PCI-E Gen 2.0 Bug X38/ X48/X58

[Ket]

Looks like you would need to do a clean install with those BIOS settings and try then.

[STEvil]

doubt a clean install will fix it. Non-ACPI HAL's are slower anyways. Remember the "Ricky Tweak" for single core CPU's on WinXP?

[Ket]

When you enable/disable some ACPI settings in the BIOS they are directly linked to some OS things and a format is the only way to set things up "correctly". Eitherway, these GTX issues are chipset specific. They can be fixed with a BIOS update but through lack of a better term most BIOS writers are dumb. They don't know how to fix problems that occur outside of normal chipset paramaters, even if their easy things to fix.

[JohnZS]

I am still awaiting a reply from ASUS, but in my previous correspondence from them in July they did say they spent a month working on the issue and could find no way to resolve it

Hi John,



Our BIOS RD and BIOS RD leader have tried many ways to solve this problem by BIOS for more than one month

But that was on the 26/7. Hopefully your and StEvil's insight will sway their way of thinking.
I have also reported the issue to nVidia too so hopefully (fingers crossed) this will be resolved for everyone

John

[Ket]

Well, to solve the bootup issues some GTX460s have you can tell Asus all they have to do is get the BIOS to force the graphics cards P8 state at boot instead of the P12 state

[JohnZS]

Thanks Ket
What constitutes a bootup issue? The card not being found on POST so leaving you with a lovely beep code? Or system booting with crippled performance (like with the GTX 295 on rare cold boots).
John

[JohnZS]

ASUS are not helping at all, in fact I am still waiting a response from the 8-8-2010
John

[JohnZS]

nVidia have responded...

There is a possibility that our software team can create a backdoor method where the graphics card won't change P states and then can always run at full speed but the downside of this is the graphics card will consume more power. Is this an ok alternative for everyone? I am still not 100% certain they will go ahead with this yet.

What would your opinions be guys?

John

[STEvil]

My opinion is use software to control clock speeds, but leave p.state at full perf.

[Maxforces]

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

problem is only in idle mode and you don`t loose performance


when you run 3D mode PCI-E returns to PCI-E 2.0


same on 9800GT,GTX480/470 on X58

[STEvil]

problem is only in idle mode and you don`t loose performance
-snip-
when you run 3D mode PCI-E returns to PCI-E 2.0
-snip-

same on 9800GT,GTX480/470 on X58

try reading the rest of the thread.

[JohnZS]

Thanks STEvil

By design Maxforces nVidia cards with variable pStates are supposed to run @ PCI-E Gen 1.x when the card is idle and then switch into PCI-E Gen 2.0 mode when any load is put onto the card, sadly on the X38 chipset there is some sort of "bug" which causes alleged stability issues when a card switches from Gen 1.x to Gen 2.0 mode. So nVidia chose to lock the cards @ Gen 1.x mode regardless of how much load was put onto a card.
If you are not convinced I can post a picture of a GTX 480 running Furmark for more than 5mins yet still locked @ PCI-E 1.x (this was kindly posted by someone who suffers from this bug on a ASUS P5E3 WS Pro (X38) motherboard).
John

[Maxforces]

Thanks STEvil


If you are not convinced I can post a picture of a GTX 480 running Furmark for more than 5mins yet still locked @ PCI-E 1.x (this was kindly posted by someone who suffers from this bug on a ASUS P5E3 WS Pro (X38) motherboard).
John

I am not convinced I chacked 10 games and every time PCI-E changed to 2.0 even with a small load.
This is one of the smallest loads 30% of GPU use

pci-e change in P3-games do not use this mode

PCI-E change to 1.0 in P -state 12 idle mode


you do not loose performance in games, only i p3 mode--when you play movies and for this you do not need PCI-E 2.0

So show me furmark or any game with pci-e 1.0 on x38/x48/x58, I do not see, only idle and p3 state which in no way interferes of any performance .

[JohnZS]

The Bug does not occur on X58 and X48, apparently those chipsets are now fully ratified to work by nVidia (I believe there was a driver version which did lock those chipsets out at one point, or some early X58 BIOSes had an issue). Anyway what motherboard are you using?
Here is the proof for you..

Thanks to MikeS on the nVidia forums for performing this test
John

[Maxforces]

Intel DX58SO and I never had a problem with PCI-E in 3D mode.

[Ket]

Thanks Ket
What constitutes a bootup issue? The card not being found on POST so leaving you with a lovely beep code? Or system booting with crippled performance (like with the GTX 295 on rare cold boots).
John

The problem is the monitor signal being dead for no reason at some bootups and on most restarts. Theres no reason why this should happen, it just does in some systems. Best I can figure out is force the mobo BIOS to only use P8 state on boot / restarts, should fix the issue 100%.

nVidia have responded...



What would your opinions be guys?

John

Instead of nV doing that get them to simplify things. They need to can the P12 state. Modify the P8 state to run GPU/mem clocks of say 200/300, unlink the P3 state and use it for UVD/DVD/Blueray playback, clocks of 400/1000 should be plenty. Finally use P0 as its always been used - for 3D clocks.

[STEvil]

Intel DX58SO and I never had a problem with PCI-E in 3D mode.

I bet that the Intel board has an ACPI table in the bios that conforms to the "standard" for ACPI a lot better than many others tend to.

edit

102c is pretty toasty there JohnZS

[JohnZS]

Thanks Ket
I have relayed this onto nVidia (and credited you for your expertise and knowledge).

Indeed STEvil, I must admit I was concerned when I saw 102C and informed Mike that his GTX 480 was running extremely toasty. For comparisons sake a GTX 295 (Single PCB) in my system runs upto 86-88C after one whole hour of Furmark @ 1920*1200 16XFSAA. I think that is "safe" as during heavy gaming the card never goes over 70C (according to eVGA precision tool).

You are right, I think the DX58BSO had this issue fixed in a BIOS update.

Fixed issue where SMBIOS System Slot structures mismatch for PCIe Express Gen2 slots

Listed as a fix for July 2009?

Also in May 2009

Set Slot Power to 300W for PCI Express x16 slot

But then again I know nothing
John

[Ket]

Oh, so its clear for the nV boys aside from the P12 state causing havok in some systems the clock going as low as 50/135 cause problems in Windows as well sometimes. Depending what you are doing, when the card goes as low as 50/135 a Windows message pops up saying aero performance is really bad do you want to disable it etc. Its generally very annoying. Incredibly, performance even on the BIOS setup menu is terrible as well. Never in my life have I had to put up with such slow scrolling and screen lag.

[donmarkoni]

Oh, so its clear for the nV boys aside from the P12 state causing havok in some systems the clock going as low as 50/135 cause problems in Windows as well sometimes. Depending what you are doing, when the card goes as low as 50/135 a Windows message pops up saying aero performance is really bad do you want to disable it etc. Its generally very annoying. Incredibly, performance even on the BIOS setup menu is terrible as well. Never in my life have I had to put up with such slow scrolling and screen lag.

Please post screenshots of it happening.
Do you really think that NF100/104 based VGA can't process text mode @ 50/135MHz?

The Bug does not occur on X58 and X48, apparently those chipsets are now fully ratified to work by nVidia (I believe there was a driver version which did lock those chipsets out at one point, or some early X58 BIOSes had an issue). Anyway what motherboard are you using?
Here is the proof for you..
http://img840.imageshack.us/i/gtx480fur3k.jpg
Thanks to MikeS on the nVidia forums for performing this test
John

Why the thread still says "PCI-E Gen 2.0 Bug X38/ X48/X58"?

[Ket]

1. Screenshots won't capture the lag
2. In the BIOS setup menu the card isn't running 50/135, if it was, there wouldn't be a problem. The card apparently dynamically downclocks to as low as 13MHz when not in Windows, that certainly would cause lag, even in a BIOS setup menu.

[donmarkoni]

1. Screenshots won't capture the lag
2. In the BIOS setup menu the card isn't running 50/135, if it was, there wouldn't be a problem. The card apparently dynamically downclocks to as low as 13MHz when not in Windows, that certainly would cause lag, even in a BIOS setup menu.

Not lag, balloon tip in Win.
13MHz? There is no performance level with 13MHz core speed.
BTW, my VGA regularly downclocks to 50/135 and runs 2D/flash/movies no problem.

[Ket]

I don't know how to reproduce the win popup so I'll try to remember to get a screenshot with what I'm doing when it next happens. As I said, there doesn't need to be a P-State with those values, they are part of the P12 state where the GPU/memory can adjust themselves it would seem from anywhere between 13/34 and 50/135.

[donmarkoni]

I don't know how to reproduce the win popup so I'll try to remember to get a screenshot with what I'm doing when it next happens. As I said, there doesn't need to be a P-State with those values, they are part of the P12 state where the GPU/memory can adjust themselves it would seem from anywhere between 13/34 and 50/135.

OK, you do that.
You are saying that VGA can set clocks other then preset performance levels? And how do you prove that?
If you prove it, I'll stand corrected, but otherwise it's just a nonsense, as I never seen it happen to anyone else, or had similar experience(s).

[Ket]

Just run nvidia inspector and select the P12 state from the OC side. You will see the apparent clock ranges. You can also use Fermi BIOS Editor to see how some values are actually set to 0MHz. Don't call nonsence on things when you don't have access to all the tools to look at things properly yourself