Anybody else unimpressed with new midrange Nvidia GPUs, and much higher MSRP?

[antihelten]

VR SM Projection, it's likely compute based, and their version of True Audio, also compute based calculations for sounds.

As far as I can tell multi-projection just sounds like a slightly fancier version of multi-resolution rendering which was present on Maxwell, so I don't really think there's anything stopping Nvidia from implementing it for Maxwell as well (other than the fact that Nvidia obviously want people to upgrade to Pascal).

 

[EightySix Four]

AMD has signal it's plan since the start of the year.

Vega is not due til Q4 2016 or Q1 2017.

Polaris 11 is entry. Polaris 10 is mainstream. Both small chips.

They aren't going to compete with 1080 and 1070 because it's in a separate market segment in terms of price, power usage, performance profile.

NV can sell the 1070 and 1080 at whatever price they deem is best for profits until Vega arrives. Even then, who here expects Vega w/ HBM2 to be cheap? Nobody should.

This is why I said get used to the new expensive mid-range.

Here's hoping Polaris 10 surprises and is mainstream the way it used to be, a reasonably priced card with midrange performance, and drags 1070 prices down too.

 

[swilli89]

AMD has signal it's plan since the start of the year.

Vega is not due til Q4 2016 or Q1 2017.

Polaris 11 is entry. Polaris 10 is mainstream. Both small chips.

They aren't going to compete with 1080 and 1070 because it's in a separate market segment in terms of price, power usage, performance profile.

NV can sell the 1070 and 1080 at whatever price they deem is best for profits until Vega arrives. Even then, who here expects Vega w/ HBM2 to be cheap? Nobody should.

This is why I said get used to the new expensive mid-range.

Based on what I saw here, AMD is going to be back in the lead as far as perf/w and perf/mm2. Its too bad, by most accounts, a Vega chip about the size of the GP104 would probably dominate.

 

[sontin]

As far as I can tell multi-projection just sounds like a slightly fancier version of multi-resolution rendering which was present on Maxwell, so I don't really think there's anything stopping Nvidia from implementing it for Maxwell as well (other than the fact that Nvidia obviously want people to upgrade to Pascal).

It is not the same:

Lens Matched Shading
Improving on Multi-Res technology this technique allows for non linear pixel density within a render target. This can be tuned to match the final output display properties of a Head Mounted Display (HMD). Scale values foreshorten the scene continuously so the viewport provides a gradual change in sampling. Fewer viewports are used, improving efficiency. The result is a simpler solution than Multi-Res shading which provides higher performance.

https://developer.nvidia.com/introducing-nvidia-geforce-gtx-1080

 

[renderstate]

VR SM Projection, it's likely compute based, and their version of True Audio, also compute based calculations for sounds.

Vertex projection happens in the vertex shader and they are talking about single pass stereo rendering. This is clearly gfx pipe lingo, compute has nothing to do with it.

 

[Azix]

Is this a paper launch? Is it a launch? or just an announcement? Specs are out so its something. We still know not a thing for sure about whats in polaris.

It sounds like polaris 10 should beat a 1070 at least. if it overclocks well it could beat a 1080. The key here is that the same dx12 advantage AMD has might still apply. If pascal is too much like maxwell polaris 10 could well give a 1080 some trouble in dx12. Its possible they went further in enabling compute and graphics to run in parallel.

VR SM Projection, it's likely compute based, and their version of True Audio, also compute based calculations for sounds.

It would be impossible on maxwell only if they use a special chip for it. It would just run on maxwell with a performance hit otherwise. remember compute still runs sequentially. My guess is it will run the same on pascal as on maxwell because they really haven't said anything that suggests pascal is a major improvement.

oh, above post suggests its not compute.

 

[EliteRetard]

Here's hoping Polaris 10 surprises and is mainstream the way it used to be, a reasonably priced card with midrange performance, and drags 1070 prices down too.

I really hope AMD can pull a big win and smash Nvidias faces in.
I'm tired of this crap. Not saying AMD can't/wont have a high priced $700 card, but I sincerely hope they will release their midrange cards at $250-400 that will compete with these Nvidia "equivalents"...and offer their true high end card/s around the $5-700 ranges that will blow the doors off this 1080.

 

[happy medium]

Based on what I saw here, AMD is going to be back in the lead as far as perf/w and perf/mm2

Unless like Fiji vs the gtx980ti ,AMD has the Polaris 10 back in the shop raising clocks to get close to the 1070?

One things for sure they have nothing to compete with the gtx1080's performance.
I thought thats where cards make the most money?

 

[antihelten]

It is not the same:

https://developer.nvidia.com/introducing-nvidia-geforce-gtx-1080

As I said, a fancier version.

 

[Silverforce11]

Here's hoping Polaris 10 surprises and is mainstream the way it used to be, a reasonably priced card with midrange performance, and drags 1070 prices down too.

Based on what I saw here, AMD is going to be back in the lead as far as perf/w and perf/mm2. Its too bad, by most accounts, a Vega chip about the size of the GP104 would probably dominate.

My expectations for Polaris is still the same, above the 390X overall, but in games with more scene complexity (Hitman, lots of crowds, dense open world games etc), it will rise above Fury X for MIN FPS. How it benches will depend a lot on the games being tested as that Discard Accelerator can either be super or barely doing anything.

Pretty good at ~110W. But not enough raw performance to really justify users already on 290X/390X/980 to upgrade to. It will come down to overclock-ability.

Vega is where it's at, but I am disappointed it's coming so late though. AMD should really aim for a Q4 release of Vega, even if it's just a paper launch. Get the hype going early.

 

[renderstate]

As far as I can tell multi-projection just sounds like a slightly fancier version of multi-resolution rendering which was present on Maxwell, so I don't really think there's anything stopping Nvidia from implementing it for Maxwell as well (other than the fact that Nvidia obviously want people to upgrade to Pascal).

Maxwell allows to send the same projected vertex to multiple viewports in a single pass, but if you want to do stereo rendering (i.e. a different projection per eye) you need to transform the geometry twice on Maxwell. Apparently they can do it in a single pass on Pascal, cutting in half the number of vertices sent to the GPU.

 

[Grooveriding]

It's about what was expected. Nvidia's marketing slides say 25% better than TX, so the observed reality in reviews will probably be 20%. A 1500 core 980ti is = to stock 980 SLI, so their claims sound about right to me. I think even a stock 1080 is going to give you 10-20% over a highly clocked 980ti, plus they are going to drop driver support for Maxwell henceforth and the Maxwell cards are going to perform like garbage within 6 months.

Yeah it's another mid range as high end, this is what the new norm is apparently. We saw it with 680 and then 980 after that, now with the 1080.

The big thing I am watching for is reviews that include the recent DX12 games where Maxwell cards like the 980 are getting crushed by the 390X. We need to see how Pascal does in these situations to get a sense of whether or not they have fixed whatever is deficient when it comes to DX12 games using compute heavy tasks.

 

[antihelten]

Maxwell allows to send the same projected vertex to multiple viewports in a single pass, but if you want to do stereo rendering (i.e. a different projection per eye) you need to transform the geometry twice on Maxwell. Apparently they can do it in a single pass on Pascal, cutting in half the number of vertices sent to the GPU.

Based on this link, it sounds like Pascal still does 2 separate viewports (with all that this entails) in VR/stereo rendering, and multi projection then takes effect within each individual viewport (via what they call sub viewports)

I don't know how reliable that link is though (it's quite slim on details after all)

 

[renderstate]

Based on this link, it sounds like Pascal still does 2 separate viewports (with all that this entails) in VR/stereo rendering, and multi projection then takes effect within each individual viewport (via what they call sub viewports)

I don't know how reliable that link is though (it's quite slim on details after all)

I think they got it wrong. A viewport only allows to scale and translate geometry, so supporting 16 viewports doesn't mean you can render to 16 point of views in a single pass. IIRC even Maxwell supports rendering to up to 16 viewports in a single pass, but the projected vertex coordinates don't change, you can only swizzle them (useuful for rendering to cubemaps and for voxelization). On Pascal they can apparently transform the vertex twice for the left and the right eye in a single pass. I assume each projected vertex can still be sent to up to 16 viewports.

 

[raghu78]

So will the partner cards be available at launch for GTX 1080 and at USD 599. The founder edition at USD 699 is a rip off. If GTX 1070 matches GTX 980 Ti perf and has similar OC headroom as 980 Ti it would be a good card to have at USD 379.

 

[tviceman]

I watched everybody at the event hollering in excitement, am I the only one who groaned in disappointment?

How loud and long will you groan when Polaris 10 is 5-10% slower than the GTX 1070 for $319-349? Do you want a new card that is the fastest or a new card with performance that has already been attainable for two years?

 

[Grooveriding]

So will the partner cards be available at launch for GTX 1080 and at USD 599. The founder edition at USD 699 is a rip off. If GTX 1070 matches GTX 980 Ti perf and has similar OC headroom as 980 Ti it would be a good card to have at USD 379.

Sounds to me like this is exactly like TX launch. Availability direct from nvidia and a wait for retailers to get cards launch. We haven't even seen any AIB leaks and the launch is in 20 days ? AIBs have nothing ready is my guess.

We might see EVGA have cards available directly as well as they seem to be a special case for nvidia, getting preferential treatment.

 

[Elixer]

Is this a paper launch? Is it a launch? or just an announcement? Specs are out so its something. We still know not a thing for sure about whats in polaris.

It basically was an announcement of a launch, with specs given.
Nobody will know how many cards they will have on hand until the date specified comes.

 

[Cookie Monster]

I think the real star of the show here is the 1070. It doesnt look gimped e.g. full 8GB of memory and its roughly Titan X speeds at $379. Think these will fly off the shelves.

Now the real question is, as any pointed out how do these fair in DX12 titles.. Reviews in 2~3 weeks time.

 

[Elixer]

I think the real star of the show here is the 1070. It doesnt look gimped e.g. full 8GB of memory and its roughly Titan X speeds at $379. Think these will fly off the shelves.

How can you tell just by looking at it, that it isn't gimped?
It still a wait and see ballgame here, we don't have the full details yet, so, going by how something looks is just speculation.

 

[nenforcer]

If Pascal was in development for nearly 3 years as was said in the presentation its possible it doesn't have the GPU Context Switching / Asynchronous Compute and is just a refined Maxwell 3 on a lower manufacturing process node. No mention of FP64 Compute Capability. GP100 is the real compute card.

I think AMD has a golden opportunity here with Polaris 10 if it comes in at $349 or lower they could steal a bunch of GTX 1070 sales since it should be announced at Computex just before the 1070 hits retail.

 

[sontin]

The GTX1080 literally killed the Fury Duo Pro for everything:
Half the power for the same performance with twice the memory and much better VR support for 46% of the price.

And people are disappointed.

 

[Azix]

Finally got a complete view of what the presentation was about. On the hardware front its not very interesting. Performance should be good, but its down to much higher clocks and more shaders. They didn't talk about hardware improvements over maxwell except faster VRAM, but even with that they are still below 980ti in bandwidth, and better power which could be down to 16nm process.

ansel is nice but its not a major thing. 360 degree is nice, regular screenshots can be had a million different ways. - not that interesting

VR audio - not that interesting unless it makes it easier to do proper audio. My guess is that designers would already be creating the audio to sound right.

Multi-projection is probably the biggest thing there for VR. Not sure how great it would work for multiple monitor folks, if they even care to setup their screens like that. Not sure how special this is to pascal considering

That covers the areas that are similar to AMD's LiquidVR, but Nvidia has one feather in their cap that AMD doesn't currently support, and that's Multi-Resolution Shading. This one is something I got to preview last year, and it's another clever use of technology. Nvidia's Maxwell 2.0 architecture (sorry, GM107/GM108 and Kepler users, you're out of luck!) has a feature called viewport multicasting, or multi-projection acceleration. They first talked about this at GDC2015, but there weren't a lot of details provided; now, however, we have a concrete example of what Nvidia can do with this feature.

Normally, graphics rendering takes place from a single viewport; if you want to render from a different perspective, you have to recalculate a bunch of stuff and set things up for a new viewport, and that takes a lot of time. Multiport viewcasting allows Nvidia's GM20x architecture to do multiple viewports in a single pass, and what that means for VR is that Nvidia can do nine different scaled viewports. Why would they want to do that? Because a lot of the data normally rendered for VR gets discarded/lost during the preparation for the VR optics.

http://www.pcgamer.com/amd-liquidvr-vs-nvidia-vrworks-the-sdk-wars/2/

seems they just bumped it up to 16 from maxwell's 9.

Nothing much on the feature front at all. Just more shaders and higher clocks on the hardware front.

Also the 1070 is starting at $449 with a founders edition as well. No date on the custom cooled ones from partners. fishy. Not how things are normally done.

 

[DarkKnightDude]

The best thing out of that ironically for me is Ansel. I wonder if all gpus can use it or only Pascal, because seems like its only driver level.

 

[R0H1T]

If the single pass multi-viewport thing actually gets used in VR (not sure if that requires Nvidia VR Works) then it's a big win for the price.

I actually paid for a Radeon Nano at 500$ for VR. The 1070 looks like it might be a better deal, use the same amount of power, and will be faster too.

Not to mention it's on 16nm FF+

Not a big fan of Nvidia, so neither impressed nor unimpressed by this release but anyone wanting 2GHz stock GPU clocks must have surely been disappointed by this :hmm: