It has been close to two weeks since I last reported on tecChannel's discovery of the VIA PCI bug. Since then, tecChannel has re-released the article in English (versus the e-zine's native language of German). Now that I've read the article in full, I realize we only managed to scratch the surface of the subject. tecChannel was unbelievably thorough in their testing, so much so that VIA's confirmation of the issue was nothing more than a formality; tecChannel had already produced concrete evidence.
While I give 110% to summarize a story for you all, the scope of tecChannel article is too big for me to hit all the relevant points. Let me urge you to read the article in full for yourself. That said, let's get to the facts.
tecChannel's testing consisted of nine motherboards, six hard drives, four different PCI IDE controllers, and one PCI SCSI controller. On top of that, they employed the assitance of a logic analyzer and PCI probe to check not only the operating paramaters of the PCI cards used, but the PCI bus itself.
The results of their testing showed Intel's core logic chipsets at the top of the heap. Chipsets from ALi and SiS were within the expected range of real-world performance, as well. The VIA chipsets, meanwhile, were significantly slower than the rest. But why?
The logic analyzer and PCI probe shed some light on this mystery. Read this:
Less than one kilobyte of data output versus four kilobytes of data from the Intel chipset. Even with George Breese's patch, VIA-based motherboards only manage transfer 32 packets of data concurrently. This slight increase in data output led to another revelation, though. Read this:
VIAs PCI bus: This bus of a P4X266A system transfers only 24 packets of data, after the initial address was sent. After these 24 packets are transferred, a new address gets requested. The PCI bus can only transfer blocks of 96 bytes of data within one burst with this solution.
Intels PCI bus: After the target address is sent, 1024 blocks of data are transferred. Only then the next address gets requested. This makes Intels PCI bus more efficient than VIAs. With Intels chipsets being able to transfer 1024 packets within one burst, this makes a whole transfer of 4096 bytes. As a consequence, Intels chipsets gain a higher transfer rate overall on their PCI bus than VIAs chipset.
Playing around with the patch a little more we found that even audio applications can gain a great benefit from it. One system with a professional DSP sound card for studio use lost all the drops of notes it showed before. Without the patch, one single note played by a software sampler was audibly interrupted frequently. This effect disappeared after applying the patch. It's another indication, that PCI transfers got interrupted to often.Perhaps some were too quick to blame Creative for the SBLive! problems a while back. Although the SBLive! demands quite a bit from the PCI bus, it's hardly Creative's fault that VIA can't seem to get data rate of its PCI bus up to par.
The problems found, especially with the frequent interruptions of data transfers on the PCI bus of VIA's chipsets, may also be responsible for many of the compatibility issues found with their products in the past.
As a consequence, we can currently not recommend VIA chipsets for professional users who demand high performance from their hard drives and think about setting up RAID configurations. This includes video editing, small- and medium-business servers and workstations for graphics- and audio-editing.Given their test results, it's hard to argue.
|AMD drops prices on the Radeon RX 460 and RX 470||15|
|Reports: Radeon RX 470D is a budget Polaris card for China||5|
|Examining reports of slow write speeds on the 32GB iPhone 7||19|
|Cellular Insights dissects iPhone 7 Plus modem performance||10|
|Deals of the week: scads of high-performance storage and more||8|
|Tobii's Eye Tracker 4C knows where your head is||0|
|GeForce driver 375.57 is prepared for Titanfall 2||6|
|Phanteks Eclipse P400 gets a tempered glass option||0|
|Radeon 16.10.2 drivers add support for October's big games||10|
|A real "console monitor" would be 720p @ 30 Hz ;P||+58|