• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

PCI bottleneck in intel 850 chipset?

J

jberke

Junior Member
:frown:

Hi - I'm trying to set up a data acquisition system (for listening to lots of brain cells) and I'm getting distinctly
subpar performance from my Pentium 4/i850 chipset machines, running under Real Time Linux. I need to perform frequent data transfers from a PCI A/D converter card (made by UEI) to memory. But the time taken by the interrupt service routine to complete one such operation seems to be far more in my machines with the 850 chipset than in older machines, and so the system can't handle the required data acquisition rate (producing a FIFO overload error from the A/D board).

the problem is seen under either a Dell Optiplex GX400 P4 1.8GHz or a Micron P4 2.0GHz (with a 66MHz PCI interface),
both with the i850 chipset. In both cases the time taken by the interrupt service routine to transfer one FIFO buffer of data from the PCI A/D board to memory is around 1.4ms. The problem does *not* occur in several older Athlon 1.0GHz
machines (transfer time around 0.6ms).

basic problem *appears* to involve the fact that on the i850 chipset the AGP video card occupies PCI slot 0, so my
PCI A/D data acquisition card ends up in slot 2. we're speculating that it takes the interrupt service routine twice as long to transfer data to memory from PCI 2 as from PCI 0. Is this right, and is there way at all around this problem?

unfortunately/stupidly we went out and bought 9 Pentium 4/i850 computers (7 Dell, 2 Microns) and
a similar number of $4k PCI A/D boards before becoming aware of this problem...
Particularly annoying as I thought this "advanced" chipset would be *less* likely to have such issues...

any advice greatly appreciated...
please forgive any displays of massive ignorance implicit in this post!

Josh
 
Do you have the latest BIOS and chipset drivers for the systems? Also try tweaking the PCI options in the BIOS.
 


<< Yes, there are documented PCI latency and timing issues with Intel's i850 chipset. >>

I heard about it haveing a documented BANDWIDTH issue. ie, 33Mhz PCI was limited to ~120MBps
Another thing:
Since the very beginning, Intel has said to never use dual video cards with the 850 (HUGE performance issue). Does there happen to be integrated video in addition to the AGP card?
 


<< Since the very beginning, Intel has said to never use dual video cards with the 850 (HUGE performance issue). Does there happen to be integrated video in addition to the AGP card? >>



Could you please provide a link? I'm currenty running a GeForce Ti200 AGP and a V3 2000 PCI. 🙁
 
ouch. That has got to hurt. I dont think there is any option for you to tweak anything on your dell systems has far has timing goes.
 


<<

<< Since the very beginning, Intel has said to never use dual video cards with the 850 (HUGE performance issue). Does there happen to be integrated video in addition to the AGP card? >>



Could you please provide a link? I'm currenty running a GeForce Ti200 AGP and a V3 2000 PCI. 🙁 >>


I agree........I ran a dual head Matrox G450 on my old S423 TH7, and run dual cards on one of my new TH7II's and never had any problems with either.........😉
 
The dual-display problem was publicized before the P4 was even released, therefore banning any possible confusion w/ the i845. The PCI being limited to 120MB thing was fairly recent. I'll try to dig it back up again (I think the thread was something like "Intel's got chipset issues too!" with some mention of Via in there somewhere.
 
Maybe that has something to do with bug in i850/860 chipsets that limits PCI bandwidth to ~80Mb/s in certain (supposedly rare) situations ?
 
not sure what BIOS PCI options would be relevant... any point in trying to reserve a high priority IRQ #?
if anyone knows of a relevant chipset driver upgrade, please let me know.
the 4x AGP card is the 32MB nVIDIA card that came with the machine. also tried swapping to
a 32MB Radeon, didn't make any difference.
 
The only "problem" the i850 has is this:



<<
5. Sustained PCI Bandwidth

Problem: During a memory read multiple operation, a PCI master will read more than one complete cache line from memory. In this situation, the MCH pre-fetches information from memory in order to provide optimal performance. However, the MCH cannot provide information to the PCI master fast enough. Therefore, the Intel® 82801BA terminates the read cycle early to free up the PCI bus for other PCI masters to claim.

Implication: The early termination limits the maximum bandwidth to ~90 MB/s.

Workaround: None

Status: There are no plans to fix this erratum. >>



This information can be found on page 11 of the Intel spec sheet fot the i850 chipset(link here). This problem only effects PCI cards reading from memory not writing to memory so the i850 isn't that cause of Jberke's problems.

I think its the conservative timings and bios' on the motherboards OEM's like Dell and Micron use thats causing your problems and sad to say their not much you can do to fix it. OEM motherboards don't have the all tweaking features in their bios' that boards from Asus, Abit or MSI have so you can't tweak those boards to suite your needs.

If Dell and Micron can't fix your problem I say return the machines and build your own using Asus or MSI motherboards.
 
In the 8 months I've owned a P4 rig with I850 chipset I've never come close to saturating the PCI bandwidth. I'm not sure what WOULD eat up that much bandwidth. Hard drive transfers and AGP data flow thru a diff hub. Even serious video editing shouldn't come close to maxing it out.
 


<< I'm using an i815EP-B chipset right now, and it's bandwidth with its ICH2 (82801BA) southbridge is only 53-max 60 MB/s. VIA 686B has a bad reputation, but man is it faster! >>



The i815 has no problem going over 115MB/s in burst mode with ATA133 hard drives and PCI controllers.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top