EXT_compiled_vertex_array still useful?

I have made some tests with an application which allows me to toggle EXT_compiled_vertex_array support.
It turns out that there’s no performance gain by enabling CVAs! Sometimes, there’s even a performance hit (often little, sometimes greater).

I am minding the idea of dropping support for it in my app.

Can someone post their experiences about that extension?
I know it depends from a variety of factors, but I don’t want to get in detail. Maybe this functionality is simply not useful anymore with today’s video cards and I would like to know what do you think about that.

Thank you!

No performance gain compared to what? Display lists?
Indeed the extension is obsolete for today s cards, but it is perhaps the most advanced supported by my old card and so, I would appreciate any app that still allows to use whatever obsolete extensions it supports (as I wont have a better one for a LONG time).

what a card is it that you have got?

GeForce4Ti4400, AthlonXP1800+, singlechannel ddr 333 cas2.
It’s hard to say if the test I did was correct (I guess writing a benchmark is the only thing to be sure of that) but the thing I have said it’s true. In fact, I didn’t even know if the program I used (very famous OpenGL game) uses display lists or vertex arrays. I heard it uses display lists when possible but I don’t think this is true since by enabling the logging I can see glLockArrays is getting called… Oh well, it’s not really important anyway since this is an ask for your opinion on the extension and not on my benchmarking session.

but it is perhaps the most advanced supported by my old card and so, I would appreciate any app that still allows to use whatever obsolete extensions it supports (as I wont have a better one for a LONG time).

Granted. Trust me, I also hate dropping support for hardware that’s still there but supporting this extension (trust me or not) makes everything more difficult for me in the internal geometry management. Nothing I can solve without some hours of work but if it does not provide additional performance it will just be trashed time.
By the way, the program simply needs a programmable vertex pipe to run right now. This hurts much more but I really could not do it without vertex program facilities. Having a set of “generic vertex attributes” instead of some specific attributes makes everything easier. In the future I may add support for fixed vertex pipes but I don’t plan to do it since you can get a GeForce2MX to run the program (with software vertex pipe) at a very low price.

CVAs are not very useful partly since the spec is ill defined and partly because you only gain something if you render from a locked vertex array repeatedly. Displaylists are a better way to get good geometry performance on hardware that doesn’t support vbos.

Originally posted by JanHH:
what a card is it that you have got?

I have an Nvidia vanta lt, a cheap, office work oriented card. Compared to the other cards I have here (i810 and Sis 300) it is good, but lately Im getting in troubles to run games (I got Neverwinter Night and I cant play it).

Originally posted by Obli:
you can get a GeForce2MX to run the program (with software vertex pipe) at a very low price.

Sorry, what is cheap for you is far from being cheap for me
I cant pay for a gf2. In fact, I cant pay for any card at all.

Ok, I decided I will consider the idea of adding support for it in the future. By now, this extension is not supported - looks like it’s a common though it does not give enough performance gain even if I have mixed feelings about some replies.

Thank you everybody for the contribution.