PCI-E Gen 2.0 Bug X38/ X48/X58

[JohnZS]

Sadly this patch does not work on the ASUS P5E3 which has a 1MBit BIOS ROM compared to other ASUS X38/X48 boards which have a 2MBit BIOS ROM.
I tried the patch posted, but I keep getting stuck in an endless loop of "Windows Startup Repair".
This is Windows 7 x64.
John

[death]

hmm should work since it dont touch u bios all it does its place files on u pc that at the start tell windows u running an x58 but hmm

[nusihb]

I have a 'problem' with SLI 480GTX on a X58 motherboard - P6X58D Premium.
GPU-Z shows one of the 2 working in 'full' PCIE 2.0 X16, but the other stays in PCIE 2.0 X8.
HWINFO32 shows v2.0 X8 for both, in 'idle mode' 2.5Gb/s and during (windowed) Furmark still X8 but at 5.0Gb/s.

Would be interesting to know if they really are running in v2.0 X16...

And yes, I have the MB BIOS set to 'x16-x16-x1' for the 3 PCIE x16 slots.

[luke997]

I have a 'problem' with SLI 480GTX on a X58 motherboard - P6X58D Premium.
GPU-Z shows one of the 2 working in 'full' PCIE 2.0 X16, but the other stays in PCIE 2.0 X8.
HWINFO32 shows v2.0 X8 for both, in 'idle mode' 2.5Gb/s and during (windowed) Furmark still X8 but at 5.0Gb/s.

Would be interesting to know if they really are running in v2.0 X16...

And yes, I have the MB BIOS set to 'x16-x16-x1' for the 3 PCIE x16 slots.

Check the details of the lanes for cards in Everest.
It'll tell you exactly where the stuff is connected, perhaps GPU-Z/HWInfo may have issues when cards downclock/upclock on your system.

[nusihb]

Hm, where in Everest (v550_2100) do I find info about current PCIE mode?
Can't find it anywhere...

[luke997]

Hm, where in Everest (v550_2100) do I find info about current PCIE mode?
Can't find it anywhere...

Motherboard --> Chipset --> PCI Express Controllers

You'll have a list here of all the available lanes, devices on it and modes the devices are in.

[nusihb]

Thanks!
Blind as usual... anyways, Everest is showing the following for the PCIE 2.0 bus:
#1 x1 Marvell SATA 6Gb/s controller
#2 x1 NEC SUB 3.0 controller
#3 x8 GTX480
#7 x16 GTX480

Seems like GTX480 in the first slot (closest to CPU) is 'stuck' in x8??
Bad card or bad m/b?

Will switch the 480s later this week to see if things changes.

[demo23019]

Thanks!
Blind as usual... anyways, Everest is showing the following for the PCIE 2.0 bus:
#1 x1 Marvell SATA 6Gb/s controller
#2 x1 NEC SUB 3.0 controller
#3 x8 GTX480
#7 x16 GTX480

Seems like GTX480 in the first slot (closest to CPU) is 'stuck' in x8??
Bad card or bad m/b?

Will switch the 480s later this week to see if things changes.

I have the same set up and mine are fine
Try loading default settings in the bios.

[MoreGooder]

I just wanted the OP to know that this thread saved me at least $400. I was going to buy a new video card for my rig to stop the slide show when I play BFBC2. Now, I play with 4XAA and 4X AF at 1680 x 1050 on my water cooled and moderately overclocked GTX 260. The patchworks perfectly. (Win 7 64bit)

Props to you, and all that contributed to this excellent thread!

[JohnZS]

No problem at all MoreGooder
Thank David/Death for finding that the patch resolved the issue.
However there is still more the motherboard manufacturers and nVidia need to do for us. To cut a long story short ASUS have "closed the case" and will not resolve this bug.
I have requested nVidia add a feature to the forceware drives to add PCI-E Gen 2.0 support to be forced by the control panel as an option. (for Gen 2.0 cards on Gen 2.0 boards), I have also asked them to fix this bug.
In the meantime I am going to send another e-mail to ASUS asking why my GTX 295 runs slow compared to a GTX 295 on a Rampage Formula (X48). They are BOTH running at PCI-E Gen 2.0, however the transfer speed is crippled on my P5E3 to 4600MB/s yet on the Rampage Forumla it is happy @ 5300MB/s (this can cause a very minor performance issue with stuttering in games which are bandwidth sensitive and also reduce CUDA performance).
John

[Ket]

GTX460 has similar PCIE issues, the customised Quadro drivers I'm now using stopped that though. Didn't solve my biggest problem however, on my M2F for absolutely no reason on bootup / restarts my monitor signal just goes. Theres nothing wrong with the GTX460 itself I verified that by putting it in a M3N78.

[JohnZS]

Well the good news (and I hope it is good news) Ket, is that the nVidia driver team are now looking at ways of either resolving this issue, or getting the wheels put into motion to fix this issue.
I personally submitted a bug report AND an nVidia technical representative has posted into my thread on the nVidia forums saying that he has informed the driver team of this issue.
Fingers crossed and all that
John

[Ket]

I don't think its a driver problem or the same issue would happen with the M3N78, I do think its a vBIOS problem though

[JohnZS]

It is sort of a combination of driver Bug and chipset bug. The last official wording from nVidia back in June was that it could not be fixed.

Hi John,
I am sorry however there is nothing we can do in this case. The Intel x38 chipset was Intel's first generation PCI-E 2.0 implementation and unfortunately has some PCI-E 2.0 bugs, one that causes the graphics card to hang when changing P states. The only Intel PCI-E 2.0 chipset which shows these bugs is the Intel x38 chipset. If you see any other Intel chipset reporting PCI-E 1.0 speeds with our GPUs, please let me know. The best way to check is to run a 3D program in the background as this will change the P state of the graphics card from PCI-E 1.0 to PCI-E 2.0. This throttling is done to conserve power when the user is running basic Windows tasks.

But after running ASUS into the ground with queries asking them to fix this "X38 Bug", they keep pointing me back towards nVidia. nVidia are now going to look into this again....

[STEvil]

All an X38 board manufacturer has to do to fix the issue is modify a bios to hold the P-state at L0.

All nVidia has to do is modify their driver to never lower the P-state for the PCI-E link.

ACPI tables in the bios of the motherboard are what to change to "fix" the P-state.

[Ket]

A lot of manufacturers of GTX460s have removed a P-State in their vBIOS. Galaxy haven't, but Gigabyte have, as have Gainward, and Palit. A possible workaround for the problem is to set GPU voltage in the vBIOS @ 1v.

[JohnZS]

Thanks for the information STEvil and Ket.
I have passed on your suggestion to ASUS STEvil. Hopefully nVidia will resolve this by the driver as it makes more sense to check for X38 and then never lower the P State as opposed to check for X38 and then revert to legacy PCI-E Gen 1,0 mode.
Am I right in assuming Single PCB GTX 295's do not have support for P-State in their vBIOS? or is it just some quirk that the only nVidia cards running happily @ Gen 2.0 on my P5E3 is a GTX 295. 280 and 285 cards only run @ Gen 1.

By the way, Gen 1 does have a detrimental effect on the GTX 295's performance. (On occasional cold boots it reports Gen 2.0 but actually transfers @ Gen 1 speeds). FYI when a GTX 295 is running @ Gen 1 I lose 1000 Vantage points, 1000 3dmark 06 points, 0.2 WEI on the Graphics card.... and GTA IV is horrifically stuttering and laggy.

I have also spoken with someone called Chris who is heavily into CUDA and he says that Gen 1.0 cripples CUDA Performance on his GTX 480 (even in Linux as nVidia have put the lock on those drivers too).
So this IS an issue which needs resolving
John

[Ket]

I should clarify what I said in my last post a bit. When I say set GPU voltage to 1v I mean for bootup / 2D clocks, 3D voltage is fine as is. As I said though, this is only a possible workaround. The problem is primarily with compatibility issues with GTX460s on a fair few boards. Both mobo manufacturers and nVidia should issue a new BIOS resolving the bootup issue. The driver for GTX460s doesn't really have anything to do with these bootup issues as the driver hasn't even had chance to initiate at that stage

[rangerone7669]

i ran the sli patch and i''m back to pci 2.0 now thanks alot. gained about 500 points in 3d06

[Ket]

I'm still testing my fix for the inexplicable signal loss on reboots with some systems and GTX460 cards so I won't go into too much detail but I believe I've nailed the problem. After some more testing I'll post further details.

[STEvil]

I wonder how much this ACPI bug (thats what it is mostly) is responsible for stuttering in SLI and Crossfire?

[Ket]

If its a ACPI bug some boards you can disabled ACPI, anybody tried that?

[JohnZS]

Thanks Ket

Which one of these options would you suggest for disabling ACPI?


STEvil
You have a very good point there, perhaps this is the cause of Microstutter?
Anyway with a GTX 295 (Single PCB), this issue manifests itself slightly differently. The card always reports as working @ PCI-E Gen 2.0.

However on cold boots and sometimes (on rare occasions) a reboot the card seems to lock itself into a crippled PCI-E transfer mode.
Take this screenshot here


That is ~20% of the transfer to BUS speed that it should be using, when a GTX 295 is in this state games stutter and performance is not very good at all. (WEI also drops by 0.2 on the Graphics card).

But when those transfer values are at 48xx MB/s everything is smooth and fine

Is this a symptom of the ACPI bug?
John

[Ket]

Try disabling each of them one at a time and see what happens, then try both

[JohnZS]

Well, I ended up getting a Blue Screen crash disabling ACPI 2.0 Support.
The annoying thing is everything was going well and very fluid in gaming, and then bham, the game froze (but sound continued), then the monitor went into an out of sync black screen, and then a BSOD popped up.

BCCode: 117
BCP1: FFFFFA800746F4E0
BCP2: FFFFF880101A3CF8
BCP3: 0000000000000000
BCP4: 0000000000000000
OS Version: 6_1_7600
Service Pack: 0_0
Product: 256_1
Bucket ID: X64_0x117_Tdr:2_IMAGE_nvlddmkm.sys
Server information: 3940acc5-e624-406e-b30b-a893a96905bb

Never had that happen before
John