Unigine Heaven 2.0 RC1

[96redformula]

I would love to upload it to another site, but I have 385k upload speeds or something like that, would take me too long. Sorry

[annihilat0r]

I added a Rapidshare link.

[Gorguls]

Direct Link Final Version.

http://unigine.com/download/files/Un...Heaven-2.0.msi

[Chickenfeed]

.

What I don't understand is why FPS goes down so much during zoomed in scenes... There's barely anything to render yet the card is starting to choke. Not that the stuff looks so amazing when you're too close either...

Tessellation is leveraged to dynamically adjust the detail level based on the scene. Scenes where less detail can be seen ( eg from a far ) have their triangle count reduced thus the performance increase were as scenes with high detail ( close up ) increase their triangle count thus the performance reduction. Its quite simple really. If the wide shots were rendered at the same quality as the close shots, it would be a power point presentation.

@Gorguls

Nice

[zalbard]

Direct Link Final Version.

http://unigine.com/download/files/Un...Heaven-2.0.msi

Thank you!

Tessellation is leveraged to dynamically adjust the detail level based on the scene. Scenes where less detail can be seen ( eg from a far ) have their triangle count reduced thus the performance increase were as scenes with high detail ( close up ) increase their triangle count thus the performance reduction. Its quite simple really. If the wide shots were rendered at the same quality as the close shots, it would be a power point presentation.

I know.
The thing is, even with reduced level of detail scenes taken from far have 2-3x more FPS yet look much better and more detailed than close up scenes the engine seems to stutter at...
Maybe it's just me...

[96redformula]

Whoops, seems like when media center is open Powerplay doesn't release the clocks back to 3d speeds.

Here is the corrected run with it at full clocks 785/1275

HD5850 785/1275
Athlon 555 Unlocked 4 cores 3.2ghz @ 1.25V

Benchmark Settings:
Default settings
1920x1080 Resolution

[s1nykuL]

Direct Link Final Version.

http://unigine.com/download/files/Un...Heaven-2.0.msi

Thanks

[Leopold Butters]

Average FPS: 34.9
Score: 880
Min FPS: 5.8
Max FPS: 127.8

Settings
1920x1080
noAA
MaxAF
Extreme Tess.

[kemo]

http://translate.google.com/translat...p%3Ft%3D164761

Can anyone check that? Apparently the dragon scene from 1.0 -> 2.0 improved by 40% for the 5800's...

This is also where the Fermi cards had their biggest lead in the benchmark

No need to check i know that guy

[annihilat0r]

Yep, that's right. There's definitely improvement for my 5870 with the dragon scene.

[Mungri]

This is with my Crossfire 5770s @ 1000 / 1445, no AA and 4x AF:

[Musho]

Thats not with Tesslation Extreme

The Tessalation Extreme setting isn't really a good real world comparable setting. On Extreme, tessalation is way over the top, slowing down the cards, while not giving any noticeable extra eyecandy.

[Mungri]

Its a good power comparison between different cards though.

[Musho]

Its a good power comparison between different cards though.

Power comparison for what? What it's most likely going to show, is that Fermi will run all over 5870, yet with real world usage, you don't need such excessive amounts of tessellation. Why benchmark something you aren't going to use in real world applications and games? Just for higher numbers? It looks like Fermi is only going to be a tad faster than 5870, while extreme tessellation might give an extremely skewed view.

[Mungri]

Yup, thats what benchmarks are for, better numbers.

But how can you be so sure how much tesselation games are going to use? Someone might decide to use a crazy amount of it.

[96redformula]

Link got pulled.....

[Musho]

Yup, thats what benchmarks are for, better numbers.

But how can you be so sure how much tesselation games are going to use? Someone might decide to use a crazy amount of it.

Why use a crazy amount of it if it's not going to make a difference in picture quality? Just to bring the FPS down? Just run the benchmark with the different tessellation options, and try to see the difference. It's extremely hard to spot the differences, so it isn't going to look better if you have to actually search and compare screen shots, just to try and find differences

Tessellation is not only there to make games look better. It's also there to make games run better. Instead of using multiple fixed models, swapping them as you move away/towards those models for either higher/lower detailed models, tessellation can actually increase polygon counts on the fly. This saves space, as you will only need 1 model per item you want to render, avoids seeing models pop into more/less detailed models as you move and can be easily scaled down and up to improve performance or to improve visuals. Enough polygons is enough, though. If you can't see the difference anymore, because you are already using enough polygons for all the details, using more won't actually net you a better looking model. It's just going to slow the game down. That's why I think using and benchmarking an extreme amount of tessellation isn't going to be really needed, and will only skew the results, as Nvidia's and ATI's architecture both handle it differently.

Nvidia uses it's shaders to produce tessellation. This means it can allocate more resources towards tessellation as needed. ATI uses a fixed function tessellator, however. This mean as you increase the amounts of tessellation, the tessellator will be bottlenecking the rest of the architecture. That's why it's unfair to benchmark using excessive amounts of tessellation, as it's obviously going to make nVidia look better than ATI, while it's not going to be any benefit to real world games.

ATI's tessellator does look to a bit too slow, though. I guess we'll find out if it's good enough once people start benchmarking heaven with the normal tessellation mode, metro 2033 and grid. I'm really looking forward to comparisons between Fermi and 5870!

[Mydog]

First run, CPU @4,5, 5970 and 2x 5870's @stock.

[annihilat0r]

Could you possibly have a faster rig?

[Musho]

First run, CPU @4,5, 5970 and 2x 5870's @stock.

Nice score, is the 5970 clocked to 5870 levels? Should improve performance by quite a bit I'd say.

[Mydog]

Nice score, is the 5970 clocked to 5870 levels? Should improve performance by quite a bit I'd say.

I usually run the 5970 @1000/1220 and the 2 5870 @1050/1330 with th 10.3 preview driver, but for some I just got "display drivers stopped..". Took the OC down to 5970 @950/1200 an 2x 5870 @1000/1300 and got this.

[Musho]

I usually run the 5970 @1000/1220 and the 2 5870 @1050/1330 with th 10.3 preview driver, but for some I just got "display drivers stopped..". Took the OC down to 5970 @950/1200 an 2x 5870 @1000/1300 and got this.

Wow, more than DOUBLE the min FPS by overclocking! That's very good indeed!

[96redformula]

Why use a crazy amount of it if it's not going to make a difference in picture quality? Just to bring the FPS down? Just run the benchmark with the different tessellation options, and try to see the difference. It's extremely hard to spot the differences, so it isn't going to look better if you have to actually search and compare screen shots, just to try and find differences

Tessellation is not only there to make games look better. It's also there to make games run better. Instead of using multiple fixed models, swapping them as you move away/towards those models for either higher/lower detailed models, tessellation can actually increase polygon counts on the fly. This saves space, as you will only need 1 model per item you want to render, avoids seeing models pop into more/less detailed models as you move and can be easily scaled down and up to improve performance or to improve visuals. Enough polygons is enough, though. If you can't see the difference anymore, because you are already using enough polygons for all the details, using more won't actually net you a better looking model. It's just going to slow the game down. That's why I think using and benchmarking an extreme amount of tessellation isn't going to be really needed, and will only skew the results, as Nvidia's and ATI's architecture both handle it differently.

Nvidia uses it's shaders to produce tessellation. This means it can allocate more resources towards tessellation as needed. ATI uses a fixed function tessellator, however. This mean as you increase the amounts of tessellation, the tessellator will be bottlenecking the rest of the architecture. That's why it's unfair to benchmark using excessive amounts of tessellation, as it's obviously going to make nVidia look better than ATI, while it's not going to be any benefit to real world games.

ATI's tessellator does look to a bit too slow, though. I guess we'll find out if it's good enough once people start benchmarking heaven with the normal tessellation mode, metro 2033 and grid. I'm really looking forward to comparisons between Fermi and 5870!

Great point, so why are we all using the extreme setting again????

[WeiT.235]

The official 2.0 version turns up at Techpowerup.

[Lightman]

Could you possibly have a faster rig?

He CAN play Crysis /jk

Impressive scores Mydog!