Introduction
In our last test we could show that VIA’s Apollo Pro 133A chipset has become a serious competitor to Intel’s i820. Especially the lower costs of the platform plus the much lower cost of the actual memory (PC133 SDRAM vs. PC800 RDRAM) speaks quite strongly for VIA’s latest solution. Still there are several issues with this chipset that I’d like to discuss here.
A — The AGP-Driver of Apollo Pro 133A
You might have seen it elsewhere or heard about it, some motherboards with the Apollo Pro 133A seem to have serious performance problems in 3D-applications. I saw references claiming that particularly the Asus P3V4X performs very badly in e.g. Quake 3 Arena. The results published by me in my article ‘Showdown at 133 MHz FSB‘ did not reflect this problem and thus I received a few questions.
When I did my first runs of the P3V4X I had used VIA’s latest set of drivers, the 4in1-drivers rev. 4.20. Those drivers include the AGP-driver (viagart.vxd), some inf-files and VIA’s latest bus master driver for ATA (IDE) hard drives. The 3D-performance with this set of drivers was very low indeed and for me much too low to be true. Therefore I spend quite a while trying to find out what was hindering the 3D-performance of the P3V4X, because it was obvious to me, that I could not possibly publish those — obviously wrong — results. After a while of trying I found the reason for the bad performance. It is VIA’s AGP-driver. I had already been puzzled why Asus delivered the motherboard with VIA’s 4in1-drivers rev. 4.17, since those drivers seemed very out of date. Installing those drivers instead of rev. 4.20 solved the mystery. The P3V4X scored perfectly well with the AGP-driver of the 4in1 rev. 4.17. The driver sets 4.18 — 4.20 lead to horrible AGP-performance though and are to be avoided. Thus I recommend using rev. 4.17 for the time being, while we are trying to find the reason behind this strange behavior.
That’s all there is to it. The wrong AGP-driver leads to horrible 3D-performance.
B — Ultra DMA 66 and the Apollo Pro 133A
VIA is claiming support for ATA66 or UDMA66 since the release of the Apollo Pro Plus chipset and of course the Apollo Pro 133A is supposed to support this standard as well. This transfer standard for IDE hard drives offers up to 66 MB/s bandwidth via DMA (Direct Memory Access). It’s also called ‘Multi word DMA Mode 4’.
Unfortunately our tests revealed that this support exists only on paper. Measuring the maximum throughput from the hard disk cache to the system produced results that were only in the area of ATA33 or UDMA33, which offers a bandwidth of only up to 33 MB/s. There’s currently no real world benchmark that would make any difference, mainly because even the fastest disks can only supply up to 30 MB/s from the physical media, but this might change very soon. Therefore I expect VIA to sort out this issue as soon as possible. We do not know if the problem is due to a driver issue, or if the actual chipset is unable to transport data fast enough, but we do know that motherboards do report ‘Multi word DMA Mode 4’ with the proper hard drives, alas without offering the expected throughput.
Intel’s i810, i810E, 820 and i840 chipsets claim UDMA66-support as well and they are actually able to reach the expected throughput. Alternatively we tested with the Promise Ultra66 PCI card, which is usually used for systems that don’t offer ATA66-support. This card produced the same results as the motherboards with Intel 8XX chipsets.
For benchmarking the maximum throughput I decided to use a nice and super old benchmark that most of you won’t know anymore. ‘Coretest’ used to be a widely used hard drive performance benchmark some 10+ years ago. It’s reading 64 kB blocks from the hard drive and does some seek time tests as well. Today it can only test how fast data can be retrieved out of the hard disk cache memory, but this is exactly what we want.
It’s obvious that VIA’s Apollo Pro 133A is not able to transfer data at ATA66 standard.
C — VIA Apollo Pro 133A with CL3 PC133 SDRAM
It’s not that easy to get the correct PC133 SDRAM nowadays. Going in a shop and asking for it can get you any kind of PC133-SDRAM. That’s not what you want though. PC133 SDRAM only makes real sense if its CAS/RAS latency is 2 clock cycles and not 3. In many cases you will find out that the guys in the computer stores haven’t even got the slightest clue what you’re talking about if you ask «is this stick PC133 SDRAM CL2 or CL3?» I advise to decline from a purchase in case that you don’t know if it is CL2 or CL3. The performance difference is rather significant and you might get rather disappointed if you should have bought the wrong memory.
SysMark2000 does not make that much of a difference if it’s CL2 or CL3.
Most 3D games take advantage of CL2, except of Unreal Tournament.
C — VIA Apollo Pro 133A with CL3 PC133 SDRAM, Continued
AGP-performance is heavily depending on memory performance. CL3 has a bit of an impact even on this benchmark.
The professional OpenGL-benchmark Viewperf is very touchy about memory bandwidth. Using CL3 instead of CL2 results in 5-8% performance loss.
Follow-up by reading the article ‘Showdown at 133 MHz FSB — Part2, The Real McCoy‘.
