvidia is quite literally just bull faced liars, and that literally everything they've said about pascal is lies and obfuscation, etc.
It's honestly just sad. I always suspected he was an AMD fanboy (he only ever recommends NVidia cards in extremely limited scenarios, just enough to make a claim he is unbiased), but these last couple weeks have absolutely proven that he is irrational on this topic.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
The 1070 will probably be released sooner than the 1080. The 1080 is definetly having supply issues if its using a memory standard that isn't ready for mass production.
http://wccftech.com/nvidia-pascal-trouble-asynchronous-compute/
http://www.overclock3d.net/articles/gpu_displays/nvidia_pascal_mia_at_ces_reportedly_in_trouble/1
Pascal has been in trouble for a while , hence this rush to showcase a working 1080.
http://www.3dmark.com/3dm11/11061015
?.?
post as 1080 , but reads the core as 980ti ..lol
http://www.3dmark.com/fs/6695719
I'm getting a feeling that Nvidia is about to take a dump.Nothing is adding up ...
Benches will show how it is, it seems i was somewhat wrong as those performance leaks seems to be legit so far.
"New" technologies are blah.
Ansel - i dont see it becoming thing
Sound improvements - AMD has had True Audio for 3-4 years now? Thats even sported in PS4 iirc
SMT - maybe in VR its "worth it" (as it has to be assessed in actual game) but basicall NVidia now makes fake resolutions - in 4k for instance, only center of the screen is rendered in 4k and edges are rendered in lower resolutions, something like "dynamic resolution" which consoles have done (successfully) for a very long time now. Con is that it actually lowers image quality. I guess if someone buys good 4k monitor he wants 4k resolution exactly because of that image quality.
And speed....thats most important and i guess people expected more as it isnt really much different from Kepler - Maxwell jump but this time it was almost twice smaller node jump as well while Kepler - Maxwell was same 28nm. Or, from other perespective, much better price/perfromance at least.
http://www.3dmark.com/fs/6695719
But hey, if plenty are available on Newegg, and Amazon, and I can walk down to Best Buy and pick one up - then I was wrong - but you can't say that I was wrong until the 27th comes and goes and we see what the situation really is. I mean, it's not like GPU manufacturers have ever lied before, right? Woodscrews, baked benchmarks in drivers, pushed release dates, claimed performance specs, etc....
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
The foundry is the god, judge, jury and executioner, everything else comes afterwards.
The GeForce GTX 1070 will be about as fast as a Radeon R9 Fury X or a GeForce GTX 980 Ti. The top Polaris 10 card will likely be about that fast, too, or perhaps a little slower. The GeForce GTX 1070 will have an MSRP of $379, or you can get a "founder's edition" card for $449. Initially, only the founder's edition cards will be available, but there will eventually be other cards available. Polaris 10 will be $300, and is likely but not certain to have widespread availability before the GTX 1070.
The GeForce GTX 1080 uses GDDR5X memory, which means it won't be available until GDDR5X is. Considering that GDDR5X started sampling in late March, for real availability, you're probably looking at somewhere near the end of this year at best. Fall is possible, but Summer is not, outside of a tiny trickle of cards.
Nvidia claims that the GeForce GTX 1080 will offer 50 GFLOPS/W (which is believable), while they had previously said that the Tesla P100 would only offer about 35 GFLOPS/W. This is in spite of the Tesla P100 being clocked much lower and having HBM2, both of which should improve its energy efficiency. So basically, Nvidia's top end GP100 chip is horribly broken. As it's waiting on HBM2, Nvidia has time even to do a base layer respin and eventually fix it without that much of a delay in the final launch date. But Nvidia does sound like they're going to sell real cards with it, and until that gets fixed, GP100 is a good candidate for the most broken discrete GPU chip in the history of discrete GPU chips, or at least in the last decade.
It's not at all clear whether Nvidia will have any Pascal chips for cards slower than the GTX 1070 anytime soon. AMD will, with Polaris 11 and lower bins of Polaris 10. It's possible that Nvidia has a smaller chip ready to go soon, but they haven't talked about it yet, nor have I seen any rumors of it. If they don't, then AMD will pretty much own the laptop GPU market for however many months it takes until a smaller Pascal chip arrives.
Meanwhile, reviews of the GeForce GTX 1080 will release on a Tuesday, and probably be highly positive, as it's going to be the fastest GPU that has ever been reviewed. But the official release doesn't come until a Friday (here we go again) 10 days later, at which point, people will see that the GTX 1080 is out of stock and you can't get one, even if you're willing to pay $700 for a "founder's edition" reference card.
Did I miss any super positive news for Nvidia in there? I would consider Nvidia putting more focus on viewports to be a good thing, and an underappreciated part of their talk. But that's a software thing, not even a driver thing, and developers have had the capability to do it since DirectX 10 in 2007. Well, they've had the available features in graphics APIs, even if most game programmers lack the math background to know what to do with it.
Now, if you think that the news that came out of the presentation is that the GeForce GTX 1080 will hard launch on May 27 for $600, you might well think that it was positive news. But if that's what you took out of it, you're not paying attention.
There is the issue of how well you can exhaust that computational capability. In simple synthetics, you can get something like 99% of theoretical peak performance on Maxwell or the various GCN cards, though not on Kepler. Nvidia does have more clever scheduling than AMD, so in real code that is doing complicated things (such as graphics), they tend to be able to come closer to exhausting theoretical peak performance than AMD. AMD generally counters by having higher theoretical peak performance available; I suspect that this is made possible by space and power savings allowed by simpler scheduling. This has been the case every single generation in the entire unified shader era.
I'm personally expecting Pascal to be about as clever about scheduling instructions as Maxwell, and Polaris about as clever as GCN, which would mean that Nvidia maintains their scheduling advantage into the next generation. That advantage is real, but it's much smaller than it was before GCN.