EDIT: updated this post so people dont have to fight their way through this post jungle
There seem to be several issues with the 790 chipset:
1. windows/hdd corruption/ system instability
2. FSB holes
3. HDD corruption/destruction
4. real voltages are notably lower than what is set in bios (mostly applies to nvidia reference boards)
------------------------------------------------------------------------
1. While on other chipsets running unstable or semi stable settings causes a BSOD, system freeze, restart or power down, on 790 it tends to corrupt the windows install very easily. Nvidia officially admitted that theese issues exist, but claims its a highly isolated issue and only happens when overclocking the chipset to very high speeds.
The problem now is that for some reason many 790 boards are not stable even at stock or below stock settings. this whole issues is even more frustrating as it seems to be very difficult to determine system stability... even when all widespread tools and benchmarks seem to run fine for hours, the system might still freeze/shut off or corrupt after some more hours or days.
It seems that if a configuration passes all stability tests beeing under full load, it might still be unstable in idle or when switching from idle to load.
The curious thing is that, as you can see from the poll, around half the people are not having any problems, while the other half are having a really hard time getting their 790 based systems stable.
There are 3 theories what could cause this:
a- VTT / GTL ref issues
b- Widespread Chipset yields (some chipsets work fine, others dont)
c- Bad BIOS coding
VTT seems to play a major role in this, it seems that tweaking vtt/gtl makes it possible for some people to get their systems 100% stable or at least improve stability notably. According to one mainboard manufacturer the 790 chipset has very small tolerances when it comes to vtt, and hitting the sweet spot to get a system stable is not easy. Depending on what cpu you use, if its a quad or dualcore, and if its made in 65nm or 45nm, the vtt window shifts. this of course makes it very difficult if not impossible to program one "auto" setting in bios which just works with all cpus.
Chipset yields, this could be the cause, especially since the 680 chipset, the predecessor of the 790 suffered from very similar issues and in this case its widely accepted that there was a high variation between the boards. some would simply clock awesome, others wouldnt even be stable at stock, which is exactly what we are seeing here. The 790 chipset has 2 settings which impact system and memory performance greatly, mysteriously referred to as P1 and P2 by nvidia. Nobody knows what exactly theese settings do... they improve system performance, mostly memory performance, but they seem to stress the chipset a lot. some people can run p1 and p2 enabled even at high speeds and still have the system stable, while some can not even enable them at stock speeds without losing stability. this again hints towards a high chipset variation.
Bad BIOS coding is unlikely to be the cause, there have been several new BIOS releases that improved the situation slightly, but the problems are not fixed at all.
After all it seems like its a combination of all 3.
the chipset has a very tight tolerance window for vtt
AND
there is a strong variaty between chipsets, some having a higher tolerance and a wider vtt/ gtl ref window than others.
AND
the issues are too complex to be fixed easily in a new BIOS, new BIOSes might improve this but a perfect fix for all configuration seems impossible.
So what can you do?
Ensure good Chipset cooling
Increase the SPP (northbridge) voltage to 1.35V+
Increase the MCP (southbridge) voltage to 1.55V+
Try to adjust vtt and gtl ref values manually and find the sweet spot
Here are some hints for you:
65nm CPUs usually like ~1.4v vtt and gtl ref settings of .63-.68
45nm CPUs usually like ~1.2v vtt and gtl ref settings of .58-.65
Here are some example configurations that work for the people that posted them. Please note you might not be able to copy them 1:1 but will have to tweak them a bit.
65nm Quad example from 5010, again thanks a lot for sharing your results man
another 65nm Quad example from tekjunkie, again thanks a lot for sharing your resultsQ6600 G0 @ 3.55GHz
444 /1777FSB x 8
Linked and Synced
P1 + P2 Enabled
Timings: 7-7-7-21-1T (All other sub-timings auto)
CPU Vcore: 1.50 (Bios)
FSB VTT: 1.35
Memory: 1.95V
SPP: 1.40V
MCP: 1.50V
GTLVREF +35mv all lanes (works at +%2.6 - +%2.7 no matter the VTT all the way down to 1.20VTT and even upto 1900FSB)
Virtualization: enable
All other CPU features: disabled
QDR/RAM=1780 Unlinked
PCIE1/PCIE2 = 100
SPP<-> MCP= Auto
vCORE=1.375(1.344 in windows per cpuz)
Loadline =Enabled
vTT = 1.28
PLL=Auto
vDIMM=1.9
vSPP =1.46
vMCP = 1.55
CPU GTL_REF0 = -29mV
CPU GTL_REF1=-32mV
------------------------------------------------------------------------
2. FSB Holes do exist for some people, while they dont exist for others...
they are not always at the same speeds... they seem to be common at around 400fsb though. basically the system will not boot or be highly unstable at a certain fsb range, and above and below this range it works just fine.
this seems to be caused by bad/too agressive BIOS tweaking from Nvidia/the manufacturers. The latest BIOS files seem to have improved this a lot...
try to set as many timings and settings to auto if you want to get rid of an fsb hole... but generally speaking its not worth the hassle, just try to find the spot where the fsb hole ends and run a higher fsb that works fine.
------------------------------------------------------------------------
3. this seems to be caused by memory related/chipset instability issues... im not sure though, and nobody really seems to be, if this is only based on the SPP, the 790 northdridge, or if it could also be caused by the southbridge.
Nvidia is recycling an older chipset part and using it as the southbridge on 790 boards. this older part is known to have caused a LOT of hdd corruption problems, so its likely that this is still the case. especially since there have been numerous reports of HDDs that were actually killed.
So far it seems around 8-10 hdds were killed... nobody really knows how this is even possible as a bad signal or corrupted data should not render a hdd unusable... but it did happen to several people...
Im keeping this as a seperate point since there might be two causes for the hdd and data corruption. the spp instability issues and the sb data corruption issues. There are a few people who used a raid controller card in a pciE slot connected to the northbridge, so its absolutely impossible the SB alone causes all the data corruption issues.
How can this be fixed? i dont know... Can this be fixed? i dont think anybody knows... the cases in which hdds died are rather insulated... only around 8 people reported this... If you want to be 100% sure it doesnt happen to you, id recommend you to use a raid controller card.
------------------------------------------------------------------------
4. there is no fix to this that im aware of, though it might be possible to reprogram the bios and just rename the voltage options to what they really are.
If you have anything to add or want to comment please post here in as much detail as you can. In any case, please keep in mind that its important to send as many details about this to nvidia and their partners so they can work on this and fix this issue. and more importantly, that they are aware it IS a serious issue and notice how widespread it is. because atm nvidia seems to be playing it down and not taking very serious.
Reply With Quote
Bookmarks