Nvidia Pascal Reviews [1080XP, 1080ti, 1080, 1070ti, 1070, 1060, 1050, and 1030]

[swaaye]

Well when you have money, you can have as much media coverage as you want. AOTS / Oxide doesn't fall into that category of corporate megadollar media overload, nor do they have that apparently popular Tomb Raider name. Tomb Raider being popular has always fascinated me because the old games sucked ass. They just relentlessly pushed out sequel after mediocre sequel though and now it's just super popular yay!

Anyway, though the charts I see make it look like AOTS is super demanding, I've found that my 970 has no problems at 2560x1440 in some crazy intense situations. RTS games just typically don't do 60 fps unless they are quite simplistic. So I would avoid the 60fps obsession with the genre. It only became possible to keep SupCom at 60 fps most of the time fairly recently.....

 

[Alessio1989]

The price for a hybrid 2012-2015 hardware (I hope the driver is wrong), with that poor PCB and stock cooling is simply insane. Only the x1950 pro did worst, but it was not a top-level GPU....

Let's wait for a sexy custom full-metal-solid PCB with a ultra power-management that would make you shout "Great Scott!". 'Cuase it is what I would expect from a 800 bucks GPU.

 

[Unknown Soldier]

I read somewhere that someone had said that the DX12 is still broken on Pascal because the tests that were done on Ashes of the Singularity in DX11 and DX12 mode had the same results so it showed that DX12 wasn't working correctly for Pascal. Can anyone confirm this?

 

[Razor1]

the frame rates are the same, or very close, but that doesn't mean its broken, nV will never show the same amount of increases from DX11 to DX12 applications, that is because AMD's DX11 drivers had issues with CPU overhead.

 

[itsmydamnation]

I love RTS's and I have to say AOTS is not a great RTS, in my opinion they should have focused more on game play and story line than their engine technology. What I feel its a great DX12 tech demo for what their engine is capable of for future RTS's on their engine.

while i agree re the game (yes i actually own it, which most bashes of it do not) , you miss the point, The "game" was there to offset the engine development cost. The point was to build a new engine and that they did. Now what do you think things like the next Sins of a solar empire etc will use.........

I have pushed insane unit counts just to see what would happen and it just doesn't slow down.........

edit: now all i need is someone to remake Red Alert 2 into the AOTS engine and im set. The other thing none of the bashers talk about is the AI, and they wouldn't because they have never played the Fing thing, The harder levels of AI actually act intelligently as opposed to just cheating...........

 

[Razor1]

I own it to, I was expecting a lot more from this game when I was playing the alpha, beta. They could have done so much more in the single player game. Oxide is an indie developer, usually the only way indie's succeed is when they make a game that is better than AAA games, they need something to offset all that marketing that AAA games have. The only place they made money from this game, is from AMD's pocket. Steam shows what, 500 active players at any given time? That would not help offset anything. But AMD's bundling the game, gave them much more though.

 

[Ryan Smith]

Hey, does anyone have (or know of) a quick and dirty CUDA 7.5 FP16 throughput benchmark? I want to quickly poke at FP16 perf on GP104, and I don't have anything like that.

 

[Razor1]

the sissandra suite I think.

http://www.sisoftware.eu/2015/11/19/sisoftware-sandra-2016-released/

GPGPU: CUDA (7.5), OpenCL (2.0, 1.2)

Under Image processing.

 

[Unknown Soldier]

the frame rates are the same, or very close, but that doesn't mean its broken, nV will never show the same amount of increases from DX11 to DX12 applications, that is because AMD's DX11 drivers had issues with CPU overhead.

Ta

 

[silent_guy]

I read somewhere that someone had said that the DX12 is still broken on Pascal because the tests that were done on Ashes of the Singularity in DX11 and DX12 mode had the same results so it showed that DX12 wasn't working correctly for Pascal.

DX12's primary reason to exist is to remove bottlenecks from the driver and to be able to load the GPU at maximum capacity. If your GPU is already fully loaded, you're not going to get any upside.

 

[Unknown Soldier]

It probably means GP104 doesn't get the same increase in registers as GP100. Not surprising but a little disappointing if true.

Compute performance vs Maxwell isn't perfectly linear but seems to scale well enough. I borrowed the other card numbers from Anandtech.

Results from my 980Ti @1400MHz GPU clock, mem clock @ def:

Face Detection -- 205 MPix/s
Optical Flow -- 39 MPix/s
Particle Simulation -- 1978 MInter/s

Here's my 980Ti results. Not sure why it's showing as Windows 8, I am running Windows 10.

 

[silent_guy]

Looking at the Hardware.fr perf/W numbers, we see the 1080 having a 2.25-2.3x ratio over Fury X.

As expected, the much heralded perf/W improvements for Polaris will simply help it to catch up with Pascal.

I don't see any reasons for optimism to expect that a Polaris based laptop design will be more efficient than one that is GP106 based.

 

[fellix]

http://videocardz.com/60151/nvidia-geforce-gtx-1080-tested-with-aftermarket-cooler

Test performed by PCHG brought a simple conclusion; it is not temperature that is keeping GP104 from achieving higher clocks, but board power limit, which can’t be increased unless more power connectors are added. So I’m guessing we need to wait 10 more days to see what custom cards can deliver in this matter.

GP104 is quite a bit gimped on the reference board design.

 

[Ryan Smith]

the sissandra suite I think.

http://www.sisoftware.eu/2015/11/19/sisoftware-sandra-2016-released/



Under Image processing.

Thanks. That does the trick.

 

[LordEC911]

Looking at the Hardware.fr perf/W numbers, we see the 1080 having a 2.25-2.3x ratio over Fury X.

As expected, the much heralded perf/W improvements for Polaris will simply help it to catch up with Pascal.

I don't see any reasons for optimism to expect that a Polaris based laptop design will be more efficient than one that is GP106 based.

That is a new tune.... people were saying it would only allow AMD to catch up to Maxwell.... and Pascal would destroy it.

 

[CarstenS]

Hey, does anyone have (or know of) a quick and dirty CUDA 7.5 FP16 throughput benchmark? I want to quickly poke at FP16 perf on GP104, and I don't have anything like that.

You don't need to. GP104 does not offer double-rate FP16.

 

[pixelio]

Hey, does anyone have (or know of) a quick and dirty CUDA 7.5 FP16 throughput benchmark? I want to quickly poke at FP16 perf on GP104, and I don't have anything like that.

Unfortunately, there isn't a good way of doing this with CUDA 7.5 or 8.0 EA since there is no support for sm_6x and previous architectures other than sm_53 only implement the conversion operations.

Assuming you have an fp16 benchmark in hand, you could try compiling for sm_53 (X1) with a "-arch compute_53" flag since that should generate PTX which _should_ get JIT'd to sm_6x at runtime.

Did anyone ever get verbal confirmation that GP104 was going to include fp16x2? I sure hope so...

 

[silent_guy]

That is a new tune.... people were saying it would only allow AMD to catch up to Maxwell.... and Pascal would destroy it.

I don't think I've seen that claim made on this forum... But we've definitely seen AMD hype Polaris as the second coming of Christ in the perf/W department.

 

[pixelio]

You don't need to. GP104 does not offer double-rate FP16.

Argh, that's super disappointing to CUDA devs. Can you reveal your source? :|

 

[fellix]

You don't need to. GP104 does not offer double-rate FP16.

Sure that? What says DX Caps Viewer?