nForce PCI writes mystery: Solved!


— 8:57 AM on June 11, 2002

Last week, we revealed the possibility that NVIDIA's nForce chipset may have a problem with PCI writes. We weren't sure at the time what exactly was the cause, so we turned to you for results. We also checked with MSI and NVIDIA to get their input on the matter. Now we have some answers.

Your testing results provided a clear pattern. Regardless of processor, OS, adapter, or hard drive differences, the MSI K7N420 Pro would not write data any faster than 20MB/s. Meanwhile, the Asus A7N266 series posted numbers down in the 3MB/s range. However, A7N results also proved that the newest beta BIOS resolved these PCI writes problems. Abit and Leadtek results stayed elusive until the last minute, but you guys came through with scores in the 61MB/s and 58MB/s range, showing that these boards are unaffected by the PCI writes problem. With those results in hand, it was beginning to look more and more like a BIOS issue—that the culprit was more Darth MSI than Darth NVIDIA.

Then we received a call from Bryan Del Rizzo, the Integrated PR Manager at NVIDIA who oversees the nForce platform. He confirmed that a fix should be available in the form of a BIOS update. Once the BIOS update has been applied, the "MRM prefetch" option must be enabled. (Depending upon the whims of the motherboard manufacturers, this setting could be obfuscated from the user and set to "enable" by default.) NVIDIA referred us to the manufacturers to confirm who did and did not have a BIOS ready.

We know Asus has their BIOS ready here. We only had two Abit results, one score of 34,779MB/s and another at 61,512MB/s. One reader, using different test parameters, presented four sets of Leadtek results. He tested both Leadtek's shipping and current BIOS, and I'm happy to say both revisions scored the same: 58MB/s.

All that leaves is MSI owners without a BIOS update and fix. Of the thirty-one sets of results you guys submitted, seventeen of them are from MSI users. Since the MSI nForce board saw the largest market penetration, MSI's unresponsiveness is especially disappointing. MSI hasn't responded to our questions about this problem, so we have no timetable for a BIOS fix from MSI.

So what was the problem with the nForce? NVIDIA didn't give us exact details. My theory is that we are seeing a prefetch issue, much like the problem with the Intel 850 and 860 chipsets. If the nForce and 850 issues are similar, then the BIOS fix will only provide PCI write bandwidth up to a certain point. The 850 and 860 stopped right at 90MB/s on PCI writes. Once MSI has released a BIOS with the fix, it will be easier to determine whether or not that is the case. TR readers like Kraquen, with his RAID-0 X15 Cheetah setup, should be able to saturate the bus.

Of course, this last bit of speculation is just that. In reality, the problem could be far more benign. It could have been a typo in the data sheets provided to manufacturers or a human error in the design of the reference BIOS. We'll probably never know for sure, and in the end, all most people want is the fix. A big thanks to the community as a whole for spreading the word and providing us the results we needed.

   
Register
Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.