[Dave Baumann]

I've just noticed the following message in my inbox:

This mail is being sent out to a wide group of hardware review sites.

In the past three days we've collected over 120,000 responses to the Source Engine Video Stress Test which is built into the Counter-Strike:
Source Beta. Any user that runs the test in the engine has the choice to share the results of this test with us and the rest of our community.

Attached to this mail is the database that contains these responses, as well as the spreadsheet that displays the default settings that the Source engine prefers on different hardware platforms.

For us, this data is useful to make sure that we're correctly configuring video settings on our customers' computers, exposing any pathological performance problems before we release Half-Life 2, and making sure that we're accounting for all of the varying hardware configurations of our customers.

For you, this data should be useful for getting an idea of the kind of performance of Source from a broad data set. Data from most benchmarks tend to be produced in a vacuum, but this data could be accurately extrapolated to millions of gamers.

We think gathering and disseminating this performance data broadly and openly is good for everyone. While our internal performance testing data has been useful as a proxy for customer's experience, it is hugely more valuable to know what their actual performance is. If you have suggestions for how we should enhance this in the future, please let us know and we'll add it to Steam. Please feel free to publish any of the data, or conclusions you draw from it.

Some information on the database and the spreadsheet:

- All of the data is anonymous.
- The spreadsheet contains three main types of configuration data, and they are applied in this order for purposes of auto configuration:
1. Card specific settings. These are the rows with the friendly names of the cards. Those names are only for readability of the spreadsheet, Source bases everything off of the Vendor and Device IDs.
2. CPU Speed.
3. Amount of RAM installed.
- Users can submit data to the SQL database an unlimited number of times.
- The database is a text dump from a MySQL database. You can use this command to import it back in:

LOAD DATA INFILE 'data.txt' INTO TABLE benchmark FIELDS TERMINATED BY ',' ENCLOSED BY '"' LINES TERMINATED BY '\n'; And here is the create table definition:

CREATE TABLE `benchmark` (
`benchmarknumber` int(11) NOT NULL auto_increment, `time` timestamp(14) NOT NULL, `framerate` float default NULL, `build` int(11) default NULL, `vendorid` int(11) default NULL, `deviceid` int(11) default NULL, `ram` int(11) default NULL, `cpu_speed` int(11) default NULL, `cpu` char(22) default NULL, `width` mediumint(9) default NULL, `height` mediumint(9) default NULL, `AASamples` smallint(6) default NULL, `AnisoLevel` smallint(6) default NULL, `SkipMipLevels` smallint(6) default NULL, `DxLevel` smallint(6) default NULL, `Windowed` tinyint(4) default NULL, `Trilinear` tinyint(4) default NULL, `Bilinear` tinyint(4) default NULL, `ForceHWSync` tinyint(4) default NULL, `NoWaitForVSync` tinyint(4) default NULL, `DisableSpecular` tinyint(4) default NULL, `DisableBumpmapping` tinyint(4) default NULL, `EnableParallaxMapping` tinyint(4) default NULL, `NoUserClipPlanes` tinyint(4) default NULL, `ZPrefill` tinyint(4) default NULL, `ReduceFillrate` tinyint(4) default NULL, `RenderToTextureShadows` tinyint(4) default NULL, `RealtimeWaterReflection` tinyint(4) default NULL, `WaterReflectEntities` tinyint(4) default NULL, PRIMARY KEY (`benchmarknumber`)
) TYPE=MyISAM;

Erik Johnson
Valve

Default board settings CSV file

[carpediem]

can we have data.txt too?

[tEd]

valve will share the data on steampowered.com at some point

[PatrickL]

Wow some people use chips that should be better in a museum

[tEd]

Originally Posted by PatrickL

Wow some people use chips that should be better in a museum

i don't think that's the list of vidcards people actually use in CSS but rather what valve can/will detect

[Demigod]

Well if you look here http://steampowered.com/status/survey.html
you will see there are still almost 2000 people using rage128, 2875 using voodoo3's, 100,000 using gf2's and 147,000 using gf4mx cards. If some of them baught cs:cz then we may get to see how it runs on their cards

[LeStoffer]

Originally Posted by DaveBaumanns inbox

For us, this data is useful to make sure that we're correctly configuring video settings on our customers' computers, exposing any pathological performance problems before we release Half-Life 2, and making sure that we're accounting for all of the varying hardware configurations of our customers.

Well, that at least mean that I will have time to finish Doom III before Half-Life 2 is released.

[Pete]

Dave, I'm wondering why Valve set AA to "6x" on the 6800 series when nV doesn't have such a mode. I'm also curious as to why the X800 defaults to 8xAF, while the 6800 defaults to 4x. Will nV's new AF optimizations reduce some of the 30% lower performance Valve stated, and allow Valve to bump the 6800 up to 8xAF by default? Or is that an issue of IQ, not performance?

[euan]

There's not many highend NV cards there from the last generation... None from either ATI or NV for the current gen.

[Pete]

The csv has both the 6800 and X800 series. :? Just look for the only cards with AA and AF enabled by default.

[euan]

Originally Posted by Pete

The csv has both the 6800 and X800 series. :? Just look for the only cards with AA and AF enabled by default.

Yeah, but there are none in the results webpage. It's as if all the benchmarks we have seen are made up, or they have been removed for a reason (IE not to highlight potential market shares).

Edit:

But what do you expect when the linked survey hasn't been updated since 6:44pm PST (02:44 GMT), May 03 2004