The new Heaven GPU benchmark

[UberGerbil]

What I am basically saying is I should be able to run the benchmark at a bunch of differnet resolutions and always get the same final score because my card didnt change.

I should be able to drive my car at a bunch of different speeds and always get the same miles per gallon because my car didn't change.

[danny e.]

What I am basically saying is I should be able to run the benchmark at a bunch of differnet resolutions and always get the same final score because my card didnt change.

I should be able to drive my car at a bunch of different speeds and always get the same miles per gallon because my car didn't change.

Nope. miles per gallon would be more equivalent to FPS which will vary. But your car will always have the same engine / power regardless.

I already admitted that it would be very difficult or impossible to implement accurately but if you're going to try dumb illustrations like that....
It would be more like running your car full throttle up a steep hill and only topping out at 45mph and then running it full throttle on a flat surface and getting 150mph.

[ChronoReverse]

That's not possible because as you change settings, not only does a benchmark start to bottleneck different parts of the video card, it also starts to bottleneck different parts of the computer.

[TurtlePerson2]

Looks pretty ugly in DX10 mode in comparison to the DX11 videos I've seen. Suddenly I want a 58xx to replace my GTX260

Don't fall for it! At least, not for this benchmark . I want a 5850 too, but I honestly don't have a need for it, with a GTX260 in my current system. I'll wait and see what Dragon Age, Mass Effect 2, and Starcraft 2 do to my system first.

Starcraft 2 won't push the graphics envelope because they're going for a very mass market. I don't think that it'll even be DX10.

The other two games are both coming to consoles, so I wouldn't count on demanding requirements.

[SNM]

[Mentawl]

OK!! Tried the software for giggles
&
I can't get pasted 4 fps & i'm running these components:

Amd phenom II x4 955 am 3
Asus M4A78 Plus 770/700 chipset Am 2
2 wd carvier blue 500 gib sata 2 16 mb
xfx gforce 8400gs 512 mb
Ozc Reaper HCP DDR- 2 800hmz 4 gib dual channel
windows 7 64bit


Puzzled??? Can any one help?

EDIT added ram info

Wouldn't be too surprised if you're only getting 4 FPS - that 8400GS is an incredibly slow card I'm afraid.

[Airmantharp]

Looks pretty ugly in DX10 mode in comparison to the DX11 videos I've seen. Suddenly I want a 58xx to replace my GTX260

Don't fall for it! At least, not for this benchmark . I want a 5850 too, but I honestly don't have a need for it, with a GTX260 in my current system. I'll wait and see what Dragon Age, Mass Effect 2, and Starcraft 2 do to my system first.

Starcraft 2 won't push the graphics envelope because they're going for a very mass market. I don't think that it'll even be DX10.

The other two games are both coming to consoles, so I wouldn't count on demanding requirements.

Pretty much; but I'm not sure what else I'll wind up playing besides the above, and given that, I may have no reason to upgrade beyond a GTX260 for a while. I'll probably save up for an SSD instead .

[leor]

here's dx10 and dx11



don't know how representative this though, i had loads of stuff open, photoshop, illustrator, 3 screens running etc. so the performance is probably a bit off, but you can get a comparison because i ran them one after the other.

[Fighterpilot]

Thanks for the post...that other stuff you had running probably would impact the score a fair bit I would think.
Here's how mine scored at the lower resolution.

[andebad]

5850 in DX 10

[lethal]

G80 representing

[Fighterpilot]

Its interesting seeing which way the numbers trend.Hard to set anything in stone yet but the G80 generation score just posted seems just about right compared to some of the higher end cards getting posted.
We probably need a few hundred configurations to really say whether this makes for a decent benchmark.So far the numbers are all over the place.
Its very accurate on my PC through repeated runs though, with only a point or so difference which is encouraging.

[UberGerbil]

Its very accurate on my PC through repeated runs though, with only a point or so difference which is encouraging.

That doesn't necessarily tell you anything about accuracy, just consistency. If I'm shooting at a target and put all the bullets in a tight group ten feet to the side of what I'm trying to hit, my shooting may be reliable and consistent but it's not accurate. Reproducibility of results is a good thing, but only if those results tell you something useful. After all, I could write a benchmark that simply produces a number out of a table based on your your CPU and GPU, or generated from your machine name, or something else equally arbitrary -- thus offering a reliable number that doesn't tell us anything.

Your previous comment was a good one: we need a lot more data points before we have a sense of how this benchmark depends on CPU, GPU, and whatever else. Even then, without knowing more about the code paths it uses on various hardware, it will be hard to use it to say anything conclusive about DX11 vs DX10/9 or even DX11 on different hardware.

[BobbinThreadbare]

What I am basically saying is I should be able to run the benchmark at a bunch of differnet resolutions and always get the same final score because my card didnt change.

I should be able to drive my car at a bunch of different speeds and always get the same miles per gallon because my car didn't change.

Nope. miles per gallon would be more equivalent to FPS which will vary. But your car will always have the same engine / power regardless.

I already admitted that it would be very difficult or impossible to implement accurately but if you're going to try dumb illustrations like that....
It would be more like running your car full throttle up a steep hill and only topping out at 45mph and then running it full throttle on a flat surface and getting 150mph.

Except engines produce different amounts of horsepower and torque and different engine speeds.

[Fighterpilot]

That doesn't necessarily tell you anything about accuracy, just consistency.

Fair enough Uber,however sometimes consistency has an accuracy all of its own.
As there is no established zero point in this test its probably good enough that it correctly reports the hardware capabilities of the machine.
Having the various DX APIs to test against also gives it some appeal to the majority out there who are probably still running DX9/10 hardware.

[StuG]

DX9 through DX11 - No Crossfire:

DX9


DX10


DX11

[Fighterpilot]

Nice post Stu...that gives a good idea of how the various APIs tax the GPU.
Having run comparative DX9/10 and 11...is the loss of framerate and /or smoothness of the video playback balanced by noticeably better visuals?
How do you perceive the differences to be...if any?
Your thoughts on that would be appreciated.

[SNM]

I don't have a DX11 part, but I'm guessing that in that mode the stairs don't look like a smooth ramp with texturing that makes them look like stairs from one angle...

[morphine]

That was a pretty massive hit going from DX10 to DX11. Does that benchmark enable OMG THOUSAND EFFECTS in DX11 mode? From what I read around, DX11 was supposed to enable the DX10 effects in a much less expensive way.

[StuG]

Yea, DX11 was still playable because it didn't feel choppy of anything...but it wasn't nearly as smooth as the DX10 and DX9 counterparts. DX11 though was by far the more beautiful benchmark, they used tessellation EVERYWHERE in it, the roads looked beautiful, the dragon was away more elegant, lighting was more noticeable (with seeing individual rays reflect off different things rather than DX10 which just looked like heavy HDR at times).

From this demo I do see how DX11 could be implemented to have similar features as DX10 but make it easier, there were times where nothing changed in the video and I got a better FPS (the rope scene) but typically it seems as though the benchmark was less focused on that and more focused on showing off the new things in DX11. If you do ever get a chance to see it in DX11, all the effects they added would have been TOO much for a regular game, the roads were WAY to bumpy, the roofs looked just plane strange (though very very detailed....which was the nice part), and the Dragon could have looked like it did in DX10 through textures, but they decided to make him look all elegant and spiky in DX11 just through tessellation, to really show you how it works and what it can do. So yes, the performance hit is balanced by way better looking visuals.

Aside from all of that, I wasn't nearly expecting such a big hit in performance. Though if we all remember DX10 was a big hitter until Drivers became more mature and games started implementing it better. Maybe we have that on our hands again? If any of you have any more questions feel free to post, or just plain leave your opinions

[lkn4trouble]

By lkn4trouble at 2009-11-06

[Flying Fox]

This is seriously for the lulz.



In 1280x1024 it seemed to have hung.

[Fighterpilot]

That's one sucky video card

[Flying Fox]

That's one sucky video card

It's not even a "card" if you pay closer attention.

[Airmantharp]

The [H] posted a 'review' of this benchmark yesterday. They looked closely at the effects of Tesselation at DX11, and compared DX11 speeds with DX10 speeds. It seems DX11 is measurably faster, though not by much, and Tesselation can add tremendous amounts of useful detail if implemented correctly- at a cost. Hopefully developers will be able to balance adding extra detail with tesselation with the associated performance hit.

[ark_angel]

dx9-11 on 5870 no overclocks

[Airmantharp]

When you guys post your DX11 benchmarks, would you mind specifying whether you had tesselation on or not? It seems that DX11 should be faster than DX10 sans tesselation, and much slower than everything else with it.

[StuG]

Mine had tessellation on

[JustAnEngineer]

Unigine Heaven Demo v1.0
FPS: 2.1
Scores: 53


Hardware
Binary: Windows 32bit Visual C++ 1500 Release Oct 22 2009
Operating system: Windows Vista (build 6002, Service Pack 2) 64bit
CPU model: AMD Phenom(tm) II X3 720 Processor
CPU flags: 2807MHz MMX+ 3DNow!+ SSE SSE2 SSE3 SSE4A HTT
GPU model: ATI Radeon HD 3200 Graphics 8.01.01.934 256Mb

Settings
Render: direct3d10
Mode: 1600x1200 fullscreen
Shaders: high
Textures: high
Filter: trilinear
Anisotropy: 4x
Occlusion: enabled
Refraction: enabled
Volumetric: enabled

[JustAnEngineer]

Unigine Heaven Demo v1.0
FPS: 18.6
Scores: 468

Hardware
Binary: Windows 32bit Visual C++ 1500 Release Oct 22 2009
Operating system: Windows Vista (build 6002, Service Pack 2) 64bit
CPU model: Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz
CPU flags: 2499MHz MMX SSE SSE2 SSE3 SSSE3 SSE41 HTT
GPU model: ATI Radeon HD 4800 Series 8.01.01.961 1024Mb

Settings
Render: direct3d10
Mode: 2560x1600 fullscreen
Shaders: high
Textures: high
Filter: trilinear
Anisotropy: 4x
Occlusion: enabled
Refraction: enabled
Volumetric: enabled


and

Unigine Heaven Demo v1.0
FPS: 50.6
Scores: 1274

Hardware
Binary: Windows 32bit Visual C++ 1500 Release Oct 22 2009
Operating system: Windows Vista (build 6002, Service Pack 2) 64bit
CPU model: Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz
CPU flags: 2499MHz MMX SSE SSE2 SSE3 SSSE3 SSE41 HTT
GPU model: ATI Radeon HD 4800 Series 8.01.01.961 1024Mb

Settings
Render: direct3d10
Mode: 1280x800 fullscreen
Shaders: high
Textures: high
Filter: trilinear
Anisotropy: 4x
Occlusion: enabled
Refraction: enabled
Volumetric: enabled
Unigine Corp. © 2005-2009