AGP bus latency.

[Evil_Donnyboy]

Mine defaults to 64clocks, according to powerstrip. Has anyone tried lower latency values? 32 clocks? I'm running on a very slightly oc'd AGP4X bus, 68mhz. Safe to try for 32?

------------------
P3 650@897 | Matrox G400 32mb SH @ 175/233 | 128mb PC133 | SB Live | winME

[FrankDC]

AGP Latency is the number of clocks an AGP device can hold the bus for, before returning control to the CPU. Higher numbers mean better performance, at the cost of compatibility with other devices in the system (esp. ISA cards).

[Evil_Donnyboy]

ahh. I've alwaz thought lower waz better. Well lower PCI latency causes stability problems anywaz.

So why don't they set the AGP latency to >128? It's got the bus all to itself.

[FrankDC]

No it doesn't. There are other devices on the PCI bus that need to be serviced (PCI<>PCI bridge, PCI<>ISA bridge etc).

[FrankDC]

What model# is his Pavilion?

[Gurm]

I have always set PCI latency to 32. Is there a separate setting for PCI and AGP latency? If not, set it to 32 in order to avoid troubles with other devices, especially ISA.

- Gurm

[Evil_Donnyboy]

The AGP bus is serparate from the PCI bus. Thats the reason why AGP was created in the first place. PCI video cards had to fight for the little bandwidth PCI offered. The original AGP not only offered vid cards their own bus to the cpu/memory, it doubled bandwidth with a 64bit bus. Tha later came AGP2X and 4X.

But my point is, AGP is indeed separate from PCI. They are similar in archetecture, but separate for sure. I don't have to point it out to you in the intel white papers do I?

[This message has been edited by Evil_Donnyboy (edited 20 March 2001).]

[Wombat]

Sit down and shut up. Don't be so rude, there are plenty of people on the forums that could smack you from one side of the room to the other with their architecture knowledge.

Anyway, nobody is arguing that AGP and PCI function separately, but they are fighting to master the same memory bus. It's all one bus using GTL+ (but not using EV6, that's a point-to-point). It's entirely possible and not improbably that the AGP and PCI bus share the same latency counter.

I usually think Gurm comes down too hard on people, and can't always admit when he's wrong, but I want to watch this one.

School is now in session.

[SCompRacer]

<font face="Verdana, Arial, Helvetica" size="2">Originally posted by Gurm:
I have always set PCI latency to 32. Is there a separate setting for PCI and AGP latency? If not, set it to 32 in order to avoid troubles with other devices, especially ISA.

- Gurm</font>

Doesn't this affect the serial ports too?

[FrankDC]

Yes. Donnyboy you're forgetting that a single CPU & memory bus services both PCI and AGP, as well as USB and others. Any device can have problems if it requires CPU attention and cannot get it because another device is not returning control to the CPU.

I don't know if you remember the "automatic PCI bus retries" on older Matrox cards, it caused problems with everything from ISA soundcards to USB cameras.

[Wombat]

Ow, nasty memories Frank. I remember how rapid scrolling with my G200 would cause static on my AWE64.

And I don't mean to be AR, but when you're talking about bus-mastering, you can often consider the CPU to be just another device, and they're all competing for the north bridge's attention.

[FrankDC]

There's a new bios for that system, dated 3/9/01 on HP's site. You might want to try it and see if it makes any difference. Click on 'drivers & downloads' here:
http://www.hp.com/cposupport/prodhom...lion24400.html