Jump to content

Dear NVIDIA


Thorburn
 Share

Recommended Posts

Please take SLI and shove it up your arse.

I've been trying to get results for the 8800GTS in SLI with Enemy Territory Quake Wars for the last 3 hours and I swear if it crashes and corrupts the hard drive image one more time I'm going to throw the whole damn rig through the lab window.

That is all SAUER0421.GIF

Link to comment
Share on other sites

[ QUOTE ]

Can we get back to talking about how much I hate SLI?

Resorted to running single tests on both the 8800GT and 8800GTS rigs as batching them means every crash means I lose two or three runs SAUER0421.GIF

[/ QUOTE ]

crazy.gif50% of what you said went straight over my head, the other 50% went even further over yelrotflmao.gif

Link to comment
Share on other sites

[ QUOTE ]

[ QUOTE ]

I have a little script I wrote to run my tests all at once with no user intervention, but when the machines crashed I'd lose the other results from the test smile.gif

[/ QUOTE ]

excuse the sacrilege - could the problem lie in the script rather than in SLi? [runs to hide!]

[/ QUOTE ]

Worth asking smile.gif

Script works fine with NVIDIA cards with SLI disabled, and with ATI cards both with and without CrossFire.

It's just a batch file that starts the game with certain parameters to run my benchmark, looping through for the resolutions I specified and reading the results out the console log.

Also the game crashes with SLI when just playing normally smile.gif

Link to comment
Share on other sites

Yep, I have one rig with 2 x Gainward GeForce 8800GTS cards and another with 2 x ASUS GeForce 8800GT cards.

[ QUOTE ]

Really stupid set of affairs when old SLI (a la 10 years ago) worked out the box fine, had 2 Voodoo2 cards linked up no probs. Get the impression SLI now just does not work unless the actual game source code supports it let alone the drivers!

[/ QUOTE ]

Back then 3D accelerators mostly just did texturing and filtering, so it was far easier to share the work load between two cards. All the Voodoo 2's did was each draw half the lines of a given frame and interleave the two together.

Modern cards have to cope with performing vertex transformations, executing code on a per vertex or pixel basis, and many other complex tasks that makes sharing the load between cards far more difficult.

Link to comment
Share on other sites

Same principles though thorburn...

in theory it's straightforward to share the bandwidth for multi-card, 1-display output...

for dx 9/10 recommended batch size is 1000 vertices etc.

typically problems are due to heat/power but with latest stuff there maybe (and usually are) driver bugs - as SLI uses lots of different algorithms to share the load (sharing GPU, sharing framebuffer, sharing textures etc)

modern nvidia SLI is similar to RAID0 with harddisks....textures are typically uploaded to both cards etc (depending on config).... as scene generation evolves, SLI algorithms can adapt more dynamically at runtime- so a variety of algorithms are used, which can change on a per-frame/per-vertex-batch basis.

usually opengl is more stable at driver level....

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...