SFF.Network NVIDIA Set to Announce Pascal GTX GPU's Today

We've seen dribs and drabs of NVIDIA's upcoming Pascal refresh for their consumer graphics cards - from leaks of reference shrouds, to rumored (if not highly questionable) benchmarks. But the most recent slew of data has enough substance to suggest that it's the real deal, and the timing of the leak just before NVIDIA's announcement of a livestream event later today all but confirms that we're about to see the next generation of consumer flagships.

Eagle-eyed folks at videocardz.com caught some benchmarks published without identifiers indicating their GPU of origin (though driver names did spill the beans), and from that they've been able to construct a semi-complete specifications table that compares the rumored GTX 1080 and 1070 against the recently-released Tesla P100...

Read more here.

 

[iFreilicht]

A 1070 would be absolutely sufficient for VR and I really want that to exist in a roughly ITX package, but yeah, I'm not seeing a short 1080 happening either. Getting one would be absolutely amazing, though, it would double the possible performance/liter ratio.

 

[PlayfulPhoenix]

I'm not really holding my breath for a short GTX 1080. A 1070 would do just fine for me, I'm not needing VR, and already would stomp on the 970 (and probably 980). I can see more non-reference cooler designs going with short PCBs for them.

A 1070 would be absolutely sufficient for VR and I really want that to exist in a roughly ITX package, but yeah, I'm not seeing a short 1080 happening either. Getting one would be absolutely amazing, though, it would double the possible performance/liter ratio.

The 1070 beats a Titan X, so yeah, it should do

 

[PlayfulPhoenix]

So, I'm seeing conflicting information on what the difference is between the base 1080 and the "Founder's Edition" one. I was under the impression that the only difference was the fancy shroud and heatsink, but I'm hearing that they've also binned chips for it, and that some of the power delivery components on the more expensive card are improved.

I very much hope that such isn't the case, but does anyone know either way?

 

[EdZ]

From what I can gather: the 'Founders Edition' will be the only one using the neo-NVTTM cooler, and will all use the reference PCB design. Non-Founders-edition cards may use either the reference PCB or an OEM custom PCB, but none will have the neo-NVTTM cooler. I haven't found anything other than rumours about chip binning.

 

[Dedaciai]

Single 8-pin is awesome for an mITX setup; fewer cables and clutter in my little NCASE.

 

[PNP]

but an OEM design could cut down the phases to a more reasonable number to produce a shortened card.

Smaller feature sizes increase the sensitivity of a transistor to voltage fluctuations so I'm not sure how many more phases could be removed.

Whether that precludes a shortened card depends on how willing OEMs are to add extra PCB layers in to ease routing. More layers = more cost, so unless they think they can sell more short GTX 1080s (or sell them at a sufficient premium) they may not bother.

Then traditionally packaged DRAM does preclude small cards, just not for technical reasons. The AIB designs for the Fury came with some pretty enormous air coolers too, but the underlying card was short. I'd be willing to accept that sort of situation since a waterblock resolves it.

 

[PNP]

There were quite a lot of graphs in there that were pretty sketchy, and this one from the GeForce website:

Spoiler: dat X-axis

But then that demo of the 1080 running 2.1 GHz at 67 degrees C? I wonder if the vapor chamber has well and truly returned to the NVTTM cooler and what the fan curve looks like.

 

[iFreilicht]

Man they must've put a lot of research into those accurate graphs.

I think they just ran the fan at full throttle through the demo as nobody could hear it anyway, but it's still quite impressive.

 

[IntoxicatedPuma]

I'm not up to date on video card performance but how does the Titan X compare to the R9 Nano? If AMD drops the price for that under $400, does a short PCB 1070 have some competition?

 

[CC Ricers]

I'm not up to date on video card performance but how does the Titan X compare to the R9 Nano? If AMD drops the price for that under $400, does a short PCB 1070 have some competition?

Titan X beats both R9 Fury and Nano. The Nano edges out the GTX 980 in performance in most cases.

 

[PlayfulPhoenix]

From what I can gather: the 'Founders Edition' will be the only one using the neo-NVTTM cooler, and will all use the reference PCB design. Non-Founders-edition cards may use either the reference PCB or an OEM custom PCB, but none will have the neo-NVTTM cooler. I haven't found anything other than rumours about chip binning.

This much I know, and I'm fine with it. But if the boards and chips themselves are different, I'll be pretty unhappy - I think $100 for a very nice shroud and a vapor-chamber cooler is totally reasonable. I think gimping the "entry-level' hardware - which is already $50 more over the last generation and pretty late-coming - is both lame and unreasonable.

Single 8-pin is awesome for an mITX setup; fewer cables and clutter in my little NCASE.

Indeed, and I'm surprised it isn't emphasized more by reporters and NVIDIA alike. It makes our lives substantially easier, especially for SLI rigs - and that's even the case for larger enclosures.

There were quite a lot of graphs in there that were pretty sketchy... But then that demo of the 1080 running 2.1 GHz at 67 degrees C? I wonder if the vapor chamber has well and truly returned to the NVTTM cooler and what the fan curve looks like.

NVIDIA was masterful at phrasing metrics in a very particular way in order to maximize the relative performance gains or efficiency gains across generations. The "2x performance, 3x efficiency for VR" was a particularly good example of this.

Still, they did spell out the TFLOPS for each unit, and that's a pretty good apples-to-apples comparison. The 1070 has 6.5 TFLOPS (basically equivalent to the Titan X), and the 1080 has 9 TFLOPS plus the significantly faster memory.

And to clarify on the cooler - yes, its a vapor chamber. Given the 180W power rating of the part, that's a lot of cooling headroom, hence the clocks they were able to demonstrate. In the past, NVIDIA has been very reasonable in terms of presenting the "OC-ability" of their cards, so I'm hopeful that such overclocking will turn out to be pretty practical for the majority of users.

Man they must've put a lot of research into those accurate graphs.

I think they just ran the fan at full throttle through the demo as nobody could hear it anyway, but it's still quite impressive.

Maybe, but at 67C I'd think the reasonable thing to do would be to trade some noise for some heat. You don't gain anything from better temps unless you're treading 80C, really.

 

[PlayfulPhoenix]

TLDR: AMD's gunna have some price adjusting to do. But they're gunna release some parts in this price range soon anyway, so they'll likely just wait for that.

Exciting times!

 

[PNP]

NVIDIA was masterful at phrasing metrics in a very particular way in order to maximize the relative performance gains or efficiency gains across generations. The "2x performance, 3x efficiency for VR" was a particularly good example of this.

I hadn't ever heard Jen-Hsun speak before so I was suspicious when he kept playing the repetition fallacy. Does he usually do that?

 

[PlayfulPhoenix]

I hadn't ever heard Jen-Hsun speak before so I was suspicious when he kept playing the repetition fallacy. Does he usually do that?

NVIDIA and AMD both are good at finding the right phrasing to build graphs that exaggerate how good their stuff is. But NVIDIA's marketing is definitely... slicker, in this regard.

It's annoying, and I honestly think they don't win anything by doing it, but I don't really mind, given that companies are pretty much always going to put their wares in as favorable a light as possible. All I ask is that they simultaneously provide enough data so as to make objective comparisons easy. And NVIDIA did that, for the most part.

 

[IntoxicatedPuma]

I thought Polaris and Vega will be at GTX 950/960 prices? The Nano still seems very competitive and doesn't have a crazy high TDP. I guess it will be rebadged as an R9 490 along with the Fury.

 

[PNP]

Vega is the Fury's successor and the Polaris(es) will be the R9 4xx. Dunno if the current R9s will become the next R7s though.

 

[PlayfulPhoenix]

The Nano still seems very competitive and doesn't have a crazy high TDP.

I'm certainly expecting that to be the price range they aim for, but AMD's been telling folks to anticipate "affordable VR", so who knows. AMD tends to be the one that pushes prices down, and they may announce cuts as a part of introducing hardware as well. There's certainly precedent, though I have no idea what's up their sleeves.

Regarding the Nano, IIRC that's a MSRP of $649. A cursory Newegg search seems to indicate that you can get it for $500 now, but it's ~7.5 TFLOPS instead of 9, and you get half the memory. And the 1080 will undoubtedly overclock better, it's just a matter of degree... Which isn't all to say that it's a bad deal - you're saving $100 bucks comparatively, after all, and it's got the shorter form factor - just that the 1080's a tad ahead of the curve in perf-per-dollar terms, especially when you consider the available VRAM.

 

[IntoxicatedPuma]

All of the rumors have been that Polaris is R9 370/380 replacement. Vega and higher level cards won't come until later.

I saw the XFX Nano for $450 on amazon., that's putting it in 1070 range. It may still have half the memory vs 1070 but a lot more bandwidth to use it and a lot faster memory. I'd say the Fury X will be more comparable to the 1080.

 

[Phuncz]

If the GTX 1080 is indeed much faster than a Titan X or GTX 980Ti, it's more within the performance of the Radeon Duo Pro.

 

[GuilleAcoustic]

I just need a decent and as quiet as possible blower type cooling system.