Suggestions needed on benchmarking the NV35

J

JF_Aidan_Pryde

Regular
We are likely to get samples on Monday/Tuesday, some suggestions would be most welcome.

- Which tests would you recommend?
- What kind of testing methodolgy?
- Any specific ideas for image quality tests?
- What kind of comparisons to run with R3xx?

Uttar:
If you have any setup scenarios to test any of the pipeline/FP unit theories, l'll give them a shot. Just be detailed on the method. :)

Thanks in advance everyone.
 
I'd be interested in seeing how it holds up in the [pure] pixel shader bench I wrote for the nv30/r300. But apparently I'm the only one.. :?

[EDIT] Here's the original thread.
 
Ilfirin said:
I'd be interested in seeing how it holds up in the [pure] pixel shader bench I wrote for the nv30/r300. But apparently I'm the only one.. :?
Click to expand...

No, you're not! I want to see how Pixel Shader performance have changed over NV30: I also want to know what is different, but give us the results first, then we can speculate from there. 8)
 
Don't forget to take other people reccomendations for the pipeline arrangement too, though :)

Okay, so first, the traditional suggestions:
One of the things I'd like to see is a IQ ( & Performance ) comparaison of the NV35's 4x AA & 8x Quality Aniso VS the NV35's 4xS AA & 8x Balanced Aniso.
Also, one of my favorite benchies is UT2003 botmatch - but only with high AA/AF, otherwise it's CPU limited and useless. Getting true realworld performance is a lot more practical than some type of timedemo or flyby :)


Now, for the pipeline arrangement...
While, the problem is that to test this correctly, you'd need a dedicated full-screen program running a fragment file you could easily edit. And well, I don't have the time to code that now because I don't even have VC++ installed anymore, since I'm installing Linux on my secondary PC and stuff ( won't go in a detail explanation here )

I'm sure some people on the forums might already have coded that, if so, please say it and maybe send it to him so I could simply give fragment files to test? Would be quite easy then.
Otherwise, maybe some of nVidia's CgFX tools might be used to easily benchmark Cg shaders... Not sure, never tried...

Anyway, without all that, it's still possible to run the programs created by some of the people on the forums. I'm sure that since the fragment file would then be public, and that nVidia wouldn't have optimized them, it would be quite objective - but figuring out what performance you SHOULD using a given architecture is the hard part then, though. Should be possible anyway, I guess...


Uttar

Uttar
 
How about using the most up to date versions of the main 3D engines to benchmark with, but use FRAPS or something similar to get the actual FPS. That way you aren't exposing yourself to the possibilty of drivers that optimise for a fixed viewpoint.

And some of the post processing done by the NV3x doesn't show in screen captures, would it be possible to take some real photographs in situations where this is the case?
 
Screw all that. ;) Set every setting to max quality then benchmark it , then set everything to lowest quality then benchmark. That will establish 4 ranges , fastest and slowest FPS and the corresponding best and worst picture quality. Make sure you get screenshots of each setting , be sure to mention and display any graphical anomalies or wierd strutters etc.
I want to know one thing , where is the highest quality at a playable framerate. Both are very subjective , so IMHO the best you can do is show them the range of the product , you might have an editorial type conclusion where you give your opinion (clearly stated as such with as much detail to how you formed your opinion as possible) as to what you think the products "sweet spot" is. Depending on the time I would use FRAPS to benchmark 2 games from each genre preferably OGL and DX respectively.

edit horrible english grammer :rolleyes:
 
edit horrible english grammer
Click to expand...

Heh, then we won't go into the spelling. (grammar) :LOL:

Sorry, couldn't resist. I'm not usually one to mention someones spelling or grammar because frankly, both of mine suck.
 
I would just LOVE to get some IL2-Sturmovik - Forgotten Battles benchmarks on both R9800Pro and NV5900.
With "perfect" landscape settings this game becomes a hog. At least on my 9700@360/320.

Also this would distinguish your benches from a lot of others since this game is seldom used for benchmarks.
Usage of Fraps is very reliable with IL2-FB and highly recommended.
 
Ilfirin said:
I'd be interested in seeing how it holds up in the [pure] pixel shader bench I wrote for the nv30/r300. But apparently I'm the only one.. :?

[EDIT] Here's the original thread.
Click to expand...

Hmm. Not sure I'm running this correctly but....(@1600x1200 - default settings)

1> 5995.45215 MIPS
2> 11258.13672 MIPS
3> 7100.32520 MIPS

As I said above, I may not be running it correctly (I alt-f4 out after a bit, get a result, then the next test starts).
 
2B-Maverick said:
I would just LOVE to get some IL2-Sturmovik - Forgotten Battles benchmarks on both R9800Pro and NV5900.
With "perfect" landscape settings this game becomes a hog. At least on my 9700@360/320.

Also this would distinguish your benches from a lot of others since this game is seldom used for benchmarks.
Usage of Fraps is very reliable with IL2-FB and highly recommended.
Click to expand...

The nvnews.net preview of the 5900U has at least some numbers for IL2 v1.2.
Go here for the nvnews preview
The second page has the IL-2 gameplay numbers based on FRAPS and some various other settings. Don't know if "perfect" landscape was used and there is no comparison to a R9800.
 
Use non-standard benchmarks and you own recorded demos in addition to the "standard" benchmarks. It's the difference between the 2 cards I'm interested in. I'm NOT interested in how many fps I will get in Quake3 (and the like), I know it will be more than enough.

As an example, create your own flyby demo in ut2k3 with a random starting point, run it on the 2 top end cards (9800pro and fx5900u). Then run a standard flyby demo on the same cards. THEN note the differences between the "standard" fps and the non-standard fps.
Would be very good to know if one company has "optimized" too much in the standard benchmark.
Repeat tests where the game will allow recording and playback of demos.
 
I'd do what NVNews does: FRAPS recording the framerate of gameplay. Of course this reduces precision greatly, but I don't consider one card much faster than another until I see a ~10% performance difference.

This would let you use games other than timedemo-equipped PFSes, too.
 
Also confirm that the driver settings are acutally being applied as selected. With the similarities of 4x and 8x who knows.
 
Here's a thought. Talk about the card itself. I'm not sure if you guys had the pleasure of using the 5800 Ultra or not so here goes nothing:

1) Compare length of card to other cards (will it fit in most cases you have lying around?)
2) How's the tempature of the card overall?
3) Is thermal grease used or is it a thermal pad?
4) How loud is the fan?
5) Is the connector held on the PCB by screws or just by the VGA & DVI connectors
6) Bitch about the lack of Dual DVI for a $499.99 card! Hehehe :D
7) Last but not least. Would you drop $500.00 on this card? Considering reviewers never have to pay for the hardware. This is something that always lack in reviews. o_O

Oh... Please at least bring up "Con's"... One site I've read the preview of the 5900 Ultra was nothing more then a fricken Nvidia brochure. Lame...
 
DadUM said:
The nvnews.net preview of the 5900U has at least some numbers for IL2 v1.2.
Go here for the nvnews preview
The second page has the IL-2 gameplay numbers based on FRAPS and some various other settings. Don't know if "perfect" landscape was used and there is no comparison to a R9800.
Click to expand...
I'd fry my comp before trusting anything they put out.
Edit: Almost forgot it.
Bench some not so standard games. As some already has said.
 
You must log in or register to reply here.

Similar threads

R
  • Locked
Reverend at The Pulpit #12
2 3 4
Replies
73
Views
12K
Dave Baumann
Dave Baumann
N
fraps Benchmarking
2 3
Replies
45
Views
9K
WaltC
W
Back
Top