Overclockers Australia - VIA Apollo Pro266 Socket370 DDR Chipset

VIA Apollo Pro266 Socket370 DDR Chipset
Asus CUV266 and AOpen AX37 Plus - Page 3
23-July-01 - Review by: James "Agg" Rolfe

Testing configuration
Intel P3-667EB CPU
Corsair XMS2400 (DDR Boards)
Corsair PC133C2 (SDR Boards) (review here)
Enermax 350W PSU
Quantum Fireball +KA 13.6GB HDD
Asus V7700U GeForce2 Ultra video card

Intel i815e drivers v2.80.010A
VIA 4-in-1's v4.29
Detonator 650
DirectX 8.0a
CUV266 and AX37Plus using latest BIOS on vendor's website at time of writing.

Sandra and 3DMark benchmarks for the Transcend board and the P3V4X are taken from our review of the Transcend - we're using the same setup/versions so results should correlate.

Performance
As this is more a chipset review than just comparing these two motherboards, I thought comparing it to some SDR-based Socket370 chipsets would be interesting. Included in the graphs are results from 2 motherboards based on Intel's i815E chipset and one using VIA's older Apollo Pro 133A. First up, the mandatory Sandra 2001 benchmarks. These results are all averaged from 3 tests taken after a warm boot:

Sandra CPU

Nothing exciting on the CPU benchmark, as you'd expect..

Sandra Multimedia

I expected there to be some change in the multimedia test as this generally shows up memory speeds.. but no matter:

Sandra Memory

Hey, where's the boost in memory speed we were expecting? Sure, the CUV266 is ahead in memory-FPU speed, but there is not a lot in it across the whole range. The i815E boards are well up there - winning in memory-CPU, in fact. But the issue is not whether one chipset is a few points ahead of or behind the other - we're talking about an an evolutionary step in memory technology which carries a price premium (even if only on the motherboards) but is not having a noticeable effect on performance.

To focus on the memory performance, I tested the boards with StreamD2. This is a DOS-mode (boot, hit F8, select "Safe Mode Command Prompt Only") benchmark that performs a range of fundamental memory manipulations to give you an idea of the raw memory throughput of a system. I don't have results for the Transcend board because I'd already sent the review unit back - the FS35 is still here and represents the i815E side of things:

StreamD2

I'm not going to pretend I understand what each type of memory function is, but I can read a comparative graph as well as the next guy - and I'm not seeing a big performance boost from DDR here, either. In fact, the solitary i815E board seems the most impressive. Time for something a little more real-world - MadOnion's 3DMark 2000 and 2001 3D-eyecandy benchmarks:

3DMark2000

3DMark2001

This time the CUV266 pulls slightly ahead. Maybe with a faster CPU (we're only using a P3-667EB remember) or higher resolution the difference would be more dramatic. Given that the AK37Plus is only running at CAS2.5 for reasons mentioned earlier, we can probably presume that, were it capable of running at CAS2 like the CUV266, it would score similarly to it. As it stands though, both the i815E SDR boards beat it.

I could throw pretty graphs at you all day but I think the point is clear - we're not seeing a spectacular increase in overall performance, or even memory performance, by moving from SDR to DDR. Granted, looking at the VIA chipsets in isolation, the Pro266 is clearly beating the Pro133A. However, the performance of the DDR boards doesn't come close to justifying the extra AUD$150 they cost.

One final section, of course..

NEXT PAGE - Overclocking and Conclusions