What is the expected fps reading from glxgears?

[chrisr]

Hi,

I have a HD 4650 connected to a 1920x1200 LCD monitor, running with fglrx 9.3. The 4650 is installed within a docking station whose PCIe slot is only capable of x1. Upgrading to a later version of fglrx is impossible unless I want to kill my laptop's video when I undock it.

The HD 4650 plays World of Warcraft in OpenGL mode, but at such a distressingly low frame rate (< 20fps outside the bank in Ironforge) that I tried something really simple like running glxgears. glxgears gives me ~1000fps, which is considerably lower than I was expecting for RV730 hardware. (Come to think of it, the 2D desktop behaviour isn't particularly amazing either.)

I was expecting the PCIe x1 slot to cause some loss in performance, but this looks beyond what I had in mind. Is this "correct" behaviour for this hardware/driver combination, please? Does anyone have any similar hardware for comparison?

Thanks,
Chris

P.S. I have tried attaching my HD 4650 to my TV via a S-Video cable, but when I do this the S-Video automatically becomes the primary display and sets the resolution for both monitors to 1024x768. Is there any way for the LCD monitor to become "Display1" instead?

[nanonyme]

glxgears gives me ~1000fps, which is considerably lower than I was expecting for RV730 hardware. Glxgears is not a benchmark.

[chrisr]

Glxgears is not a benchmark.

No it's not. But a) everyone has it, b) it's quick and easy to run, and c) its performance is related to both the graphics hardware and driver implementation. So if people with similar hardware are getting disimilar results then that would merit further investigation.

[Ex-Cyber]

No it's not. But a) everyone has it, b) it's quick and easy to run, and c) its performance is related to both the graphics hardware and driver implementation. So if people with similar hardware are getting disimilar results then that would merit further investigation.In that respect, it's a very rough diagnostic. If there's a huge difference, it might indicate a configuration problem. That kind of use is uncontroversial. However, pretty big differences (even 2x or more) have been known to occur between different working configurations without anything close to a proportional effect on actual application performance.

As for the performance being related to "both the graphics hardware and driver implementation", part of the problem is that it's also related to color depth, window size, CPU load, number of running processes, compositor configuration, kernel configuration, and so on. 1000FPS is the right order of magnitude; trying to pin things down closer than that is pretty pointless and tells you nothing about how real applications will run.

[chrisr]

1000FPS is the right order of magnitude; trying to pin things down closer than that is pretty pointless and tells you nothing about how real applications will run.

Perhaps you would care to suggest an alternative, then? Maybe a relevant subset of x11perf tests, since (I believe) the R700 series uses 3D hardware for 2D acceleration?

[adamk]

http://dri.freedesktop.org/wiki/Benchmarking

The openarena benchmark is my personal favorite.

Adam

[nanonyme]

No it's not. But a) everyone has it, b) it's quick and easy to run, and c) its performance is related to both the graphics hardware and driver implementation. So if people with similar hardware are getting disimilar results then that would merit further investigation.Well, you can easily get thousand or more fps in glxgears by buying a faster CPU so meh. ;) Seriously, rather run any game as a benchmark.

[nanonyme]

1000FPS is the right order of magnitude; trying to pin things down closer than that is pretty pointless and tells you nothing about how real applications will run.With modern PC hardware 300fps-30000fps is probably the right order of magnitude... There is no normal ratio for a single card, just what a particular computer setup manages to do. We're talking about a (mostly, motherboard bandwidths and memory bandwidths probably play their part too) CPU-bound technical demo, not a GPU benchmark.

[whizse]

Perhaps you would care to suggest an alternative, then? Maybe a relevant subset of x11perf tests, since (I believe) the R700 series uses 3D hardware for 2D acceleration?

x11perf is not recommended:
http://cworth.org/intel/performance_measurement/

[Qaridarium]

Hi,

I have a HD 4650 connected to a 1920x1200 LCD monitor, running with fglrx 9.3. The 4650 is installed within a docking station whose PCIe slot is only capable of x1. Upgrading to a later version of fglrx is impossible unless I want to kill my laptop's video when I undock it.

The HD 4650 plays World of Warcraft in OpenGL mode, but at such a distressingly low frame rate (< 20fps outside the bank in Ironforge) that I tried something really simple like running glxgears. glxgears gives me ~1000fps, which is considerably lower than I was expecting for RV730 hardware. (Come to think of it, the 2D desktop behaviour isn't particularly amazing either.)

I was expecting the PCIe x1 slot to cause some loss in performance, but this looks beyond what I had in mind. Is this "correct" behaviour for this hardware/driver combination, please? Does anyone have any similar hardware for comparison?

Thanks,
Chris

P.S. I have tried attaching my HD 4650 to my TV via a S-Video cable, but when I do this the S-Video automatically becomes the primary display and sets the resolution for both monitors to 1024x768. Is there any way for the LCD monitor to become "Display1" instead?

install 9-10 catalyst and you will have 40 000 frames in 5 seconds in glxgears.. (i have this on my system)

[forum1793]

Several times now I've seen someone broadcasting something about catalyst 9-10. But I only see 9-8 at AMD's site. How are you getting 9-10?

[Ant P.]

`LIBGL_ALWAYS_SOFTWARE=1 glxgears` gives you software Mesa fps, `glxgears` should either give you either way more or exactly 60 fps.

[bridgman]

Several times now I've seen someone broadcasting something about catalyst 9-10. But I only see 9-8 at AMD's site. How are you getting 9-10?

A version of the same base release that will be used for Catalyst 9.10 was included in the latest Ubuntu 9.10 update.

[Qaridarium]

Several times now I've seen someone broadcasting something about catalyst 9-10. But I only see 9-8 at AMD's site. How are you getting 9-10?

install ubuntu 9.10... its the driver for ubuntu 9.10!

best catalyst ever! kernel 2.6.31 support! and a lot of improvements

[Kano]

The driver is not for U 9.10, thats the Xserver 1.5+ part from a normal ati-installer package - the Xserver 1.4 and lower part was stripped from the orig file, so it is not really "original" in the normal meaning. If you want to use the created packages then you would need just any Debian (with dkms installed) or Ubuntu with Xserver 1.5 or newer. That means Debian sid or Ubuntu since intrepid/9.04.

[cutterjohn]

`LIBGL_ALWAYS_SOFTWARE=1 glxgears` gives you software Mesa fps, `glxgears` should either give you either way more or exactly 60 fps.Actually that's not quite correct, if you are referring to exactly 60FPS as in VSYNC is enabled in the driver. The actual FPS if VSYNC is enabled should be exactly whatever the display frequency is, as the idea is to have the GPU have a new frame ready to go when the display refreshes it screen.

Anyways, yeah glxgears and fgl_glxgears are pretty useless as benchmarks, but they are a q&d way to see if the driver is behaving mostly normally, i.e. you get better relative *gears performance with the driver properly installed and loading, and if you peg display freq, lets you know that VSYNC has been enabled. (I just disabled VSYNC and let apps decide, as a video player would want to VSYNC whereas a game you may not want VSYNC...)

Anyways, with a 512MB GDDR3 Radeon Mobility 4850 (500/550) I get about 8000 FPS glxgears, and c. 2200 FPS fgl_glxgears w/catalyst 9.9

As to with a fast CPU == high glxgears, I seem to remember always having <<1000 FPS with various 2.4GHz multi-core CPUs(P8600, 4800+ X2)

I always like games as benchmark tools as recent games will tend to stress the system more than ANY other app will as a whole. The problem is that only some games will, either really cutting edge designed for the futures games or really badly written/poorly optimized ones. (I'm entirely unimpressed by synthetic benchmark scores, e.g. Vantage, SiSoft, etc. as raw scores, but they can be helpful when doing a high level comparison of say various CPU models & architectures, but in the end the games are what are rely on for the bottom line. )

[nanonyme]

As to with a fast CPU == high glxgears, I seem to remember always having <<1000 FPS with various 2.4GHz multi-core CPUs(P8600, 4800+ X2)Well, AMD64 3500+ -> Intel Core 2 Duo E8500 caused a leap from 1000fps to 2000fps for me. Go figure. (might not as direct with closed drivers though since they use interrupts and stress the CPU less whereas open drivers currently do polling; there should be clear boundaries though which are impossible to exceed without buying a faster CPU if we assume optimal drivers since it seriously uses a lot of CPU still) Then again, with open drivers with interrupts you should be getting vsync in KMS by default. :D