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]

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.

Actually I'd heavily disagree there. The Nano costs $499 USD and offers similar performance to the GTX 980 at a TDP of 175W.
The GTX 1070 will be launched at $380 USD from partners, so even if an ITX version requires more layers or something, it will probably stay below $450 USD. For that, you get higher performance than a Titan X at an unknown TDP somewhere below 180W.

The R9 Nano isn't only not competitive at this price point, it is practically dead. If an ITX 1070 is released (which we don't know will happen but seems probable), it will offer much higher performance for less money and less power draw. There would be no reason whatsoever to use the R9 Nano, apart from the power connector in the front maybe.

AMD will have to cut the price on the Nano quite significantly to something in the range of 399$, maybe less.

I don't particularly like AMDs strategy of rebadging GPUs when a new generation is released, it's just confusing to consumers and doesn't serve any purpose other than artificially bloating their "current-gen" lineup.

 

[iFreilicht]

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.

Of course, and I don't think the fan curves will be aimed at getting the temps down so far in the production units, but it sure is interesting to overclockers who will replace the stock cooler with a loop anyway.

 

[PlayfulPhoenix]

Of course, and I don't think the fan curves will be aimed at getting the temps down so far in the production units, but it sure is interesting to overclockers who will replace the stock cooler with a loop anyway.

*raises hand*

Literally waiting for OC benchmarks with bated breath. Remember the 980?

Just imagine a repeat of that

 

[PlayfulPhoenix]

....

Generally agree with all of this. I compared the Nano to the 1080 because I was recalling the launch MSRP, but you're right that a comparison to the 1070 is more apt.

To me, the 4GB of VRAM is the elephant in the room. 4GB is really the bare minimum at the performance level of the Nano today, and that means that games moving forward are going to make it a real bottleneck to performance pretty soon. It doesn't really matter how fast that memory is if you can't even store all the textures you need on it, frankly.

I wouldn't go so far to say that it's dead - the 1070 doesn't land for a month, anyways - but the "right" price point for it dropped quite a bit on Friday, that's for sure.

 

[Phuncz]

I merged the two Nvidia GTX 10x0 topics because they are basically the same.

 

[CC Ricers]

So here's something that I've been thinking about now. We know (well, according to NVidia's claims) that the 1070 can still beat the Titan X in performance. Some of us run low-wattage CPUs. Would low-wattage i3's and i5's start becoming a huge bottleneck for gaming performance with the 10 series? Is it gonna be a waste of power situation where the 900's would still suffice?

I'm not expecting to jump on a 1070 any time soon. In fact I'm planning a modest upgrade from GTX 950 to a 960 with a Core i3 4130T. For the 10 series though, I wonder if I can still do fine with just the 4130T or also move up to Broadwell CPUs.

 

[iFreilicht]

I merged the two Nvidia GTX 10x0 topics because they are basically the same.

Holy shit you can merge topics with XenForo?! I wasn't aware.

Would low-wattage i3's and i5's start becoming a huge bottleneck for gaming performance with the 10 series?

I wouldn't be terribly concerned. The CPU is responsible for physics and logic, which are time-critical tasks but not influenced by the graphical fidelity of the game. And with 8GB of VRAM instead of 4GB, you'll be able to preload a lot of textures which reduces the dependency on the CPU to instantly provide missing ones. I think as long as your current rigs CPU is not clinbing over 60% load during gaming, it shouldn't be an issue at all.

 

[Phuncz]

Holy shit you can merge topics with XenForo?! I wasn't aware.

Yep, although I needed to look for it. You have to tag two threads/topics in the sub forum and than merge them. But I first had to move one of them for this to work, because one was in another sub forum.

 

[IntoxicatedPuma]

I don't see how 8gb of RAM is going to make a big difference over 4GB on a 1070, it's running GDDR5 only through a 256 bit bus. It sounds a bit of a waste, like the 960 4GB. Maybe in SLI with the new bridge this will give it a boost.

I will be surprised if this 1070 performs that much better than a 980 at 4k - which doesn't perform better than a Nano at 4k. Nvidia pointed out that this superior performance is in VR and using their services.

 

[PlayfulPhoenix]

At last, some clarification (kudos to confusis for mentioning a post by @jtrias21 on Instagram, which cites the below):

• Every single instance of “Founder's Edition” can be replaced with the word “Reference,” using previous-gen nomenclature. There is not one difference in its market positioning. They are synonymous. NVidia has replaced its “Reference” name with “Founder's Edition.”
  
  
• There are not two GTX 1080 models made by nVidia. Only the “Founder's Edition” exists; there is not a cheaper card made by nVidia than the $700 Founder's Edition, which ships first.
  
  
• Just to be clear: nVidia is making one official GTX 1080 and one official GTX 1070 model.
  
  
• The “Founder's Edition” is not specially binned.
  
  
• The “Founder's Edition” is not pre-overclocked.
  
  
• The “Founder's Edition” uses the new industrial design and cooler from nVidia. Historically, this is what we would call the “reference cooler.” The cooler is more-or-less identical to the previous reference models. It's got vapor chamber cooling, a VRM blower fan, and a large alloy heatsink under the shroud. There is a backplate on the GTX 1080 Founder's Edition.
  
  
• This card is not "limited edition," despite its name that would indicate as much, and will run production through the life of the GTX 1080 product line.

So indeed, it's basically $700 for a GTX 1080 with NVIDIA's fancy cooler, and then "non-reference" cards will start at $600. The silicon between the two isn't different, at least in terms of NVIDIA's specs and chips. And that makes sense, given that although I was fearful that there was some binning going on, NVIDIA would have probably seriously alienated their partners had they done that. Not doing so, and then charging $100 for the fancy cooler, likely help them not compete directly with the likes of EVGA and co.

My only concern at this point is getting a blocked 1080 without having to spring for the Founder's Edition I'm on the fence about buying, let alone building, but if I do I'd rather not have to pay $100 for a cooler I'm not using.

 

[EdZ]

One slight difference: non-Nvidia (i.e. OEM) cards can still use the reference PCB design (though are not required to), they just do not use the reference cooler.

 

[IntoxicatedPuma]

Some reports/benchmarks are showing that the 1080 is performing about 0-10% better than a 980 ti. In other words, about the same or slightly better than a FuryX. Nvidia waved that 2x 980 number around alot but I think people should have paid attention to the part where they say it's while in VR using Nvidias new optimization. I will be interested to see more benchmarks in the next few weeks and find out how it really performs.

In any case, I spent 3 hours trying to update the driver's for my R9 370 and still haven't been able to successfully complete it, so if AMD can't get a working app to download drivers then it won't matter how well the FuryX performs.

 

[PlayfulPhoenix]

Some reports/benchmarks are showing that the 1080 is performing about 0-10% better than a 980 ti. In other words, about the same or slightly better than a FuryX. Nvidia waved that 2x 980 number around alot but I think people should have paid attention to the part where they say it's while in VR using Nvidias new optimization. I will be interested to see more benchmarks in the next few weeks and find out how it really performs.

Can you link to these? That sounds very dubious considering the following (which are known or claimed TFLOPS by NVIDIA):

GTX 980: 4.61 TFLOPS
GTX 980 Ti: 5.63 TFLOPS
GTX 1080: "9" TFLOPS (I'm gunna assume that they've rounded this, so it's probably a bit more or less)

Raw compute performance is basically the best objective measure of relative performance, so claiming the 1080 as being as fast as 980 SLI generally seems reasonable (if not a bit of a stretch) when considering the fact that multi-GPU configurations are inherently inefficient. This would also lead us to expect the 1080 to be ~60% higher performing than the 980 Ti.

Granted, NVIDIA's own limited real-world examples show that this isn't consistent:

...so we'll see what composite real-world benchmarks ultimately determine the real benefit to be. It could be that NVIDIA's going with 2X because the average when including VR among non-VR uses nudges the average up to that figure, but who knows.

To be clear, though, they did not qualify the "1080 > 980 SLI" statement in their presentation with the VR stuff. Although the initial y-axis on the chart they used was incredibly stupid, they were speaking to relative game performance, period, not specific to VR and not "cheating" by incorporating new technologies in the comparison (or at least, not across the board):

We know this because it was this stupid chart where they did exaggerate by qualifying towards "VR Gaming Performance" (see the y-axis again):

In any case, given the limited data and ambiguity with respect to what "Relative Gaming Performance" even is, I'd stick with the objective and unadulterated TFLOPS figures to compare performance until we have full reviews that provide a thorough analysis. Those suggest that the GTX 1080 should be a touch below SLI 980's, and around 60% faster than the 980 Ti.

 

[IntoxicatedPuma]

http://videocardz.com/59725/nvidia-gtx-1080-polaris-10-11-directx12-benchmarks

http://www.overclock3d.net/articles/gpu_displays/gtx_1080_ashes_of_the_singularity_benchmarks/1

they're using the same source i think and it's only for Ashes of Singularity.

I am hoping Nvidia is right and it's at least 1.5x better than a 980, and a 1070 is at least 980 ti level of performance. I am skeptical that it will be much better but still hopeful I am also hopeful AMD can release an R9 490 that truly matches 980 ti for $300, that means the 1070 will need to come down in price fast if it's not as fast as they say!

 

[PlayfulPhoenix]

...

Nice, thanks for sharing! These certainly aren't very exciting, but this is a sample size of one game, and a "leaked" sample at that. So we can't really draw conclusions until we have official benchmarks from lots of folks, covering lots of games and tests.

I also recall that NVIDIA's cards have pretty consistently benched a lot worse than AMD's on DirectX 12 titles (probably a driver/optimization thing?), but I don't really know enough about that to say whether or not it's a factor in this instance. Perhaps someone else can illuminate that a bit.

 

[IntoxicatedPuma]

Agreed. I'm eager to see some more benchmarks. At the very least, it's exciting to see that Nvidia is ahead of the game for VR, and from what they described about reducing unneeded pixels when using the distorted image, i think they're going to deliver in terms of VR performance.

After seeing so many GTX 980 ti special editions release in the last few months, I can't help but think a lot of people will be royally pissed if they were sold a totally obsolete card at such a high price that was released only months before a brand new model (if the 1070 is really superior to it). If the 980 ti is at least competitive with the 1080 in non VR gaming, then I think they can at least tell themselves that it's still a good card.

I'm mostly excited about Ansel though, that looked awesome. Nvidia is doing good work even if the 1080 isn't 2x a 980.

 

[K888D]

It looks like AMD have responded to NVidia's recent 1070/1080 reveal by bringing forward the launch of Vega10 (with HBM2 memory)from early 2017 to October 2016.

It is believed that Vega10 will compete with the 1070/1080 market sector, whereas the upcoming Polaris will compete in the low to mid range market.

With regards to benchmarks - I believe that Microsoft were inspired by AMD's Mantle API with DirectX12, so perhaps titles that are geared towards DirectX12 may perform more efficiently on AMD hardware?

Also, if benchmarks are being carried out on new graphics hardware, would NVidia have supplied the testers with drivers that take full advantage of the new hardware/technology? I'm not sure how it all works, but if they are using existing public drivers then I would think the new cards would not be running efficiently.

 

[iFreilicht]

It looks like AMD have responded to NVidia's recent 1070/1080 reveal by bringing forward the launch of Vega10 (with HBM2 memory)from early 2017 to October 2016.

Source?

 

[IntoxicatedPuma]

I saw several articles today about Vega releasing in October as well.

 

[K888D]

http://hexus.net/tech/news/graphics/92747-amd-vega-launch-brought-forward-october-says-report/

Its not been fully confirmed yet.

The Polaris architecture may only compete at the mid to low end of the market, but the 14nm could mean even better fps per watt performance than what NVidia have just announced based on 16nm, can't wait for the first reviews and comparisons to start coming in once both sides have products on the market.