Express performance 1100 Vs 1200

[Thetargos]

Is it normal that an eXpress 1100 is faster than an X1200? I know both graphics chips are at the lowest end of the spectrum, especially being IGPs, but on laptops there's not much of a choice (especially down here). At any rate, I'm really surprised that an X1100 is almost twice as fast as an X1200 with the latest drivers. That can't possibly be normal, or is it? Maybe there's something wrong with my setup?

Edit

Performance measured through the nothing specific and absolutely not sensitive glxgears app on two laptops were amazing...

The laptops hardware is similar, one is an Acer Aspire 5050-4697 (X1100) and a Toshiba Satellite A215-S7437 (X1200)

To rule out the difference in CPU speeds in the laptops, I locked the CPU speed of both machines to 800MHz (their lowest rate by K8PowerNow), and ran glxgears for 25 seconds (i.e 5 FPS counts), after only 5 seconds the differences are remarkable:

X1100: 1170 FPS
X1200: 529 FPS

[bridgman]

I have never looked hard at the internals but my understanding has always been that the 1200 (690) should be faster in 3D than the 1100 although I guess there are cases where it could be slower. I don't need to mention that glxgears is not a real typical graphics app.

The 1200 has 4 pipes vs 2 pipes on the 1100, and is either similar or faster in most other respects, but the 1200 does access memory through the HyperTransport link which is fast but probably has higher latency. I think the 1200 should be faster on serious graphics tasks (games for example). If it's not, let us know.

[Extreme Coder]

You have a wierd GLXGears rate with the X1200..
My X1200 reports this on GLXGEARS:
5571 frames in 5.0 seconds = 1114.147 FPS
5701 frames in 5.0 seconds = 1140.181 FPS
5684 frames in 5.0 seconds = 1136.707 FPS
5703 frames in 5.0 seconds = 1140.428 FPS
5699 frames in 5.0 seconds = 1139.641 FPS

[Thetargos]

Thanks to both John and Extreme Coder. I will perform some gaming tests tonight and report back to this same thread. Just for the record, the X1200 is running on:

Toshiba Satellite A215-S7437
CPU AMD Turion X2 TL-58 1.9GHz
RAM 2 GB 667MHz
X1200 RS690 as reported by lspci, and amdcccle

Software:

OS: Fedora 8
Kernel: 2.6.24.3 (distro stock)
fglrx: 8.47.3 (8.3)
Arch: x86_64

One thing I didn't mention in the original post was that both laptops are running 64-bit Fedora 8, the Acer (X1100) is running Livna's packaged fglrx driver (same version), and the Toshiba (X1200) is running AMD's reference fglrx driver. Both laptops are fully updated to distro's last batch of updates. Will report back with some gaming tests.

[c247]

I'm running XFCE with AIGLX off ( at the time of the test a firefox that has been playing flash videos for hours is running, along with mysql query browser amarok others ) Linux rules

2958 frames in 5.0 seconds = 591.514 FPS
4501 frames in 5.0 seconds = 900.037 FPS
4524 frames in 5.0 seconds = 904.655 FPS
4627 frames in 5.0 seconds = 925.294 FPS
4254 frames in 5.0 seconds = 850.777 FPS
4274 frames in 5.0 seconds = 854.714 FPS
-- Here I dragged glxgears to a fullscreen (almost 1440x900) window --
713 frames in 5.0 seconds = 142.566 FPS
356 frames in 5.0 seconds = 71.155 FPS
360 frames in 5.0 seconds = 71.936 FPS
362 frames in 5.0 seconds = 72.350 FPS
356 frames in 5.0 seconds = 71.096 FPS
355 frames in 5.0 seconds = 70.892 FPS

I played urbanterror this morning after yum installing the latest fglrx packages from livna.fc8 It looked awesome, and played semi-smooth with the games default settings. Fiddle with the game settings and it gets very smooth and usable.

Barebones desktop I built:
Gigabyte GA-MA69G-S3H
2 gig ram
IGP uses 128MB of that
Using all the other integrated devices (audio, network,etc)

BTW, using the previous fglrx packages I was having hard locks where I would have to hard reboot using the reset button. I hadn't had this problem until I upgraded from fedora 7 to 8. I'm starting to believe that uninstalling everything video related and starting from default x config helped more than getting the latest packages.

[Thetargos]

Ran some gaming "benchmarks". Nothing definitive, as I'll go from bottom up. I ran Quake 3 (ioquake) on both machines (did not lock the CPU speed) and the difference as at first of about 20FPS (favoring the 1100), after modifying the config so the com_max_fps variable was 999 the results were more like ~15 FPS appart. The first run was:

Errata

I should have mentioned that the screen resolution I used for the tests was 856x480.


| run 1 | run 2 |
--------|-------|-------|
X1100 | 111.5 | 110.9 |
--------|-------|-------|
X1200 | 88.5 | 95 |
-------------------------

I know this game is not much of a demanding application or accurate, particularly for the features. I was wondering which other games could I test that more accurately measure performance. I attribute the performance delta in my preliminary tests due to one CPU being single-core 2.2 GHz and the other a dual-core CPU, but 1.9 GHz. I'm not sure if the 300MHz difference justifies the FPS difference (supposing the X1200 is beefier). Will test again with the CPU speeds locked to 800MHz, and see how big the performance difference this time is.

Now, if there is another game that I should test, and you could recommend (I was thinking about UT2004 Demo, but I fear it'd be too "heavy") I'd be most grateful.

Edit

Well, I just ran Q3A with the CPUs locked @800MHz, there is still a difference:


q3a@800 | run 1 | run 2 |
--------|-------|-------|
X1100 | 56.1 | 56.2 |
--------|-------|-------|
X1200 | 45.5 | 45 |
-------------------------

The difference is small, but still the X1200 is consistently under the figures of the X1100. Now IIRC the X1100 has the 3D engine from an R400 without vertex engine and reduced pixel shaders (the actual code number is RS485)

[Gretschy]

My Catalyst Control Center reports "PCI" as Bustype for my x1200 inside my notebook. Is this correct? I thought it should be pcie.
BTW... i get around 930 FPS in glxgears
Thanks, Fabian

[Thetargos]

You are absolutely right! I also see as Bustype PCI instead of PCI-Express.

Edit

By the way, Control Center in Windows Vista shows the same, PCI instead of PCI-E or similar.

[Kano]

If you want to test raw gfx speed with quake3 add a - behind the name of the demo like:

timedemo demo1-

[Thetargos]

If you want to test raw gfx speed with quake3 add a - behind the name of the demo like:

timedemo demo1-

Thanks for the suggestion, Kano.

Edit

I tried running the Demo like that, but I got error messages about not being ale to open the demo file... Will try with that exact syntax and see how it goes (it should be something like timedemo four-, right? Since the demo file included with ioquake is the "four.dm_68" demo)

[Kano]

Hmm maybe it was -four then.

[Thetargos]

Will try that. Thanks.

[damentz]

I noticed that you are underclocking the cpus, which creates a cpu bottleneck while benchmarking. You should run the cores at their full designed speed otherwise your benchmarks are worthless.

For all I know, one laptop's motherboard and processor combo could be more efficient than the other, thus the source of incorrect benchmarks.

If you really need to test them, run the games at the highest resolution possible and disable features that are cpu intensive, such as shadows and other software powered doodads.

[Thetargos]

The idea to lock the CPUs was precisely to create that bottleneck. As the bottleneck would yield more representative figures of raw graphics performance. Also one CPU is faster than the other, hence it helps to test with a common CPU speed. Both systems share even the system's chipset and memory type, brand and speed. So creating that bottleneck actually rules out any CPU aid.

[Thetargos]

Rather than opening a new thread about this very same issue, I thought I'd add a bit more information here and hopefully have some feedback from John or anyone else knowledgeable.

I ran 'lspci -vvnn' on my laptop and here's the output from it. Maybe this can tell something to someone about the lousy performance of fglrx on this setup...

01:05.0 VGA compatible controller [0300]: ATI Technologies Inc RS690M [Radeon X1200 Series] [1002:791f] (prog-if 00 [VGA controller])
Subsystem: Toshiba America Info Systems Unknown device [1179:ff1a]
Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 64, Cache Line Size: 32 bytes
Interrupt: pin A routed to IRQ 11
Region 0: Memory at f0000000 (64-bit, prefetchable) [size=128M]
Region 2: Memory at f8100000 (64-bit, non-prefetchable) [size=64K]
Region 4: I/O ports at 9000 [size=256]
Region 5: Memory at f8000000 (32-bit, non-prefetchable) [size=1M]
Capabilities: [50] Power Management version 2
Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-)
Status: D0 PME-Enable- DSel=0 DScale=0 PME-
Capabilities: [80] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable-
Address: 0000000000000000 Data: 0000
Kernel driver in use: fglrx_pci
Kernel modules: fglrx

At this point I'm pulling my hair off at how lack luster performance is with the fglrx drivers on this machine and how fast it is on "less capable" hardware (I'm starting to assume that the less capable hardware is actually orders of magnitude more capable than this hardware). However it does strike me as surprising that all in all Compiz-Fusion performance is not THAT bad, and is actually rather good... However, for anything else it is really bad, be it real-time rendering for games, or for visualization purposes (which I need for work). To top it off, I can't even run Compiz all the time due to the XVideo issue with it (even though at least fullscreen works).

The fact that some users besides me have also reported similar performance issues with allegedly the same graphics chip (I assume there may be slight configuration discrepancies depending on how the manufacturers actually integrate the chips into the rest of the system), I think that identifying what is causing this reduced performance will eventually help other users as well. So anything I can provide to help narrow down the issue?

[Melcar]

That's some really low performance. This is what I get in glxgears running in a window on a 200M:

5066 frames in 5.0 seconds = 1013.075 FPS
5459 frames in 5.0 seconds = 1091.701 FPS
5422 frames in 5.0 seconds = 1084.331 FPS
5462 frames in 5.0 seconds = 1092.340 FPS
5422 frames in 5.0 seconds = 1084.242 FPS
5460 frames in 5.0 seconds = 1091.860 FPS
5415 frames in 5.0 seconds = 1082.974 FPS
5464 frames in 5.0 seconds = 1092.792 FPS
5418 frames in 5.0 seconds = 1083.551 FPS
5408 frames in 5.0 seconds = 1081.471 FPS
.. and this are my Graphics scores (http://global.phoronix-test-suite.com/index.php?k=profile&u=togo-19197-13316-24976).

[Thetargos]

That's some really low performance. This is what I get in glxgears running in a window on a 200M:

5066 frames in 5.0 seconds = 1013.075 FPS
5459 frames in 5.0 seconds = 1091.701 FPS
5422 frames in 5.0 seconds = 1084.331 FPS
5462 frames in 5.0 seconds = 1092.340 FPS
5422 frames in 5.0 seconds = 1084.242 FPS
5460 frames in 5.0 seconds = 1091.860 FPS
5415 frames in 5.0 seconds = 1082.974 FPS
5464 frames in 5.0 seconds = 1092.792 FPS
5418 frames in 5.0 seconds = 1083.551 FPS
5408 frames in 5.0 seconds = 1081.471 FPS
.. and this are my Graphics scores (http://global.phoronix-test-suite.com/index.php?k=profile&u=togo-19197-13316-24976).
My numbers are actually very similar, but with an RS690M (X1200). If I lock the CPU to 800MHz, at most I get 525fps in glxgears. Will have to install PTS and Nexuiz and Doom3 on this laptop to check the performance of the benchmarks.

[Melcar]

This is glxgears with the CPU capped at 800MHz:

3156 frames in 5.0 seconds = 631.086 FPS
3445 frames in 5.0 seconds = 688.851 FPS
3469 frames in 5.0 seconds = 693.660 FPS
3405 frames in 5.0 seconds = 680.977 FPS
3497 frames in 5.0 seconds = 699.376 FPS
3673 frames in 5.0 seconds = 734.493 FPS
3803 frames in 5.0 seconds = 760.547 FPS
I believe that both IGPs share similar specifications. Despite the PTS scores, this laptop can game at low settings rather well (AA 2008, Urban Terror, Nexuiz). Compiz ran fine when I used it (except for the more demanding plugins and effects) and runs decently with the KDE transparencies (a bit on the sluggish side when I start moving windows around, though). Back when I bothered to keep Windows on this laptop I used to run opengl benchmarks and starting with the 7.12 drivers performance got pretty close between the two, so I doubt the problem is the drivers not performing, but rather the chip itself (it's an IGP after all).

[Thetargos]

My fiancé's laptop is an Acer with an RS485, aka eXpress 1100. However, in that setup, it actually makes little difference whatosever if the CPU is locked to 800MHz or set to on-demand. In that setup on-demand yields almost 1800FPS in glxgears, in 800MHz it yields 1200FPS. On my setup performance is almost cut in half if I lock the CPU speed to 800MHz, unlocked it gives 1070FPS at most.

[Thetargos]

Here's the lspci -vvnn VGA section of the machine with the Xpress 1100:

01:05.0 VGA compatible controller [0300]: ATI Technologies Inc RS485 [Radeon Xpress 1100 IGP] [1002:5975] (prog-if 00 [VGA controller])
Subsystem: Acer Incorporated [ALI] Unknown device [1025:010f]
Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 255 (2000ns min), Cache Line Size: 32 bytes
Interrupt: pin A routed to IRQ 21
Region 0: Memory at c8000000 (32-bit, prefetchable) [size=128M]
Region 1: I/O ports at 9000 [size=256]
Region 2: Memory at c0100000 (32-bit, non-prefetchable) [size=64K]
[virtual] Expansion ROM at c0120000 [disabled] [size=128K]
Capabilities: [50] Power Management version 2
Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0-,D1-,D2-,D3hot-,D3cold-)
Status: D0 PME-Enable- DSel=0 DScale=0 PME-
Kernel driver in use: fglrx_pci
Kernel modules: fglrx, radeonfb

Hope this is helpful.

[Kano]

Maybe install nexuiz and use

nexuiz -benchmark demos/demo1

to compare. Then you use at least a "real" game not just glxgears which is useless.

[Shae]

To put them on a more even level, maybe you should disable one of the cores on the dual core machine by passing nosmp in the kernel line in grub. You should also lock the speed at 800MHz and turn off the CPU intensive features as mentioned before.

[Thetargos]

Maybe install nexuiz and use

nexuiz -benchmark demos/demo1

to compare. Then you use at least a "real" game not just glxgears which is useless.

I had to re-install the Acer one (single-core Xpress 1100) in 32-bits. I'll conduct a test with Nexuiz as soon as I have the system fully updated and what not.

Shae:

I have already run similar tests disabling SMP, virtually identical results.

By the way, one thing I have noticed about glxgears (and while we all agree it is NOT a real test, more like a demo) the fact that it does not use any actual texture data, but only colors to draw the gears on screen, should yield much faster renderings.

Last time I tested with ioquake3, the results were a consistent ~20 FPS delta using the default demo. Nexuiz is a bit more demanding a game, so it should give a better picture at which system is actually more capable (bloom, shadows, HDR, glare, etc). As soon as I can, I'll test it.