Recommended video cards

[esteel]

I'd just like to note, that I have a 7600GS (Asus Top Silent model with 550/950 clocks instead of the standard 400/800) and I can't enable all the effects. I can enable everything besides "realtime world lights" which causes framerate drops below 30fps in 1024x768 without AA.
(Linux x86, SDL binary)

Do the nvidia drivers support that card already? I have a 6600gt and can run nexuiz FINE! For example i can run Nexuiz 2.1 map runningman with 'ultra.cfg' setting and get 30-80 fps. Using the normal (startup) config i get 100-250fps. Of cource that is alone in the map and for online play i use the lowest possible settings but your Card should be able to run Nexuiz WAY better then 30fps..
Do other games work? Does glxinfo | grep -i direct say 'direct rendering: yes' ?

[Ed]

Also, what do you get on glxgears? The 7600GS should beat a 6600GT which will give around 8000.

[Zygfryd]

Yes, direct rendering is enabled. I have ~7500fps in glxgears, but it's a lousy benchmark. I got the 7600GS exactly because it was better than 6600GT in reviews, so I'm a bit confused here.
I've benchmarked the game with all effects enabled on the q1dm6 campaign level. Spectating before it starts gives framerates from 30 to 70, but in actual gameplay it falls to 20 very often. (again 1024x768x32, no AA, best textures).
Xorg 7.0, driver version 8762, nvidia-settings quality control set to 'high quality' (though it still cannot keep 30fps in all places with 'high performance' instead).
Are you saying there might be something wrong with my setup or my card? I started thinking that Nexuiz's effects are simply so demanding.

PS. Still Nexuiz 2.0, but I haven't seen performance improvements in the changelog.

[Ed]

glxgears is actually a very good benchmark. That's because it only needs OpenGL 1.2 and hence should be comparable across most hardware from the last decade and also (as intended) software renderers. It also has little dependence on CPU, memory and HDD thus giving a good graphical benchmark. Things like 3DMark aren't actually very good benchmarks because they depend upon far too many factors which means you get losers disabling the minorest of Windows features just so that they get 1 point higher and they think that gives them a bigger penis despite them having no understanding of statistical significance.

What are the specs of your computer? Nexuiz is also restricted by things other than graphics card. As you have overclocked your card, did you try Nexuiz with the card at stock settings? If you get exactly the same results, it proves that Nexuiz is not graphics card limited on your computer, in which case you should look at other things.

[Zygfryd]

glxgears is actually a very good benchmark. That's because it only needs OpenGL 1.2 and hence should be comparable across most hardware from the last decade and also (as intended) software renderers. It also has little dependence on CPU, memory and HDD thus giving a good graphical benchmark. Things like 3DMark aren't actually very good benchmarks because they depend upon far too many factors which means you get losers disabling the minorest of Windows features just so that they get 1 point higher and they think that gives them a bigger penis despite them having no understanding of statistical significance.

What are the specs of your computer? Nexuiz is also restricted by things other than graphics card. As you have overclocked your card, did you try Nexuiz with the card at stock settings? If you get exactly the same results, it proves that Nexuiz is not graphics card limited on your computer, in which case you should look at other things.

Glxgears seems very CPU-bound because it saturates one of my cores (with 50% user and 50% system time on that core).
My computer is a month old. It has a Core2Duo 2.13GHz, 2GB RAM, two fast drives. I haven't changed the clocks on the graphics card, they should be at 550/950 set by the manufacturer, but I don't know the actual speeds because nvclock doesn't support the card.

PS. I meant no AF, not AA, in my earlier posts. But there's no AA either.

[Ed]

Glxgears seems very CPU-bound because it saturates one of my cores (with 50% user and 50% system time on that core).

That would be the root of your problem then, not a problem with glxgears. Despite having enormous processor power, you will be CPU bound in Nexuiz too. The stock speeds for a 7600GS are 400/800MHz, downclock it and you'll get nearly identical FPS. Whatever it is, it's a porblem with processor usage, not the graphics card itself.

[kyre]

In defense of the widely abused "glxgears" app, I'll try to sum up what it does and does not measure.

glxgears tests the following things:

How fast glFlush() is (implicitely called by glXSwapBuffers()). Swapping buffers several thousand times a second isn't really a useful benchmark parameter, real applications do rather a lot more rendering between swaps.

How good the driver is at throttling the command submission rate - try dragging windows around while glxgears is running, if it's choppy the driver isn't doing a good job of throttling back the submission rate to a sane value. Note that this means high values can be a bad thing...

How fast glClear() is. This is really the only thing that translates well into actual performance in games such as Nexuiz, and even then it's only about 10-20% of the rendering time for a frame. You need to maximize the window to get any useful values here.

glxgears does NOT test the following, which tend to be deciding factors for performance in many apps, Nexuiz included:

Fill rate, which is the limiting factor for many games. The time spent drawing the gears is dwarfed by the frame buffer clear and command submission overhead.

CPU/GPU bus transfer speed. The gears are stored in a display list which is usually uploaded by the driver to graphics card memory, so the size of the rendering commands sent by the CPU for a frame is ridiculously small.

How fast the driver/hardware handles state changes such as changing textures, lighting parameters etc. This is often a bottleneck in actual applications.

CPU performance. In glxgears the CPU just spends all it's time submitting lots of tiny command buffers to the GPU, and waiting (often with a CPU-hogging spinlock) for those commands to complete. Real apps submit up to several megabytes per buffer (including a fair amount of processing to build these buffers), not the around 1kB glxgears does, and naturally do so a lot less often than glxgears.

To sum it up: glxgears performance doesn't really tell you anything about how real apps will perform. Benchmarking those real apps doing real things will.

PS: Sorry about the long somewhat off-topic post...

[esteel]

Hmm your fps problems are very strange.. Could you try those options and see if they help? Just open the console (ingame with ` the key left of 1 or shift-escape does also work) and type the bold text? You can use the number mentioned as default to switch back to the default
gl_finish 1 (default 0)
gl_lockarrays 0 (default 1)
gl_combine 0 (default 1)
Does one or a combination of those help? However IF so this points to some faulty drivers or or other faulty stuff..

[Cyr4x]

My FX 5700 LE works very fine. All OpenGL 2.0 effects are off, but textures at high, and af at 16x. Usually 50-150 fps. Good for online play.

[Ed]

Think this list needs updating. The Geforce 8 will theoretically be a lot quicker but I'd like to see some Nexuiz benchmarks some time. The Geforce 7300 is not on this list either. There will be a lot of range with these cards as the 7300GT is quite different from the GS and then the LE is someway behind the standard 7300. Anyone got any of these cards for comment?

I think it may also be a good idea to periodically list cards that people can recommend to buy, many of those cards no longer are available or are old top end cards that offer poor value for money.

Here's what I'd recommend at the moment, descending in price on different interfaces:

PCI-E:
8800
7950GT
7900GS
7600
7300

AGP:
Gainward 7800GS+ (7900GT on AGP)
7600GT
7600GS
7300GT
6200
5200

PCI:
FX5200/5500
MX4000

Yes, it's only NVIDIA cards and no, not every card they have ever made is in there but only some that are worth buying. Yes, you could buy a 6800 Ultra but you'd struggle to find one new (eBay doesn't count as being generally available for sale, an item is a one off) and you'd pay more for it than a 7900GS which would beat it. I have included some cheaper cards that aren't that highly recommended for performance but at the price point, it's the best you'd get.