Serious Engine 2 early build results for x800&6800

[silence]

link is here and here

this is croatian site, they just got x800pro for testing and they got early build of Serious Engine 2 (SS2). i know that Croteam showed 3Dc on SE2 and that ATi used SS2 for showing their new cards, so i am little bit surprised with these results and i am posting just to hear some comments.

 

[Killer-Kris]

link is here and here

this is croatian site, they just got x800pro for testing and they got early build of Serious Engine 2 (SS2). i know that Croteam showed 3Dc on SE2 and that ATi used SS2 for showing their new cards, so i am little bit surprised with these results and i am posting just to hear some comments.

Well the first thing that definitly jumped out at me was how much of a hit every card, with the exception being the 6800, took when using OpenGL.

Anyone have any ideas as to why? Even the FX series took a huge hit, which leads me to believe that it's not just a matter of nvidia's better GL drivers.

 

[mikechai]

NV3x suffered a HUGE performance hit.
I also noticed in SS2 DX9 vertex shader test, NV40 performed worse than NV35.

Experts, please explain...
________
Crown comfort lpg

 

[Hyp-X]

If these are really pixel and vertex processing stress tests, then it's unacceptable to have this much speed parity between the two APIs.

There are either serious engine (npi) or driver issues.
Until these are worked out I would take these result with a ton of salt...

 

[tEd]

could be that they used pp for dx9 which never seemed to work in opengl if they used arb_fragment_pogram which could explain the results for fx5950 in d3d and opengl

 

[Bjorn]

Until these are worked out I would take these result with a ton of salt...

Or two

 

[silence]

they say they used early build of engine, which might explain some of the results.

 

[DemoCoder]

OGL deals much better with smaller batch sizes compared to DX9 (one reason why geometry instancing is a win) The driver model is more efficient than DX which is why Longhorn will have a new model.

The bad vertex results for the 6800 (worse than NV3x) are a compiler issue.

 

[Hyp-X]

NV3x suffered a HUGE performance hit.
I also noticed in SS2 DX9 vertex shader test, NV40 performed worse than NV35.

Experts, please explain...

Well, suffering a performance hit is expected because the DX9 version is likely using half's all over the place, and GL is full precision.

Yet this hit is too large - it's likely not the same workload.
If you look at the ratio of FX5950 : 9800XT
DX9 211 : 206.8 = 102%
OGL 33.2 : 83.1 = 40%
that looks somewhat closer to the truth altough a 2.5x speed difference between half and full precision is still surprising.

Either the same performance of the 6800 or the differing performance of all the other cards are likely is an error of the one who did the benchmark/tables.

btw, you got to love how 14.4 has a longer bar than 28 ...

 

[DemoCoder]

I think the OGL pixel shader result for 6800 is probably an error.

 

[Hyp-X]

OGL deals much better with smaller batch sizes compared to DX9 (one reason why geometry instancing is a win) The driver model is more efficient than DX which is why Longhorn will have a new model.

What you say is true - but I don't think it has anything to do with these results.

1.) If one wants to do a 'pixel shader' or a 'vertex shader' test one is likely to make sure it is PS and VS limited - and that means large enough batches.

2.) Too small batch sizes -> CPU limited. Those results doesn't look like CPU limited.

3.) Everything except the 6800 seems to have worse performance in OGL than in DX9.

The bad vertex results for the 6800 (worse than NV3x) are a compiler issue.

Seems likely.

I think the OGL pixel shader result for 6800 is probably an error.

Me too.

 

[991060]

Any idea on wich language is the page in?

 

[chavvdarrr]

Any idea on wich language is the page in?

10 years ago it was called serbo-croation... now.... hmmm, if that was yugoslavian site that would be serbian, but the site is croation, so ...

 

[oddfellow]

Any idea on wich language is the page in?

10 years ago it was called serbo-croation... now.... hmmm, if that was yugoslavian site that would be serbian, but the site is croation, so ...

French??

 

[silence]

link is here and here

this is croatian site, they just got x800pro for testing and they got early build of Serious Engine 2 (SS2). i know that Croteam showed 3Dc on SE2 and that ATi used SS2 for showing their new cards, so i am little bit surprised with these results and i am posting just to hear some comments.

i guess that could help give you a clue
and Cro in Croteam also helps....

 

[K.I.L.E.R]

Anyone in need of a translation?

My Croatian is a little rough however.

X800 PRO nam je pak pokazala da ni OpenGL vertex performanse nisu baš onako dobre kao što smo mislili.

X800 PRO shown that OpenGL vertex performanse isn't as good as we thought.

 

[Zvekan]

Anyone in need of a translation?

My Croatian is a little rough however.

X800 PRO nam je pak pokazala da ni OpenGL vertex performanse nisu baš onako dobre kao što smo mislili.

X800 PRO shown that OpenGL vertex performanse isn't as good as we thought.

Not bad

If you are really interested I can translate it whole, but from results I think that there is too much difference between tested cards and therefore results aren't too interesting yet.

But I'm interested why there is such a large disparency between different chips.

Zvekan

 

[K.I.L.E.R]

I don't have that much of a problem reading it as is.

However a translation would save me a bit of work.

Anyone in need of a translation?

My Croatian is a little rough however.

X800 PRO nam je pak pokazala da ni OpenGL vertex performanse nisu baš onako dobre kao što smo mislili.

X800 PRO shown that OpenGL vertex performanse isn't as good as we thought.

Not bad

If you are really interested I can translate it whole, but from results I think that there is too much difference between tested cards and therefore results aren't too interesting yet.

But I'm interested why there is such a large disparency between different chips.

Zvekan

 

[991060]

We need translation of the 2 pages, somebody please

 

[silence]

We need translation of the 2 pages, somebody please

nothing really interesting there. one of reasons i posted link is that i wanted to hear opinion on you guys over those graphs. they have few mistakes in their review, at least i think they do.