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

[Deleted member 2197]

NVIDIA GeForce GTX 1080 On Linux: OpenGL, OpenCL, Vulkan Performance

If you looked at all of the results shown in this article, there isn't too much need for a conclusion... Many of the GeForce GTX 1080 numbers were phenomenal and represented huge improvements over the GeForce 900 "Maxwell" graphics cards and only furthering the performance advantage NVIDIA has over AMD on Linux, particularly for OpenGL when factoring in the respective company's drivers.

Not only was the GeForce GTX 1080 performing great with the raw performance in OpenGL, OpenCL, and Vulkan games and applications, but the performance-per-Watt was also nothing short of astonishing. In a few of the performance-per-Watt results, it was incredible to see how much further ahead Pascal has put NVIDIA even compared to just Maxwell cards from last year.

http://www.phoronix.com/scan.php?page=article&item=nvidia-gtx-1080&num=1

 

[egoless]

Edit: Removed

 

[Deleted member 11852]

If you were ever a RTS fan and cringed at an artificially limited amount of playable units, this picture would leave you really happy.

Sorry for O/T but this looks like no more units on screen than Total War: Warhammer, which my girlfriend is currently playing through. What am I missing?

 

[Malo]

Sorry for O/T but this looks like no more units on screen than Total War: Warhammer, which my girlfriend is currently playing through. What am I missing?

Maybe compare the CPU and GPU usage of the titles when massive amounts of units are onscreen?

 

[Deleted member 2197]

GeForce GTX 1080 2-way SLI review

Before we start please read this a HUGE note; this will be article is revision one. The early 368.25 driver does not seem to support SLI for the majority of games, hence we'll look at a handful of games in this preliminary revision of the article first, once a proper final SLI supporting driver is out, we'll add the rest okay ? So this is us jumping the gun to push some results out, not blaming Nvidia.

http://www.guru3d.com/articles_pages/geforce_gtx_1080_2_way_sli_review,1.html

 

[Silent_Buddha]

Sorry for O/T but this looks like no more units on screen than Total War: Warhammer, which my girlfriend is currently playing through. What am I missing?

There some notable differences at least at the AI level. Each individual unit in AOTS is driven by an AI. In Warhammer, each regiment (which consists of multiple units) is driven by an AI. So for example, coarse pathfinding is used for the regiment when traveling any distance of note. Within the regiment each unit will path find in the much more limited area that the regiment immediately takes up. Hence the AI load is less for Warhammer.

That may also have ramifications on rendering of each unit. IE - with regiments many of the units will have the same models, facing the same direction, doing the same animations, reusing the same shadows, shaders, etc. In AOTS each unit is doing something completely unrelated to other units even if they share the same model as other units, so everything will have to be rendered separately for each individual unit.

Regards,
SB

 

[Deleted member 11852]

There some notable differences at least at the AI level. Each individual unit in AOTS is driven by an AI. In Warhammer, each regiment (which consists of multiple units) is driven by an AI. So for example, coarse pathfinding is used for the regiment when traveling any distance of note. Within the regiment each unit will path find in the much more limited area that the regiment immediately takes up. Hence the AI load is less for Warhammer

Thank, thats helpful. And impressive!

 

[Silent_Buddha]

Thank, thats helpful. And impressive!

Yes, and I hope AOTS is something that will influence future titles to a greater extent.

Imagine if GTA V actually had a city that felt alive and full of people, each with their own individual AI (probably not anything too complex). For gameplay reasons they'd likely have to limit the number of vehicles in your vicinity when driving (trying to drive in a traffic jam wouldn't be fun). But imagine if you were flying in a helicopter or plane and looked down to see streets just bustling will traffic pedestrians...like a real city. With thousands upon thousands of people and cars. Or when walking around, there's hundreds and hundreds of people walking to and from work during the times normally associated with people going to work or coming home from work. Case out a target for a robbery during the busy times of the day when it's very crowded and your activities won't be noticed. Then execute the robbery when logically there's very little pedestrians out and about.

I think it's one of the reasons I don't like most open world games. The locales are all empty and mostly dead feeling. If open world games could have at least the capacity to have the numbers that we can see in AOTS that'd go one step towards making games more immersive. The next problem that would need to be solved would be diversity of models (Dead Rising 3 did some good things in that area). Which, come to think of it. I wonder if a Dx12 version with async compute would have performed a lot better on PC?

I'll hope that developers take advantage of Dx12 and do something along those lines, but I'm prepared to be disappointed by most games set in a city yet again.

Regards,
SB

 

[Deleted member 11852]

Imagine if GTA V actually had a city that felt alive and full of people, each with their own individual AI (probably not anything too complex). For gameplay reasons they'd likely have to limit the number of vehicles in your vicinity when driving (trying to drive in a traffic jam wouldn't be fun). But imagine if you were flying in a helicopter or plane and looked down to see streets just bustling will traffic pedestrians...like a real city.

This is close to what Ubisoft were promising with WATCH_DOGS but that city felt less alive than GTA V, which actually does a fairly good job of giving the appearance of individual AI, as long as you don't look too closely. Pull a weapon in public and some people will run, others will attack, others will hold their ground and perhaps attack if others do first. But follow a pet or car and you'll realise they don't do anywhere.

Ditto Skyrim, where NPCs have residences and places of work. We're getting there, slowly, but we're getting these.

 

[sebbbi]

Ashes snow comfirmed to be a driver bug:
https://twitter.com/dankbaker/status/739880981612625920

 

[Anarchist4000]

Driver bug seemed like a given, the real question was if it impacted performance in a meaningful way?

But imagine if you were flying in a helicopter or plane and looked down to see streets just bustling will traffic pedestrians...like a real city. Without thousands upon thousands of people and cars.

With hairworks and 64x tessellation applied to all of them!

 

[Silent_Buddha]

Ashes snow comfirmed to be a driver bug:
https://twitter.com/dankbaker/status/739880981612625920

So a bug that only existed in the drivers that the Press were explicitly going to use to benchmark the game for their reviews? Well, hopefully retesting with the publicly available 368.25 drivers won't show a huge performance discrepancy with the bug fixed. Assuming any review site bothers to retest with the fixed drivers.

I think I'm growing overly cynical in my old age.

Regards,
SB

 

[SimBy]

Performance difference, if there is one, is likely only a couple %. Anything more and it would raise too many eyebrows and unpleasant questions.

But it seems 1080 does have questionable drivers atm. Latest example is Mirror's Edge Catalyst which is supposedly unplayable on 1080 (massive stuttering).

 

[Jawed]

Wow SLI is broken in launch drivers and it's not NVidia's fault.

 

[OlegSH]

Driver bug seemed like a given, the real question was if it impacted performance in a meaningful way?

There is a review of MSI's GTX 1080 with 368.25 - http://www.guru3d.com/articles_pages/msi_geforce_gtx_1080_gaming_x_8g_review,15.html
MSI GTX1080 is actually a little bit faster than theoretical 7% from higher boost frequency, probably due to more consistent boost frequency with TwinFrozr VI cooler, anyway, I don't see any meaningful impact on perf here

 

[lanek]

If this driver correct it.. it was out on 25th may..

 

[OlegSH]

https://twitter.com/dankbaker/status/739880981612625920
368.25 is the day 1 driver

 

[CSI PC]

If this driver correct it.. it was out on 25th may..

Good catch.
So AMD was not using the right driver for their presentation.
Although timeframe was a bit tight, so giving them benefit of doubt but maybe they should had pulled that segment.
If Nvidia marketing had wanted to be nasty (nasty being they know it is probably a genuine mistake) they could had gone with a public narrative AMD deliberately used old driver and did nothing to stop the wrong conclusion being published by Stardock (where it was inferred Nvidia altered this for performance gain, emphasis inferred).
Unless the issue is still with 368.25 and Dan has wrong info, should be easy for anyone to test though.
Must be at least 10 Nvidia owners who bought this game

Cheers

 

[gamervivek]

AMD were using the right driver alright.

 

[CSI PC]

It does also raise the question where AMD got the 1080 from as well if using the reviewer drivers - I assumed it was a retail card.
That may be a bit more naughty if they borrowed a reviewer/developer card that was under some kind of NDA/sharing agreement.
Although this may have some leeway as the event was after the 1080 went on sale.
Cheers