Granite Bay performance: Should we be concerned?

[Elvis2]

What's this all about?

 

[wicktron]

This problem is in fact true.

 

[mechBgon]

The i850e and i845 chipsets had/have a bug which limits the PCI bus to an effective 90Mb/sec, but I haven't really heard people having a real-world problem with it. A SCSI RAID card and two fast hard drives would make it very apparent that there was an extra bottleneck somewhere, but evidently not too many people have that type of setup. Then again, the AGP 3.0 spec apparently allows for dual AGP cards (wow) and maybe this will throw a curve into mobo manufacturers' plans for some kind of wild dual-AGP workstation board. Sounds like GB's successor is already in the works, so they might wait.

 

[Thor86]

Ed from Overclockers.com has a good outlook/read about the GB chipsets and the uderlying issues on their homepage.

 

[ManojM]

Ok now I understand why I am having a problem with my Radeon 9700 and Asus P4G8X (Sample).

I have been having lots of lock ups in games and also my display all of a sudden gets all messed up.

I disabled AGP Fast writes by using the Smartgart utility and it sort of solved the problem.

Hmmmm!!!

 

[naddicott]

Then there's this tidbit at 2CPU.com forums (on page2)

Just got a response. The first trace errata was not seen on any production board, only an Intel tester.

The prefetch is disabled but no performance difference have been seen. Intel's benchmarks were also done with prefetch disabled.

I'm still concerned. Was planning on putting together a GB system for a friend, but I don't want to put together a lemon.

 

[Alphi]

I would like to know what this means in terms of video/system performance as well. Intel seems to think that it will not effect performance, but I have to assume that the AGP prefetch cache does something! If you disable it, what does it NOT do? Does anyone know how to translate this? It is a bit beyond my scope.

1. tVAL(min) Timing Is Not Meeting AGP 3.0 Specification
Problem: During simulation studies, it was determined that the AGP signals were not meeting the tVAL(min) of
1ns.
Implication: AGP interface is susceptible to latching invalid data.
Workaround: Minimum trace lengths for the AGP common clock and source synchronous signals have increased to
offset this timing violation. Refer to latest platform design guide.
Status: No fix planned for this erratum. Issue resolved by board workaround.
2. AGP Prefetch Cache Must Be Disabled
Problem: System hangs can occur during AGP transactions when the AGP Prefetcher is enabled.
Implication: System hang.
Workaround: Disable AGP Prefetch Cache by clearing bit 0 of AGP Miscellaneous Register (Device 1 Function 0
Offset A0-A3h).
Status: No fix planned for this erratum. System performance is not affected by this erratum.