Radeon 8500 and memory bandwidth.

[JSSheridan]

As the video card market stands right now, the Radeon 8500 is the budget card to buy. That will probably be changing in a month however as the newest releases from nVidia and ATI find there way to the shelves. For now though, it has become almost impossible to find 8500 cards from OEM's that are clocked as high as the Built-by-ATI versions were produced. For anyone who has considered buying a 8500, I ran some test on my Built-by-ATI Radeon 8500LE 128MB to help to show you what performance you might expect from the 8500's currently available.

System Information:

Windows XP Pro | Cat. 2.5 | DX 9
A7N8X-D | 2400+ @ 2.0GHz | PC2700 @ 133, CAS 2-2-2-4 | WD800JB

The core is clocked at 250MHz for each test. The video was set to "Optimal Performance" in the ATI control panel. I varied the memory speed between 187 and 295MHz to see how memory bandwidth affected the performance of the card. I used 3DMark2001 at default settings and UT2003 Demo at 1024x768. Here are my results.

Last edit: Added UT2003 Flyby benchmarks and graphs for the balanced and optimal quality settings.

Optimal Perfomance
Speed (MHz): 3DMark2001 | UT2003 Demo (Flyby / Botmatch);

187MHz: 8294 | 109.827 / 54.469;
200MHz: 8648 | 116.685 / 56.365;
217MHz: 8801 | 119.807 / 56.938;
232MHz: 8964 | 123.284 / 57.297;
250MHz: 9025 | 123.836 / 57.088;
266MHz: 9200 | 127.106 / 57.146;
275MHz: 9260 | 127.795 / 57.395;
295MHz: 9336 | 129.580 / 57.257;

Graph for 3DMark2001
Graph for the UT2003 Flyby

Balanced
Speed (MHz): 3DMark2001 | UT2003 Demo (Flyby / Botmatch);

187MHz: N/A | 093.064 / 51.837;
200MHz: N/A | 098.064 / 53.461;
217MHz: N/A | 103.674 / 54.445;
232MHz: N/A | 107.666 / 55.074;
250MHz: N/A | 110.161 / 55.667;
266MHz: N/A | 111.780 / 55.991;
275MHz: N/A | 113.778 / 56.250;
295MHz: N/A | 116.270 / 56.931;

Graph for the UT2003 Flyby

Optimal Quality
Speed (MHz): 3DMark2001 | UT2003 Demo (Flyby / Botmatch);

187MHz: N/A | 44.458 / 29.415;
200MHz: N/A | 48.266 / 32.412;
217MHz: N/A | 50.200 / 33.964;
232MHz: N/A | 52.016 / 34.837;
250MHz: N/A | 53.514 / 36.176;
266MHz: N/A | 24.709 / 37.050;
275MHz: N/A | 55.242 / 37.173;
295MHz: N/A | 56.319 / 38.129;

Graph for the UT2003 Flyby

Hope you can find this useful. If you would like anymore information about my system, let me know. Peace.

 

[titanmiller]

Thanks for the benchmarks JSSheridan. Interesting.

A little bit off topic but is there anything to say that card manufactures cant make a FX5200 Ultra (Ahk, I said Pro first time through) and smack on some memory that can run at 400mhz (800mhz) to boost the preformance?

 

[JSSheridan]

I don't believe so. I think there are only minimum requirements that the card makers are expected to meet. That's why you could have Golden Sample cards that would be released a few months after the standard card. Those promised better overclocks because the memory was faster and were targeted at overclockers. Peace.

 

[Eug]

Thanks. Useful information.

I too found that varying memory speeds did change the benches on my Radeon 9100, but it never seemed like much of a change in real gameplay at my settings (which are 1280x1024 with most of the stuff turned down, on a Celeron 1.4). It seemed to me that changing the core speed made more of a difference. Perhaps at these settings, the card is not memory bandwidth limited. Your botmatch results seem to support that contention.

Kinda moot on my card, since I can't get the 3.5 ns memory much past its stock 230. My core overclocks easily to 275 from 250 though. I haven't tried any higher yet.

 

[JSSheridan]

I think that at 250MHz core, the 8500 doesn't scale with bandwidth after around 230MHz. Since it was at performance settings however, it wouldn't use as much bandwidth as it would at balance or quality either. The UT2003 Botmatch is mostly CPU limited at 1024x768. I've uploaded a couple of graphs.

Graph for 3DMark2001
Graph for the UT2003 Flyby

If I have time this weekend, I will upload benches for balanced and optimal quality settings also. Peace.

 

[JSSheridan]

Added UT2003 Flyby benchmarks and graphs for the balanced and optimal quality settings. The balanced setting responds to added bandwidth better than the quality setting. The quality test are GPU bound almost from the beginning, so I will oc the core to 295MHz and run that set of test again. That and 3DMark2001 numbers coming tomorrow afternoon, maybe. Peace.

 

[HappyNic]

crucials has 8500 le at 250/166 can you do bench on that one?

 

[VirginiaDonkey]

Very nice job!

is there any reason in paticular you used cat 2.5 vs cat 3.1?

I have noticed wih my 8500 that the memory performance starts to degrade after 315mhz. I have the 8500LE with the 3.3ns memory. Also overclockability depends on the bios you have. from everything I have read, the original 1009 bios and the 7006 bios seem to have the highest scores (maybe more agressive timings?)

 

[JSSheridan]

I can't go quite that low on my memory speeds. If I go much lower than 187. I get some screen corruption even in 2D mode, but I'll keep trying to get it lower. I'm using Cat. 2.5 since Mafia doesn't work too well when I have the Cat 3.x's installed. Peace.