MX440 PCI and 9000 PCI benchies

[Mloot]

I received the Powercolor 9000 PCI today and ran it against the Inno3D MX440 PCI I was using at the time. The 9000 is a low profile card, not as tall as the 440. It also has a passive heatsink, compared to the 440's HSF. Both cards have 64mb of DDR RAM. The 9000 has CRT/DVI/TV out and the 440 has only CRT/TV out. For the testing, I used the latest Catalyst drivers (from ATI's webiste) on the 9000 and Nvidia's 40.72 Detonator drivers on the 440. You can view my rig specs in my sig. I tried playing Dungeon Siege on both cards, and the 440 was faster than the 9000 by about 10fps. The 440 was also faster on several of the tests, especially the flybys. It pretty much spanked the 9000 on those. I did notice, however, that at least on 3DMark, the 9000 did bring out more detail in the picture. As far as 2D performance, I really couldn't tell a difference. Perhaps I should have run more benchies (for Deeko at least), but the real world intruded on my tests (namely my kids demanding attention). I ran 3DMark2001SE twice on the 9000, at stock speed and overclocked speed (using 3dRage tweaker). I wanted to see if overclocking the 9000 would yield any performance benefits. I ran the 440 at stock speeds only, because I determined a couple of weeks ago that oc'ing the 440 had absolutely no effect. Perhaps it was due to mediocre RAM on the card, or perhaps the PCI bus speed kills any overclocked performance gains. The UT2K3 demo was run only at stock speeds. The oc'ed UT2K3 results for the 9000 were identical to the stock speed, so I included only the one. I suppose I will keep the 9000 because it has more display options and I plan on adding a DVD drive to my pc in the future and using the TV/out. I am counting on ATI's reputation for superiority in their TV/out and DVD playback. I'll have a hard time giving up the 440, though. I've become somewhat attached to it. Anyhoo, that's enough rambling. On to the benchies:


MX440 (270/400) at default settings of 1024x768x32:

3DMark Score 3536

Game 1 - Car Chase - Low Detail 84.3 fps
Game 1 - Car Chase - High Detail 21.8 fps
Game 2 - Dragothic - Low Detail 50.1 fps
Game 2 - Dragothic - High Detail 20.2 fps
Game 3 - Lobby - Low Detail 72.9 fps
Game 3 - Lobby - High Detail 31.2 fps
Game 4 - Nature Not supported by hardware
Fill Rate (Single-Texturing) 449.5 MTexels/s
Fill Rate (Multi-Texturing) 860.7 MTexels/s
High Polygon Count (1 Light) 18.5 MTriangles/s
High Polygon Count (8 Lights) 6.4 MTriangles/s
Environment Bump Mapping Not supported by hardware
DOT3 Bump Mapping 34.4 fps
Vertex Shader 4.4 fps
Pixel Shader Not supported by hardware
Advanced Pixel Shader Not supported by hardware
Point Sprites 9.4 MSprites/s


Radeon 9000 (250/380): Also at 1024x768x32

3DMark Score 4004

Game 1 - Car Chase - Low Detail 57.9 fps
Game 1 - Car Chase - High Detail 23.4 fps
Game 2 - Dragothic - Low Detail 61.7 fps
Game 2 - Dragothic - High Detail 37.2 fps
Game 3 - Lobby - Low Detail 64.3 fps
Game 3 - Lobby - High Detail 31.2 fps
Game 4 - Nature 16.5 fps
Fill Rate (Single-Texturing) 283.8 MTexels/s
Fill Rate (Multi-Texturing) 973.4 MTexels/s
High Polygon Count (1 Light) 17.4 MTriangles/s
High Polygon Count (8 Lights) 4.4 MTriangles/s
Environment Bump Mapping 66.8 fps
DOT3 Bump Mapping 29.6 fps
Vertex Shader 33.3 fps
Pixel Shader 50.1 fps
Advanced Pixel Shader 44.3 fps
Point Sprites 7.6 MSprites/s


Radeon 9000 (o/c 325/450) same resolution as above

3DMark Score 4017

Game 1 - Car Chase - Low Detail 58.1 fps
Game 1 - Car Chase - High Detail 23.8 fps
Game 2 - Dragothic - Low Detail 61.5 fps
Game 2 - Dragothic - High Detail 37.4 fps
Game 3 - Lobby - Low Detail 64.7 fps
Game 3 - Lobby - High Detail 31.0 fps
Game 4 - Nature 16.5 fps
Fill Rate (Single-Texturing) 283.8 MTexels/s
Fill Rate (Multi-Texturing) 971.4 MTexels/s
High Polygon Count (1 Light) 17.4 MTriangles/s
High Polygon Count (8 Lights) 4.4 MTriangles/s
Environment Bump Mapping 66.8 fps
DOT3 Bump Mapping 29.6 fps
Vertex Shader 33.2 fps
Pixel Shader 50.2 fps
Advanced Pixel Shader 44.3 fps
Point Sprites 7.6 MSprites/s


UT2K3 Demo - Benchmarks

MX440 (270/400) at 800x600x32

9000 (250/380) at 800x600x32


BOTMATCHES:


dm-antalus (440)
12.433115 / 23.195126 / 47.225388 fps
Score = 23.206699

dm-antalus (9000)
2.383149 / 21.795589 / 47.621185 fps
Score = 21.807034


br-anubis (440)
12.819999 / 34.939400 / 75.155396 fps
Score = 34.940479

br-anubis (9000)
11.555211 / 32.781055 / 74.442085 fps
Score = 32.782658


dm-asbestos (440)
11.744531 / 33.621410 / 64.977249 fps
Score = 33.639816

dm-asbestos (9000)
14.129875 / 30.629017 / 60.693741 fps
Score = 30.646111


ctf-citadel (440)
9.666398 / 26.203394 / 58.013371 fps
Score = 26.247728

ctf-citadel (9000)
4.350246 / 24.284803 / 53.248810 fps
Score = 24.329578



FLYBYS:


dm-asbestos (440)
46.110634 / 113.006927 / 337.401794 fps
Score = 113.099304

dm-asbestos (9000)
46.592003 / 102.465843 / 311.117950 fps
Score = 102.582336


dm-antalus (440)
15.478856 / 86.139664 / 387.796997 fps
Score = 86.198776

dm-antalus (9000)
13.737094 / 58.379696 / 245.740601 fps
Score = 58.401047


ctf-citadel (440)
22.124439 / 70.649155 / 229.053223 fps
Score = 70.835602

ctf-citadel (9000)
18.830097 / 59.680187 / 222.750961 fps
Score = 59.840492

 

[Peter]

Good work! Looks to me as if the 9000 could have used a little more CPU power, otherwise performance would have changed when overclocked ... other than that, from the results it looks like the 9000 is the much more capable card, feature-wise.

 

[Jonestown]

Concerning the overclocking, there is a big thread about how the Radeon 9000(non-pro) will simply not overclock, even though the tweaker says it is. There is something in the 9000 pro driver and/or bios that is preventing or overriding the acceptance of the increased core/mem clock. People have gone into powerstrip and determined that the clock speeds are continually reset to stock.

No Radeon OC thread

Jonestown

 

[AnAndAustin]

Well the MX440 certainly seems the faster card, though not hugely faster, while the Rad9000 supports more functions and should be slightly better for image quality and TVout too. Since the Rad9000PRO can reportedly work passively I'm a little surprised we haven't seen these ported over to PCI yet (it'd be great to get top speed and features), but perhaps we are now reaching a point were PCI is simply limiting things too much in modern games and modern GPUs. It isn't nice having to make a compromise but it seems if you're stuck with PCI and can't / won't upgrade the mobo you can either get Rad7500 or GF4MX420 for cheap or else for the extra perf (and in Rad9000 features) the GF4MX440 and Rad9000 are fine choices. It is a bit of a shame that they cost as much as AGP versions of GF3, Rad8500 and in some cases GF4TI4200 but then PCI has seen it's day, just a shame that so many mobos were released lacking an AGP slot really.

 

[rogue1979]

Overclocking a video card by a significant margin and not seeing any tangible gains in fps, usually means your system is cpu limited. The Celeron 1.7GHz overclocked to 3GHz still had trouble keeping up with a 1.6a Northwood at stock speed, so I'm gonna guess it is a pretty low performer when running stock. I think the Radeon 9000 would scale a little better than the G4 440MX at higher cpu speeds. As far as overclocking the 9000, I simply use R3Dtweak3.6 and load the overclocker only. Then go into the registry under HKEYLocalMachine, software, game utility, Radeon overclocker. There are categories for default, actual and reset speeds. Simply open them and manually type in the core and memory overclock desired and you are ready to rock.

 

[13bells]

I have a Powercolor 9000 PCI, and it's clocked at 250/380 not the 250/400 specified on their website. This is not a big deal if you could change the settings, but you can't. I used powerstrip and underclocked as far as possible and ran 3DMark 2001. I got a lower score, but opening Powerstrip I found the memory settings had reverted back to 190 while the engine clock stayed at 125 that I set.

Any solutions anyone?

I recommend getting the Hercules 9000 PCI, at least they will deliver on their specs.

Powercolor FRAUD!!

 

[kenclops]

I was considering the Powercolor 9000, however, I'm seriously considering the 440 MX 128 MB video card to be the card to satisfy me for the next year. Is Cable Mart a reliable source for hardware? I notice that there is no return from them.

 

[Mloot]

Search for Cablemart, Inc. at www.resellerratings.com to get consumer reviews of various companies. I only saw two reviews for them, but at least both were positive. The 128mb PCI 440 should be a good card. What will be really interesting is a few months from now when Nvidia releases their new budget line (to replace the MX series), which should be DX8.1 compliant, if Innovision will produce a PCI version of whatever budget card(s) are released. Granted, that's alot of ifs, but it might be something to look forward to. If you get the 128mb 440, let us know how you like it.

 

[JAV1]

Mloot,

Very nice review! The MX440 seems to only have an advantage in Car Chase low detail & Lobby low in 3D Mark & in Dragothic & the high detail tests the 9000 is better. The other tests of 2-3fps are pretty insignificant & the 58-59fps are more than adequate for game play in Antalus & Citadel, IMO. Throw in the DX8.1 support & the Rad 9000 is the clear winner, IMO.

You've made my mind up > Rad9000 & it's better GQ & DX8.1 support over the faster in low detail & no DX8.x supported MX440.

Thanks & I followed you from MO, to TR to here to read this. Well worth the trip. [smile]

 

[Mloot]

Thanks. Just curious, I know TR, but what is MO?

EDIT: Duh, Madonion. Got it.

 

[trikster2]

What's TR?

Just wanted to thank Mloot for all the hard work.

Mloot was there any difference in 2D quaility at high resolutions (say 1600x1200 or whatever your monitor will support at a decent refresh rate)?

The ATI cards have a rep in some circles for being better at high res 2D, but I don't know if that is caried over into the power color cards.....