Pascal 1080 Benchmarks

[biostud]

So basically 30% faster than 980Ti both standard clock and in OC.

 

[Silverforce11]

So basically 30% faster than 980Ti both standard clock and in OC.

No.

~25% above stock 980Ti.

OC v OC, gap is smaller, due to higher OC % (~20% to 25%) on the 980Ti vs the Reference "Founder's Ed" 1080 which has a ~11% OC headroom only.

 

[coercitiv]

JayZ just summed it up: NV isn't officially supporting 3/4 way, but they allow enthusiasts to unlock it with a special code.

Oh boy, we get both Founder's Edition and Enthusiast's Key with this launch!

 

[tential]

So basically 30% faster than 980Ti both standard clock and in OC.

Depends on the review suite/games you're looking at. Anywhere from 25-40% faster.

 

[Face2Face]

I don't think power delivery is an issue with overclocking this card. I think the cooler and NVIDIA's Greenlight program is.

Adding to that, it appears overclocking Pascal is kind of a pain as well. From Guru3D

The new overclocking features that Nvidia introduced are a bit of a mess. They require a specific driver and the tools available crash more than they work. We went back to the old way, we'll be adding a bit of extra voltage which allows a 1.1V stage, but then the GPU keeps throttling that number down. Very difficult and, if anything, tweaking and overclocking has become more complicated starting with Pascal.

We applied the following settings:

Temp Target 95 Degrees C
CPU clock +200 MHz
Mem clock +500 MHz
Voltage up-to 1.1v
FAN RPM 60 %

The Boost clock will now render at roughly 1950~2050 MHz depending on the power and temperature signature. The GPU will continuously be dynamically altered on voltage and clock frequency to match the power and temperature targets versus the increased core clock. In FireStrike we are now hovering at the 2 GHz marker on the Boost frequency for example, but some games jumped to roughly 2.1 GHz one second and dipped below 2 GHz the other

It's odd he mentions Power, but doesn't mention if there is an adjustment for Power limit... Do we have an adjustment for that? If we can't adjust the power limit, then we're stuck with a 180w card, which is very limiting for overclocking.

 

[AtenRa]

Yeah and there's nothing wrong with TPU's review, it's accurate because they have a skew towards newer games (a good thing IMO), whereas other sites have a lot of older titles.

You'll notice sites with older games, the gap is smaller, 20% or so, lots of new games, gap grows.

Pascal blows away Maxwell when it comes to GCN-optimize console ports that we've seen. The trend will grow, compute heavy games will gimp Maxwell while Pascal's new features allow it to steam onwards.

Really have a look again at those Games used by TPU, the other sites used more newer games.

From 16 games on the TPU benchmark list, 10 are Gameworks, 5 are neutral and only a single one is AMD Gaming Evolved and this one was tested with DX-11. :sneaky:

From TPU.

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/18.html

Hitman 2016

The game uses an in-house game engine by IO Interactive, that is one of the first to leverage DirectX 12. The game's DirectX 12 implementation, however, is too riddled with bugs to be integrated into our test-bench for now. In this test we're testing the game in its DirectX 11 mode, which is still reasonably taxing on high-end GPUs.

Rise of the Tomb Raider

Based on an advanced Crystal Dynamics Engine, the game takes advantage of DirectX 11 and DirectX 12. We are using DirectX 12 for our testing.

I guess DX-12 performance regression was fixed in Rise of the Tomb Rider ??? nope, still there :whiste:

 

[Spanners]

I don't think power delivery is an issue with overclocking this card, I think the cooler and NVIDIA's Greenlight program is.

Adding to that, it appears overclocking Pascal is kind of a pain as well. From Guru3D



It's odd he mentions Power, but doesn't mention if there is an adjustment for Power limit... Do we have an adjustment for that? If we can't adjust the power limit, then we're stuck with a 180w card, which is very limiting for overclocking.

Other reviews I've read said that you can adjust up to 120%.

 

[Silverforce11]

@Face2Face

The cooler can be resolved with a fan ramping up to 100%, so it's not thermals. It's TDP related. 5 phase PCB is good to 200W. After that, it's no guarantee and depending on the mosfet quality, could be risky. So ofc, it need to be locked down to avoid burnt vrms.

 

[railven]

I don't think power delivery is an issue with overclocking this card. I think the cooler and NVIDIA's Greenlight program is.

Adding to that, it appears overclocking Pascal is kind of a pain as well. From Guru3D



It's odd he mentions Power, but doesn't mention if there is an adjustment for Power limit... Do we have an adjustment for that? If we can't adjust the power limit, then we're stuck with a 180w card, which is very limiting for overclocking.

Sounds like teething issues with GPU Boost 3.0.

Does MSI AB even support it properly, yet?

 

[caswow]

7970 reloaded but with a even more outrageous price. hardwarecanucks gave this card even a value award. if this doesnt scream "we got bought out"...

 

[Silverforce11]

https://youtu.be/RqK4xGimR7A?t=14m12s

^ Ashes DX11 vs 12 comparison, the only title with moderate Async Compute usage.

Very surprised to see the 1080 regressing by a margin similar to Maxwell.

What the heck is going on here, I was expecting the new fast context switching to minimize the impact of non-Async Compute hardware.

Hmm.. I suspect their software scheduler has overhead for handling Async Compute. Oh dear, NV's being very generous with the truth here putting Async Compute slides for Pascal... or perhaps this driver hasn't activated it yet.

 

[Grooveriding]

https://youtu.be/RqK4xGimR7A?t=14m12s

^ Ashes DX11 vs 12 comparison, the only title with moderate Async Compute usage.

Very surprised to see the 1080 regressing by a margin similar to Maxwell.

What the heck is going on here, I was expecting the new fast context switching to minimize the impact of non-Async Compute hardware.

Hmm.. I suspect their software scheduler has overhead for handling Async Compute.

Any benches of gtx1080 in DX12 Hitman available ?

 

[ioni]

Damn, I was hoping 1070 info would be out today too. We still don't even know if it's 8GB, right?

 

[Spanners]

Sounds like teething issues with GPU Boost 3.0.

Does MSI AB even support it properly, yet?

The reviews I've seen have been using Precision XOC (Precision X), must be new version as the current one on their website doesn't mention Pascal support.

Edit: Seems to be version X16 6.0.4 which they are going to release to the public soon.

 

[Silverforce11]

Any benches of gtx1080 in DX12 Hitman available ?

It's in the same video. It regresses at 4K (GPU bottlencked), while lower resolution is CPU limited and so it gains a small perf, which is a good sign that the Hitman devs fixed DX12 optimizations for Pascal.

 

[AtenRa]

Also very interesting,

What the hell, GTX980Ti is gaining this much in DX-12 in Ashes vs DX-11 ???

http://www.hardwarecanucks.com/foru.../72619-nvidia-geforce-gtx-1080-review-10.html

 

[Silverforce11]

Damn, I was hoping 1070 info would be out today too. We still don't even know if it's 8GB, right?

It's 8GB GDDR5.

We don't know the Shader, ROPs, TMU and whether it's real 256 bit bus or fake 224 + 32 bit MC like the 970. Keep and eye for that. If the 1070 has chopped ROPs, it will have a neutered MC partition just like the 970 Maxwell.

 

[Borealis7]

Damn, I was hoping 1070 info would be out today too. We still don't even know if it's 8GB, right?

Guru3D posts 1070 as 8GB GDDR5 (non X)

EDIT: SilverForce beat me to it!

 

[Silverforce11]

Also very interesting,

What the hell, GTX980Ti is gaining this much in DX-12 in Ashes vs DX-11 ???

http://www.hardwarecanucks.com/foru.../72619-nvidia-geforce-gtx-1080-review-10.html

Calling BS on HWC, those guys are now just a laugh. Their numbers are made up. They think Pasal does 400mhz + OC.

Plus, I linked the video from Digital Foundry at least you can see what's going on and it regresses badly in Ashes in DX12.

 

[railven]

Also very interesting,

What the hell, GTX980Ti is gaining this much in DX-12 in Ashes vs DX-11 ???

"never attribute to malice that which is adequately explained by stupidity"

 

[zentan]

An average ~37% more performance over 980ti at 4k and 1440p according to TPU test suite. Tomshardware gets around 34% at 4k. Still waiting for a full review by AT and also custom cards reviews.

 

[Spanners]

"never attribute to malice that which is adequately explained by stupidity"

Hanlon's razor always delivers.

 

[Dribble]

https://youtu.be/RqK4xGimR7A?t=14m12s

^ Ashes DX11 vs 12 comparison, the only title with moderate Async Compute usage.

Very surprised to see the 1080 regressing by a margin similar to Maxwell.

What the heck is going on here, I was expecting the new fast context switching to minimize the impact of non-Async Compute hardware.

Hmm.. I suspect their software scheduler has overhead for handling Async Compute. Oh dear, NV's being very generous with the truth here putting Async Compute slides for Pascal... or perhaps this driver hasn't activated it yet.

I thought AS and other DX12 settings had to be hand coded for each card by the dev to maximise performance (it can no longer be done by gpu maker in drivers), one of DX12's issues - it probably hasn't been for the 1080, it's probably using settings optimised for some Maxwell card.

On a side note performance is very impressive for a 314mm2 chip, considering they are bound to build a chip twice this size in the future (they already have done with the P100) that's a lot of future potential. For now Nvidia is going to raking it in being able to top the performance charts with a fairly simple, low power 314mm2 gpu.

 

[96Firebird]

Every video card launch we have people shifting their opinions on which review sites are paid off...

Maybe you guys should just stop reading reviews.

 

[Face2Face]

@Face2Face

The cooler can be resolved with a fan ramping up to 100%, so it's not thermals. It's TDP related. 5 phase PCB is good to 200W. After that, it's no guarantee and depending on the mosfet quality, could be risky. So ofc, it need to be locked down to avoid burnt vrms.

Fair point on the cooler, but how do we know that this card's 5+1 phase arrangement is only good for 200w? Like you said, it's definitely a TDP/Power limit issue, but I think the stock phases can give more then what NVIDIA is allowing.

I think we'll have to wait for SKYN3T to unlock the power limit and put it under water.


You guys were right - I found this on HWC - Looks like 120% PL