So, I just bought a new GTX 460 1GB card so that I could play qw and newer games as well. My old card for qw was nvidia gt9600.

With my old card, I could get a pretty stable 1000 fps (although I ran at a lower maxfps).

With my new card, FPS is jumping all over the place from like 600-2000 with maxfps 0 and won't hold even a stable 847 fps.

I did however change my monitor at the same time, so I went from 1680x1050 to 1920x1080. Not that that would really matter _this_ much. I also tried running at 1680, didn't help really.

Murdoc acted as my reference point as he has pretty much the same setup as me in terms of cpu, gpu etc.

System Specs
windows 7 home premium64bit
intel i7 930 stock clock
6gb ram
ssd harddrive
ezquake-gl.exe 2.0.1 stable and 2.1 alpha

What I've tried
- run as administrator
- run in compatibility mode
- ran driver sweep
- installed newest drivers 260.99
- installed older drivers 260.89
- sys_highpriority 1
- sys_yieldcpu 0
- murdoc.cfg (he has about same system specs as me)
- zero.cfg (just for comparison)
- reinstalled windows, clean reinstall
- tried nQuake stock installation
- threaded optimization off (global and specifically for ezquake)
- removed g_weapons
- adjusted process affinity (removed cores being used for ezquake)
- r_dynamic 0

What I've noticed
- Mass Effect runs smoothly with all settings at max, at 1920x1080
- I get pretty much the same score in 3dmark 11 as Murdoc
- GPU-z seems to report that gpu isn't used at nearly 100%, more like under 50% (when running Mass Effect, it was a lot higher, although the gpu-z utility's gpu utilization indicator is a bit suspect)
- GPU fan doesn't seem too active when I play qw, it is set to automatic and doesn't seem to go up much
- even standing in the same spot shooting (dm3 lowbridge / rl stairs with a few bots) the fps can swing between like 900-1500 (edit: same with old card, so np)
- I've read a lot of the forums discussing similar problems with other games, especially bfbc2, no solution from those boards helped

wtf, lol help?

Update

Put in my old card and tried with new monitor, not much difference in fps, but maybe I was a tad optimistic about a pretty stable 1000fps. Not stable enough that I would keep it there, but still quite stable in most situations.

I didn't paste my config here, since I tried many other configs and had a clear reference point with old gpu and same config.

IIRC from f_system reports Murdoc's cpu is overclocked to 3.2ghz(?). I would have maybe said you were cpu limited if it wasn't for the fact that it worked ok with a 9600GT.
If it was me I'd still do some tests at different clockspeeds though.

i7 950, 4gb, win7 home 64, asus gtx 460

i have the same thing, my fps fluctuates from 500-1000. can't offer any help but you aren't alone i guess. although i've been playing with this for some months now and i haven't really noticed it being unsmooth despite the fluctuations.

Are you seriously talking about CPU bottleneck, with my cpu and especially if you've been aware of ddk's problems? Well, any help and guesses are welcome and I don't want to discourage that, but for the record:
I don't want to seem ungrateful for any input, just struck me as a bit outlandish (I'm a bit protective of my cpu <3)

While running qw, my CPU load is around 25-40%, of which ezquake uses about 15-25%. (yes, inactivesleep is off)

One thing I did notice now is that half of the cores are spiking at full load at times and half are barely activating.

Some more affinity tests (high priority)
- 1 core: no noticeable difference in fps, sounds stuttering a lot
- 2 cores: same, but sound stuttering a bit less
- 3 cores: seems the same as running with all 8 cores (the third core doesn't seem to get maxed out).

Oh, and another thing I've done is disabled core parking, so that no cores should be parked when not needed, just to make sure

Well, it's not unsmooth at all with a maxfps of say 616. But if I can't get this fixed soon, I'm taking it back and have to rethink things. No point in paying some 180e for the ability to play newer games with the same card, but getting no benefit in qw

ezQuake doesn't stress out your card that much, especially since it uses no shaders. Also GTX460 are silent cards that don't increase the fanrpm too much even when stressed with programs like furmark. You can ignore these two things totally, they don't matter.

Your machine should be able to keep the fps up to 1000fps easily. Note that when you are actually in QTV, your fps will be significantly lower than when you are speccing or playing normally (perfhit can be more than 50%, I tested this two weeks ago myself, it is because the client actually has to do more with CPU than when you are playing).


So, try running DPC latency checker if you haven't yet. Check the values when your computer is idling, watching youtube and finally when playing QW. There shouldn't be any problems with i7-class hardware generally, but still worth checking.


You can also disable MMCSS if you have integrated soundcard and see if it affects anything, at least some people have gotten rid of fps-problems doing so (xenic for example), and for some it has helped to ping/pl related problems. (this requires that registry change, otherwise you will lose all sound)


Since 260-drivers are buggy with GTX400 cards, I recommend testing something like 258.96 WHQL since it has been the last really good working driver for my GTX470. Any 260-driver will cause high GPU-clocks when idling on the desktop using 120Hz refreshrate, so fuck that. I also think there were some OpenGL related problems with 260.xx but I can't remember what it was about, probably something threaded optimization related.



These won't help you at all, also note that you do NOT have 8 cores, you have 4 cores that are hyper-threaded (capable of running 2 threads per core). Also note that there was change in the W7 kernel regarding hyper-threading: in certain cases where Vista or XP could lose performance when hyper-threaded CPU is used, W7 will not, this is especially seen on different games (some games like arma2 still have the problem though).

I did some of the testing just speccing in qtv, but enough outside of that to verify that it is not simply a matter of QTV MVD processing overhead.


Forgot about the latency check issue. In my previous install I had had MMCSS disabled, but I might have turned it back on as I didn't have any similar problems. Have to try that again.


Will do.


These were mostly just checked out because I was curious, especially after HT's doubts concerning cpu bottlenecking.

Thanks for the answers, I'll get to these tonight.

Yes I am seriously talking about a CPU bottleneck
I7-930 only runs at 2.8ghz. ddk has i7-950 which runs at 3.06ghz. My cpu is at 4.1ghz (albeit only an i5 so slower clock-for-clock) and gives 1001fps stable (on ‘normal’ conditions; not in QTV or at quad on dark-terror-ffa.bsp).
Remember that while you might normally scoff at the idea of an i7 being a bottleneck, bear in mind that:

1) 1000fps is far from the ‘normal’ conditions you see people testing/reviewing/benching at. It may be capable of 100fps in modern games but that doesn’t mean to say it is capable of (stable) 1000fps in ezquake.
2) The faster your GPU, the more likely you are to be cpu-bottlenecked.
3) Having 4 hyperthreaded cores doesn’t necessarily help as much as you might expect for gaming because not all threads are created equal. The fact that you state “half of the cores are spiking at full load at times” definitely sets alarm bells ringing for me. If you have 1-2 extremely demanding threads (this is fairly typical of most games, especially older ones) then the other cores will simply have to sit around waiting for the ones that are fully loaded. On a system with 8 logical cpus 25-40% utilisation could easily mean that there is a cpu bottleneck i.e. one core maxed out.

Again, I’m not saying you definitely have a cpu bottleneck (I would have expected performance at least on par with 9600GT). But rather than simply dismissing the idea out of hand, why not test it? If you don’t want to overclock, try underclocking and see if the problem gets worse &#61514;

While im using the same config as niomic i get stable 1000, when using maxfps 0 it jumps from 2400 to 1100.

Im on a E7500 / GT 450 1GB / 2GB RAM / WIN7 32bits tho.

I don't see how he gets lower fps then me. Even with CPU bottleneck as you said.. ?

Just to set this straight. I would say the performance with gtx 460 is _a tad_ better after doing more objective and lengthy comparisons. Especially with 1920x res. Sure, it doesn't hurt to try, and I can admit that I'm not as up-to-date with how modern cpu architectures work, especially in terms of older applications and GHz. I can get my CPU to at least 3,5-3,6GHz with stock cooler, but if OCing turns out to be a requirement, then I'd prolly have to get a good cooler.

Hmm I have to agree with Hangtime here, I tried to oc even more to 3.67ghz and it gave me another 250+ fps, i also installed 258 driver and did mmcss tweak, i can now get 1500fps stable but i am using 1080.

If you allow unused cores to idle/park, they won't generate so much heat, and your cpu cores that are actually being used will not throttle themselves at all (or will overclock slightly).
I've an i7. Presumably your cpu usage as reported by windows task manager should probably max out at around 16%. Activate vsync and it should drop to showing between 1 and 3%, or so.

If you're running at 25% cpu usage, you have some other thread consuming 100% of the time it possibly can. If you have 50% cpu usage, you're basically maxing out 4 cpu-loads of quake. That's a lot of memory bandwidth lost, and a lot more heat. A lot more heat means your CPU will throttle itself resulting in each core running more slowly.

Technically with hyperthreading, there is no significant difference between 50% cpu usage and 100%, except for cpu cache/pipeline stalls, and that won't speed up quake, it'll only make it run slower, and definitely more juddery.

The beauty of C-states and Core I7. For example my I7-860 clocks to 2933MHz when three or four cores are fully utilized, 3333MHz with two cores and 3466MHz with a single core active. You can even set both min/max state in W7 power options to 100% and the CPU is still able to idle normally (at 1200MHz) while utilizing C-states properly (high clocks). Why do they even bother marketing the 860 as 2,8GHz...


NVIDIA's OpenGL ICD (nvoglv32.dll) will be taking 100% of the one core if threaded optimization is enabled by the driver in the NVCPL, hence 25% in this particular case. Yes, it uses an extra thread to accelerate OpenGL rendering with CPU, causing a bit less than 10% performance increase in framerates.


Threaded optimization in action (OpenGL only)

The thing is the i7-860 turbos a lot more than the i7-930, that only goes to 3.06ghz (less than 10% increase in clockspeed), which is why I'd recommend at least experimenting with manually set clockspeeds on the s1366 platform.

Yep, that's why I got myself a 800-series i7. The extra memory bandwidth with 900-series i7 has doesn't really help in anything else than rendering and video processing, basically a stuff that I don't do. And the clock difference makes 800-series i7s actually faster in games and certain other situations (there are exceptions). Oh boy the amount of offtopic here

Thanks for all the feedback. Did all the things Renzo recommended and didn't seem to help much.

I had also forgotten that I was actually running at like 3,4GHz before, so I wasn't even near 2,8 stock clock. I've now bumped up to 3,64GHz and oc'd the gpu a bit too. Nothing major changed with fps.

Also the GPU's fan starts to give out this high pitch whine at certain rpm. Dunno if that would be enough cause to just take it back and switch to another brand. Dunno why that would help, but this is after all IT shit we're talking about, might as well be astrology And today I've actually played quite a few real games to see how the FPS works out, and it isn't that much better than through qtv (for me).

Some other things I've noticed a long the way:

g_weapons can really affect fps quite dramatically, like MatriX was trying to say. Standing on dm4 ledge next to green armor so that you can see lg, ssg, gl and ng:
g_weapons: 1050fps
normal models: 1750fps

r_particle_count number seems to affect FPS even if you don't have any particle stuff enabled. Also particles seem to be the easiest way to fuck up your fps.
timedemo dm3 (4on4 20min), r_particle_count 4096: 657fps
timedemo dm3 (4on4 20min), r_particle_count 0: 1198fps


How to get rid of death and console fps drops?
Console down seems to drop fps from like 2000 to 1800. (all effects and pictures have been turned off, and alpha doesn't seem to affect it)
When dying (dm3 lg diss), fps drops from 2600 to 1200 ?

Threaded optimization didn't seem to cause me any problems, simply added about what renzo said, 10% more fps

With your machine you should be able to get 1000fps stable. Maybe it is really something cfg-related or even something that you are running on your machine (some service or program maybe, like badly behaving anti-virus, it should NOT happen but you never know...)

Shouldn't be CFG, since he tested with other ppl's CFG, that obtain 1000+FPS with same PC setup. Hard to think it's anti-virus/services, since he tested with clean re-install.

He IS getting more than 1000fps, but not having it stable. Since it's not about CPU/GPU power, it has to be somewhere else, so the remaining options are 1) the client's configuration or 2) something within the operating system (includes drivers/driver settings).

Also why would re-install make the problem go away if it's windows service related problem, like MMCSS, since it is always enabled by default? Not like MMCSS is the only problematic service W7 has, for example indexing service can cause performance problems right after windows is installed, since it indexes a lot of data (unless you had empty HD).

Well, niomic can show me the cfg (or send it via IRC/DCC) and I can test it on my W7 and see if it has something that causes the unstable fps. It's always too much guessing when it comes to someone else's computer via a msgboard.

I think I can get stable 1000 fps with Threaded optimization on. If I limit my particle count and effects a bit and remove g_weapons, I can probably reach 1500 fps stable. g_weapons and Threaded optimization off, in combination with a bit smaller clocks for both gpu and cpu could explain the difference when I compared murrie's and zero's configs before.

1000 fps stable is all that I need and I know the tools to get a lot more, if I'm willing to forgo some particle effects and g_weapons, so in that sense I'm quite satisfied atm.

I am still very interested in knowing how to get rid of console and death fps drops or even limit them, because I want stable to be stable in all situations. If anyone can explain why g_weapons are so resource hungry, and if they could be fixed, that would be nice too.

Once again, thanks for all the input, many knowledgeable people in the scene

And sure, I can post my CFG's relevant parts here tonight (GFX etc.) since it seems that there are just too many variables involved even with reference cfg's.

Are you saying you want to run cl_maxfps 1000 while playing?

I can't get 1000fps stable either. Don't really care cos I only need 500fps.

I have

i7-875K stock speed
GTX 285
Win 7 64bit.

I Would try fuhquake with indep phys just to check .

I think console fps drops are something pretty much everyone gets(?)…. I guess the game has to renderer everything as normal (irrespective of alpha) and then the console on top. It’s not something that bothers me particularly because if my console is down I’m not playing and thus don’t really need stable FPS.

I would assume everyone gets a noticeable fps drop from g_weapons and especially dying, doesn't mean the issue shouldn't be addressed? I agree that the console fps drop isn't that bad, but it's not even close in magnitude to the drop you get when you die. And there the problem is that it takes a little time for qw to get back up-to-speed after you respawn, as in it is noticeable when I play, if I don't play with a low enough cl_maxfps to mitigate even that problem and that would mean that I'm running at 3/4 or 3/5 of what I could get stable without this dying fps drop.

And what exactly makes the console so special? It's a 2d element. By that logic everything in your hud should be slowing shit down a bunch (and sure enough I did try removing all new hud stuff and didn't make much difference)

From all that I know of smoothness in qw, my assumption is that for now, stable 1000 fps is as high as is meaningful to even go (due to usb hz, or so I understood from Renzo's samsung review). I'm perfectly fine running with something like 616 maxfps. This was originally more about "hey, I got this card and I'm not getting comparable fps with people who have an almost identical system and I'm using their configs". Also the fact that even though I bought this card so that I could play qw AND newer games with the same card, I was expecting a bit more bang for my buck than what the initial FPS readings were.

The fps drop from dying, is because the scoreboard is drawn ye?

GTX460 is actually around as fast as GTX285 in most of the games, plus it has DX11/OpenGL4 support. Older games can actually be faster with GTX200 cards (GTX260/275/285) because of few things. For example:


GTX285: (core: 648MHz, shaders: 1476MHz, memory: 2484MHz)

pixel fillrate: 20,7 Gpixels/s
texture fillrate: 47,4 Gtexels/s
memory bw: 159 GB/s


GTX460: (core: 675MHz, shaders: 1350MHz, memory: 1800MHz)

pixel fillrate: 21,6 Gpixels/s
texture fillrate: 37,8 Gtexels/s
memory bw: 115 GB/s


The numbers don't tell everything (like efficiency), but most of the games that are just multitextured and don't use pixel shading, will be faster with GTX285. GTX460, even if it has less memory bw, handles anti-aliasing modes better (less performance hit). GTX460 is very good/efficient for DX11 games and especially good when tesselation is being used, so it's kinda aimed towards DX11 and OpenGL 4.0.

so basically a gtx285 would be better for qw performance? =)