Unigine DX11 Benchmark

[Origin_Unknown]

BTW, is any1 else thinking the same as me, Fallout 4 or TES 5? I bet Bethesda is heavily concidering implementing a wide tesselation support into their future engine. These games if anything would benefit from it and would be in prime position to make use of the DX11 features IMO and prolly shock us again graphical wise like Oblivion did when it first arrived (I remember running the game on a 6800GT wasn't the best possible experience with fps drops to ~25 out in the wild ).

indeed - i cant wait for the next installment of TES!

[zalbard]

indeed - i cant wait for the next installment of TES!

Damn TES games take too long to develop... And the world keeps shrinking too, it seems.

[Asylum1]

Heres my run with everthing maxed out and a GTX285.
Not to bad for a single 285.

[initialised]

Here's mine:

Q6600 @ 3.6GHz (450x8), 4GB DDR2 4-4-4-12@900MHz, 4870x2@775/1100



In DX10 mode I got 53.7 (+0.2) so it looks like on DX10.1 hardware there is no difference in performance. DX9 mode gave 61.7 (10%) so here is the rub, adding DX11 gives a significant performance hit to your game engine and may increase costs if XP compatibility is kept (extra compilation for DX9 on XP vs single for DX9,10,11 on Vista/7) so either we will see adoption of DX11 with no XP compatibility or continued XP support with DX9.

However, I also ran it on a 3.66GHz i7, 12GB 9-9-9-24@1.4GHz, 5870 and 3.6GHz i7, 12GHz 9-9-9-24@1.6GHz, 2x5870 getting 47 and 79.5 respectively at 1280x1024 @ default IQ settings. The difference from tessellation really stands out, but running it in DX10 on the CF rig got 112.3fps so it is a significant performance hit which might be enough to make the lower end cards unable to exploit it without trading off resolution and IQ.

I had problems getting it to run on Vista in DX11 with a pair of 5870s, I had the Aug2009 patch installed but it still didn't play.

[andressergio]

here is mine just as it comes...awesome images

my Q9650 @ 4500/1200 GTX285 800/1800/3024




Cheers
Sergio

[xXlAinXx]

Here is a video comparison of the benchmark with and without tessellation: ---> http://www.youtube.com/watch?v=bkKtY2G3FbU

Im sorry for being a bit harsh, however there no geometry patches and/or instanced tessellation involved on the DX10 (tessOFF) benchmark/video.. therefore, it show all and nothing as u might notice there no landscape.. so, a pretty good smoke and mirror code.

Reference:
Instanced Tessellation in DirectX10
Approximating Catmull-Clark Subdivision Surfaces with Bicubic Patches

[Stukov]

Yes and yes.

The catch is the tesselation hardware previous to 57/58/59 is non-programmable I think... or something to that effect which makes it more "fixed function" so a few DX11 calls dont work and its just not worth going back to make DX10 hardware emulate those last few bits for a decrease in performance.

Does tesselation work on this benchmark?

[So3oL4Nu]

And one from me.

e8600 @ 9x475FSB (4275mhz)
2x2 Mushkin @ 950mhz 5-5-5-12
EVGA 260 216 core @ 760/1560/1290

I will rise the CPU and Memory when i receive my 780i from warranty. I should get around 45frames


Regards,
Sobo.

[donvitobg]

Everything maxed out, 5850@1000/1220, Q6600@3.9ghz.

[jmke]

this benchmark is funny
as there is no default setting

[initialised]

this benchmark is funny
as there is no default setting

This is the default:



But running in windowed mode makes no sense.

[initialised]

Would be interesting to see a performance hit with/without DX11 features.

Without DX11 hardware there is no performance hit, with DX11 hardware the hit is around 30%

[zalbard]

But running in windowed mode makes no sense.

For this resolution it does. Not like anyone uses such tiny monitors for benching any more!

[Bootup05]

Specs in sig.

Run benchy at default settings

[Manicdan]

not to sound dumb, but how do you guys run in dx11 mode without a dx11 card, and what does it look like

[orangekiwii]

And one from me.

e8600 @ 9x475FSB (4275mhz)
2x2 Mushkin @ 950mhz 5-5-5-12
EVGA 260 216 core @ 760/1560/1290

I will rise the CPU and Memory when i receive my 780i from warranty. I should get around 45frames


Regards,
Sobo.

Ok, thats not fair -.-

My gtx 285 only gets 43 fps vs your gtx 260 with 40

I wish I could over clock my card even a little but it instantly crashes


I think the tim wore out on my card as on totally stock settings... and any games it gets well over 100c and on 100% fan speed it gets up to 90c

[NaMcO]

Everything maxed out, CPU clock is 4200 and not 4000.

[orangekiwii]

It just uses the dx11 path, even though it's not doing tessalation. It looks no different than dx10, and I get the same performance between the two. It's probably because I don't have a dx11 card.

[NaMcO]

I noticed heavy slowdown on a 5850 yesterday as i switched tesselation on, didn't they claim impact on performance was near zero?

[orangekiwii]

It probably will be near zero in games. This benchmark uses excessive tessalation. Theres no way a game could pass that off. The stairs aren't even stairs. It's using tessalation to do simple stuff, so the performance hit is massive compared to games in which it will be negligible since most of that geometry will already be there, it will just add more depth and detail.

The stairs are a prime example. So's the roof.

[NaMcO]

Yeah, that is correct. Excessive usage leads to slowdown

[Lightman]

I think 1.4GHz HD5870 and this bench will fly @1920x1200 noAA

I love how it looks, but obviously those with CF setups need to have something to stress their cards properly which this bench is doing!
That's the reason for excessive tesselation in it...

[Farinorco]

I noticed heavy slowdown on a 5850 yesterday as i switched tesselation on, didn't they claim impact on performance was near zero?

I'd interpret it in other way:

What would be the performance like if all that geometry was actually fixed geometry and not tesselated from more simple meshes with this technique?

IMHO, tesselation doesn't exactly have a performance cost, because what it is actually doing is reducing the cost of rendering complex geometry (with high polygon count). So what it gives is a performance improvement, in the end...

Of course, you can then use it to render scenes with a complexity you could not do without tesselation, and continue adding more, and more, and more polygons to the scene until you bring any graphic hardware you want to its knees... but that doesn't mean that the technique per se has any performance cost.

You could use tesselation to keep the same or similar level of detail with an increased performance, to get a higher level of detail with the same performance, or to accept a performance penalty to go even further in detail... that's up to the programmer... and it seems like in this demo, the Unigine developers have gone through the path of "the more bling we can do to make it the more shocking we can".

[kl0012]

I noticed heavy slowdown on a 5850 yesterday as i switched tesselation on, didn't they claim impact on performance was near zero?

Do you noticed any image quality differences between DX10 and DX11?

[initialised]

Do you noticed any image quality differences between DX10 and DX11?

That's kind of the point of the project but you wont see a difference unless you have DX11 hardware.