ASUS P3C 2000 Camino Board - Page 6

System Performance

a. Overclockability

As one can tell from my system specs, I managed to get my PIII550E @ 803MHz stable by altering the FSB in SoftBIOS to 146MHz, which also renders a SDRAM speed of 109MHz. To achieve this, I had to increase my VCore slightly to 1.65V from BIOS.

Actually, my PIII550E worked fine up to 825MHz (5.5 x 150) at default 1.60V for normal Windows usage (and even a few hours of Q3A). However, during more intense apps (like 3DMark2000), it would become unstable and randomly kick me out to Windows desktop. I originally thought it was my outdated I820 Inf file that was the issue (being pre-advised earlier).  But even after updating to the latest INF from INTEL’s developer’s site (dated Nov’99), it still rendered similar results.

It was only after I reduced from 825MHz to 803MHz (at 1.65V), did the intermittent “kick-backs” become non-existent. Now, based on the above results, the board should have no difficulties satiating most inane CPU overclockers.

As a side note, I had no further gains in altering MTH/MCH/VIO jumper settings for both CPU / DDR GeForce overclocking. In any case, I was already happy with default overclock speeds of 803MHz, which offered a decent boost in performance, albeit my DDR mem speeds had to be reduced slightly. 

b. Benchmark Results

Sisoft Sandra’99 v3.5
Running several benchmarks modules off Sisoft Sandra’99 rendered the summary graphs below. Note that for the drive benchmarks, I benched my existing HotRod ATA66 controller card’s transfer rate against the onboard ATA66 controller.

The onboard ATA66 controller shows consistent results after extrapolation, but my existing HotRod card shows extremely lackluster performance, even slower than normal ATA33 speed. Seems this is an inherent problem with the HighPoint controller on my HotRod card (so stay away from the HotRod, as my buddies Wilfred & Kai Ping will unanimously agree)

 

 

Almost all benchmarks indicate an outstanding performance from the overclocked Coppermine system. However, on reaching the memory benchmark in particular, all rosy impressions turn gravely sour. Look at the graph below for its anomalous and faltering memory performance. 

Here, we witness the weakness in using the Memory Translation Hub of the Camino, where approximately 20% degradation in performance is incurred (compared to a standard BX system running PC100 SDRAM) owing to added latency from the RAMBUS-SDRAM adaptation. Do note that my SDRAM onboard is overclocked to 109MHz, which would then make the performance discrepancy even more pronounced.

Winbench99
I have attached a summary of my Winbench99 results below in comparison to the Athlon system I had reviewed earlier:

Judging from the raw benchmarks above, a jump between 700MHz and 803MHz is significant. However, in reality I couldn’t really tell any stark difference during normal Windows activities / apps. It did feel slightly nippier at 803MHz, but that’s about it.

Winstone99 v1.2
Under Winstone 99, the PIII 550E @ 803MHz achieved a Business Winstone score of 29.7 Winstone units. This is right up there with the rest of the PIII scores around on the web.  In comparison, a PIII 733MHz based HP Pavilion 8595c (VIA Apollo Pro 133A chipset) produced a slightly lower Business Winstone score of 29.4 Winstone units.

Quake 3 Arena Retail (patched to v1.15c)
(LightMap / Max Textures / Trilinear Filtering / Marks On Walls / Eject Brass)

As one can tell, the higher latency introduced by the MTH and lower clocked DDR GeForce card, has invariably degraded performance especially at higher resolutions, as compared to what I can typically achieve (about 60 fps at 1024x768 / 32bit) on my DDR GeForce in a similar BX + Coppermine combo. Even the presence of AGP 4x + Fast Writes couldn’t make up for the difference.


The Memory Translator Hub: Culprit of slow memory performance

In addition, even with AGP 4x + Fast Writes On, it still failed to show significant improvement in larger-textured Q3 levels. Running Hero’s Keep in 32bit / 1024x768 / Max textures, still exemplified unplayable frame-rates at certain points of the level. So if you hope this board would do wonders via AGP 4x in Q3A, I think you should consider otherwise.

This is in part, a result of the limited memory bandwidth offered by PC100 SDRAM (PC133 SDRAM won’t be very much better). Hence, I feel AGP 4x would inevitably become the “White Elephant” feature and a pure “Marketing Gimmick” for this board and many others based on PC100/133 SDRAM.  Hopefully, with RDRAM or even the upcoming DDRAM renditions, this may turn out to be a different story.

3D Mark 2000
In 3D Mark 2000, I only benchmarked using 32bit res & textures, to stress the card to its max and also test for stability. (Tested with 24bit Z-Buffer; Triple Frame Buffer)

[Link to 3DMark databases]

In comparison, I believe the same o/c Coppermine at 800MHz (though with a slightly higher DDR mem setting) under a BX config, typically hits about mid-4000s 3D Marks in 1024x728/32bit. Again, the lower overclocked DDR settings + weaker memory performance has invariable tainted and degraded performance slightly. But I guess the figures are just about proportional.

< Previous

Next >

 

Content