New info on 7900 series faults!!! - DEFECTIVE voltage regulators!

NoxTek

NoxTek

The Geek Redneck
Joined
May 27, 2002
Messages
9,300
[EDIT: I'd like to AGAIN remind everyone that this COULD be a load of complete BS. Anti nVidia propoganda being smartly written by someone with lots of time on their hands and the urge to "stir shit up". Take ALL of this with a grain of salt. Mmmmkay? Need I re-iterate this point any further?]

This was just posted on the eVGA forums:

This IS a cross-post from the general forum. Mods, if you dont delete it, feel free to leave it here or move it to the General Forum (where I posted it first):

To all,

There are a batch of 7900 cards with bad voltage regulators on them. These cards were shipped from China. Nvidia IS aware of the bad batch of voltage regulators and is waiting for the batch to run out before updating the regulators with a batch of higher quality regulators. They are ignoring higher than normal failure rate (50-60%) in hopes of filling demand, and that is why prices are dropping on 7900 series video cards.

These voltage regulators are located on your card directly below the GPU on the underside of your card. The regulators are a 3-pronged small chip with the terminals bent at a 90-degree angle.

These regulators will and can degrade over time (esp. 3-6 months). The OC or Superclocked versions of these 7900 cards will exibit artifacting and graphical errors before the standard clocked version due to the higher current draw of overclocked cards.

This problem effects ALL Nvidia 7900 series cards, regardless of make (IE: BFG, XFX, EVGA...) and is the direct fault of Nvidia, NOT any card distributor.

This problem was originally addressed from HardOCP.com under the mis-understanding that it was a problem caused by overclocking from vendors, which it is not. BFG, XFX and EVGA ALL are trying to improve quality control to solve this problem, but EVGA has shown the best effort so far.

Please do NOT return ANY defective 7900 series card to any retailer (store) that the card was purchased from. You may get another defective card. Instead, please use any RMA procedure from the vendor that distributed your card as they will test your replacement card and try to assure your RMA will be trouble-free.

I have forwarded this issue to HardOCP, TomsHardware, Anandtech and Overclockers.com for further investigation.

Your welcome.


Al Capone[CC]
Click to expand...

Now apparently this information comes from an internal source in the eVGA software engineering department. If this is true it sure does provide a good explanation for the problems that seem to affect most of the current run of 7900 series boards.

This isn't a far fetched explanation by any means, especially when you consider that ALL 7900 boards are actually made at a singular facility (Flextronics) and then distributed to the board partners (eVGA, BFG, XFX, etc) for cooler installation and branding/SKU.

I'd be willing to bet the VRM in question is the one just to the right of the GPU area when looking at the backside of the card. On my defective 7900GT thats on it's way back to eVGA, the voltage regulator chip in question is an AZ7805D linear regulator made by Advanced Analogue Circuits. The datasheet is here (PDF).

Here are photos of the VRM on my defective 7900GT:



It looks like the sane or similar VRM is on the GTX boards in the same spot on the backside of the card. :eek:

<EDIT>

Below is a photo of the same voltage regulator chip on my replacement card sent by eVGA on Thursday evening by next day air:





As you can see it's the same chip from the same manufacturer, but the week and batch number is obviously newer. There are also some pretty obvious signs of re-soldering so I'm thinking that eVGA might have replaced this chip before sending the card out to me.

AGAIN - I don't know for sure that this is the voltage regulator in question, but it's the only one on the underside of the card. (There is an similar one on topside of the card near the DVI outputs). This could be the regulator in question, it might not be, or the whole story might be a very legitimate looking false rumour.
 
What kind of ticks me off about this, if it ends up being true, is that passage about [H]ardOCP addressing the issue under a misunderstanding. If there was any misunderstanding going on it sure as heck was on nVidia's part as they are the one who placed the blame squarely on the board partners for shipping pre-overclocked cards. Hmph!

I'm glad the issues are finally being revealed as they truly are instead of all this white noise we've been seeing recently on the topic. I've been pointing out all along that these issues don't just affect the pre-overclocked cards. They affect the stock clocked cards just as badly, it just doesn't manifest as fast as it does on an overclocked card.

I'd be willing to bet this same VRM is used on the GTX somewhere, as there seems to be almost as many GTX users with failing cards the past couple of weeks. [EDIT: Yup, same damn voltage regulator in the same exact spot on the GTX cards]
 
Thankfully I bought a 6800GS :D

This is going to be a pain in the ass for retailers/suppliers.
 
I am glad that I didnt upgrade my system. I was planning on buying the eVGA 7900GTX Superclocked :D. And a new Cpu and Mobo as well.
 
Doesn't make sense to me. You telling me that nvidia is not only supplying the GPU, but also the supporting electrickery?

Doesn't make sense to me at all. A voltage regulator isn't a new or especcially hi-tech or proprietary piece of equipment. It's a fucking voltage regulator. Theyv'e been around for decades. Why in hell would nVidia supply the V regulators to card manufacturers? There are many manufacturers of high-quality electronics components...

Is nVidia forcing card makers to go with a certain brand or component? If so, WHY?

Doesn't make sense to me at all, and I call BS. While the vregs may be the cause of the artifacting, it looks to me more like the component SPECIFICATION is limiting the overclock, simply because the voltage regulators (as specified in the reference design) simply cannot provide more juice. I don't think it's a case of bad binning or anything like that. The vregs simply need to be more beefy in order to support non-reference clock speeds.

EDIT: Ah, nevermind, I need to learn to fucking read! Okay, so the boards were all made at the same facility, so it's certainly possible.
 
PhyberOptic said:
Is nVidia forcing card makers to go with a certain brand or component? If so, WHY?
Click to expand...

No.

As I understand it all of the 7900 boards (and most of the first runs of the other nVidia boards) are manufactured by a company named FlexTronics. The boards are almost completely assembled there except for the coolers which are added by the board partners (what you called "Board Makers") once they recieve them. The board partners do the branding (and BIOS alterations if needed) and then box them up for retail.

nVidias reference design is what dictates the components used. As for wether or not it dictates this particular brand and model number of voltage regulator, who knows. The component may have just been the cheapest and easiest for flextronics to get hold of.

This doesn't mean the board partners can't make their own cards, and a few of them actually do months after a particular GPU is released. I think BFG has a couple of 7900 models that are actually made in-house.
 
Heh, yeah :eek: I read the quoted "bulletin" but not the text that followed. Certainly is possible

This actually makes me wonder about a lot of the computing industry, and the wool that is pulled over our eyes in terms of what we think we are buying, and what brand we think we support.
 
How many reference cards do you reckon stay at reference clock speeds?

And what about those hidden "nvidia/Ati speedup" settings that you find in many bioses...
 
"The vregs simply need to be more beefy in order to support non-reference clock speeds."

this would only make sense if only the overclocked were breaking.. i think
 
Well my 7900GT aint that old and it has AS7805 VRM not a AZ one. Also batch nummber is starting with letter A.
Mine is a XFX Vivo version with stock speeds(black PCB).

Well We all look at VRM module nummber but crusial nummbers are 7805 and those nummbers state it's positve line 5 V regulator. Like 7905 would be negative line and lets say 7812 would be 12 V regulator.

It's 1 A part hence not realy a monster. I wasn't tracing lines to find out whats purpose of it but place on a PCB and it's current and voltage rating tell's me it ain't that important VRM as once found on PCB on other side next to a power connector.
Still if it mailfunctions it will make odd things.

Any 78xx can be rated for different current and voltage rating. How is it going to preform deppends on it's specs, cooling solution as wel as quality of input side. In thi case it's SMD part and cooling is provided by PCB it self and it should be in specs. Though on grafic cards there are so many components using PCB for a cooling. Even a GPU transfers lot's of it's thermal energy thrue it's BGA. OC it and situation gett's even worse.

If VRM is overloaded it will try to supply how much it can. If a input is stable and can supply demanded current everything deppends on VRM. There are safety margins hence VRM can deliver bit more then it's rated for. Though if it's non stop overloaded and badly cooled it will die. There is a reason why Ultra cards have had double VRM circuits and GT cards have had single one. Although it was posible to OC easy GT to Ultra speeds VRM cicruit was suffering. thats also a reson why Ultras where better OC-ers although they have had same core. Mostly PCB is a same just higher models have more components.

On other hand if input side can't deliver stable voltage VRM will work even worse making fluctuations.

Overloading VRM will cause voltage drop, instability and eventual failure.

Bad batch of VRM will crash and burn.

It doesn't suprise me all Nvida cards are build at one place.
Many products are made in single Factory and then delivered as semi product to a branded factory.
Like cores and CCD's and TFT and even a optical drives and mediums come just from few factories. Dell is not making their own TFT panel for 2405, they just use egsisting panel.

I see also why nvidia would need to have controll over a components choise specialy about VRMs and elcos on PCB. Those are crusial for cards stability. Also if part supports a 1 A current and current on stock speed is like 800mA there is a small margine for OC. Though if they use 5A part on same place there will be enough current to burn something behind it. Also if 5A part is like 30 cents more exprencive on 1 million of cards using same part it's a lot's of money for unneeded capacity.

If 1A rated 7805 works and they have used bad batch it's just bad luck. Hence Factory that made VRM's is to blame. Nvidia is one to blame covering it up.

Well I gues all new cards are made with good VRM's and cards that fail will be RMA-ed so We doo not need to suffer to much.

Damn shame cause 7900GT are great cards for godd price.

BTW although AC Silencer and Zalman VF coolers cool core with almost same efficiency and dispite a fact that AC cooler has exhaust system I've went last few times for Zalman as it cools pcb as well as GPU and ocupies less place. My 7900GT is cooled with VF900 and I'm very satisfyed. It fits my SN25P as well.



MD
 
Would be nice if Kyle and Brent commented on the issue or contacted Nvidia for a more detailed explanation. If this is true, Nvidia should have done a full recall, kind of like Ford did with thier explorer series a few years back, instead of knowingly sell thier customers faulty equipment. AEG and now this. I am quickly loosing faith in Nvidia. :confused:
 
MD said:
Well my 7900GT aint that old and it has AS7805 VRM not a AZ one. Also batch nummber is starting with letter A.

>snip<

MD
Click to expand...

good post! i'm wondering if sticking a ramsink on the VRM will help cure problems.
 
PhyberOptic said:
good post! i'm wondering if sticking a ramsink on the VRM will help cure problems.
Click to expand...

Doubtful.....
The instability isn't coming from heat. Most VRMs are pretty insensitive to heat. Even the elcheapo 5v DC hobbyist VRM from Radio Shack will operate up to almost 90C before requiring a heatsink, and it's got a metal tab for installing to a heatsink.

The instability is coming from the bad VRM's inability to carry current, and/or maintain regulated voltage, which could be caused by a lot of internal faults in the semi-conductor.
Like narrow traces, excessive resistance, bad silicon, etc ad nauseum.

If temp of the VRM is getting too high then it's a symptom of the problem, not the problem itself.
Kind of like getting a fever when you have the flu. If you just treat the fever, then you might feel a little better for a while but you still have the flu and your illness might just get worse.


 
FearTec said:
Are AGP 7800GS with 7900GT cores (24 pipes) broken too?
Click to expand...

I believe Gainward (I can't remember where I read it) makes their own PCB's, at least for their 7800GS (7800GT/7900GT core)...

Terra - I guess I could turn off my PC and look at the numbers on my card...but..I mean...turn off my PC :eek:
 
NVIDIA does not supply or make the IC components. They buy the items at spec from suppliers. This regulator is made by BCD. Also, the part is spec'd to 125C. So, an the 80C temps people have noted under load is well within spec, as EVGA has noted.

The poster at EVGA did not disclose any verified validity to this announcement other than the fact they posted it. So, it could all be some personal take on the issue that the poster has supplied as a known fact. BCD has not announced any defective part recalls or the like.
 
HeavyH20 said:
NVIDIA does not supply or make the IC components. They buy the items at spec from suppliers. This regulator is made by BCD. Also, the part is spec'd to 125C. So, an the 80C temps people have noted under load is well within spec, as EVGA has noted.
Click to expand...


I don't think it's the same component as the one you're referring to, and it has nothing to do with temperatures.

Anyway the whole story may just be made up, I just saw it posted and it sounded quite logical to me, so I figured I would share it and my opinions. :)
 
That's odd.... that's exactly what I've been thinking the culprit was when I made my post yesterday on x-3Dfx.

You see nVIDIA dictates what components (specs for components) are to be used on the boards themselves before they're shipped off to 3rd party manufacturers for branding. Trying to keep the price down on the 7900 series (to compete with the x1900 series) the pressure to find cheaper components must have been quite high.

In other words nVIDIA should come out and announce this. Stop making 7900 series cards or selling them until the problem is fixed. They should also apologise to all 7900 series owners (myself included) and announce a FULL recall.

Of course they'll never do that. nVIDIA isn't the most honest customer focused company (GeForceFX).
 
I'd like to AGAIN remind everyone that this COULD be a load of complete BS. Anti nVidia propoganda being smartly written by someone with lots of time on their hands and the urge to "stir shit up". Take ALL of this with a grain of salt. Mmmmkay? Need I re-iterate this point any further?
 
Blue Falcon said:
I'd like to AGAIN remind everyone that this COULD be a load of complete BS.
Click to expand...
That's what it sounds like. Obviously there is a problem, but several of the points in that post ring shens alarms. For someone with "inside info," he seems pretty ignorant of even basic information.

But the Internet is like that, even a mental case has a wide audience. ;)

(edit) FWIW, the VR on my card (same manufacturer as in your picture) has a date/batch code of 524BJD and I haven't had any problems. I guess that's my prize for ordering on launch morning. :D
 
Blue Falcon said:
They are ignoring higher than normal failure rate (50-60%) in hopes of filling demand, and that is why prices are dropping on 7900 series video cards.
Click to expand...

I doubt that number is accurate.
 
aldamon said:
I doubt that number is accurate.
Click to expand...

Considering what I've been seeing here and elsewhere I would say the failure rate is closer to 30-40%.

Maybe even as high as he's claiming. I bet if we made a post that asked if users experienced issues with tehre 7900 series we'd be astounded by the results.. in fact.. let me start one.
 
Blue Falcon said:
I'd like to AGAIN remind everyone that this COULD be a load of complete BS. Anti nVidia propoganda being smartly written by someone with lots of time on their hands and the urge to "stir shit up". Take ALL of this with a grain of salt. Mmmmkay? Need I re-iterate this point any further?
Click to expand...


Yes, what do you mean? Complete BS? I thought everything posted on forums was 100% fact. :p
 
I received my PNY 7900GTX's yesterday, PNY was late to the game with this series, and the only reason I went with them was that I had a Dell.ca credit from a laptop, and that was the only brand they carried. They weren't due to ship from PNY until june 12th. I have tested them in F.E.A.R. and 3dmark 06, which are supposed to be 2 apps that can get the prob going. I haven't seen any artifacting yet, just some weird colors when using the 87.25 betas, which others were having with the same drivers.

I just pulled the heatsink to check out the VR, it is a completely different brand/model. The symbol on it looks like a combined AP and it carries the # D408 BH6A17, both VR's in the positions mentioned in this thread, are the same. The card auto-detected to 672/854, but I've only ran it in apps at stock speeds. Only time will tell I guess
 
He said COULD be BS, but frankly it's the most plausible explanation so far..I didn't buy Nvidias passing the buck onto the board partners a bit, especially considering there's plenty of people even with stock clocked cards having their card affected. I don't think we'll get the truth though, even if it is correct, and it (the latest 'explanation') did originate @ eVGA, they're not going to piss off NV and out them..Not unless (like I suggested elsewhere) there's a class action lawsuit started due to knowingly making/selling defective parts on NVs side with board partners included since they are the ones ultimately selling them, but have nothing really todo it seem w/ the manufacture ofthe cards...[cue french accent](America - the land of sue anyone if you fart in their general direction).
 
There's a response from eVGA in the thread now: http://www.evga.com/community/messageboard/topic.asp?TOPIC_ID=15800

They are calling the post baseless but gave some interesting information:
We would like to state beyond that post the following official statements:

1. EVGA has sold tens of thousands of 7900 GT Series cards since the launch. Because of this EVGA will not comprise even 1 customer being unhappy with his purchase of an EVGA product; that is why we have setup a special web page, making out bound calls, and offering direct cross-shipments with highly tested cards to those users who experienced some issues. We would like to share with you the RMA submission rate is 0.74% for all 7900 series.
Click to expand...
"tens of thousands" at a 0.74% defect rate is still a lot of bad cards. That would be 740 bad cards out of each 10,000 sold. What's worse is that RMA rate doesn't seem to count cards that are returned to the vendor from the "RMA submission rate" figure. The defect rate could be multiple times higher, possibly as bad as mid-single digits.

1 out of 20 cards having problems would be a huge failure rate. And don't forget that people mostly go to support boards when they are having problems. So the posts you read tend to be about problems.
 
pxc said:
There's a response from eVGA in the thread now: http://www.evga.com/community/messageboard/topic.asp?TOPIC_ID=15800

They are calling the post baseless but gave some interesting information:
"tens of thousands" at a 0.74% defect rate is still a lot of bad cards. That would be 740 bad cards out of each 10,000 sold. What's worse is that RMA rate doesn't seem to count cards that are returned to the vendor from the "RMA submission rate" figure. The defect rate could be multiple times higher, possibly as bad as mid-single digits.

1 out of 20 cards having problems would be a huge failure rate. And don't forget that people mostly go to support boards when they are having problems. So the posts you read tend to be about problems.
Click to expand...

Well last week they reported a ~2% RMA on the 7900 series and I highly doubt that the number has gone DOWN since then. There may be some public relations damage control at work here.
 
I wonder how many cards were sold to Dell or the likes and 3dmark06 or Oblivion have never been ran. There are probably MANY "closet" broken cards out there that will never surface.
 
Blue Falcon said:
Well last week they reported a ~2% RMA on the 7900 series and I highly doubt that the number has gone DOWN since then. There may be some public relations damage control at work here.
Click to expand...
If you read the link they break it down: "EVGA’s 7900GT Series total RMA rate is from 0.04%-1.9% per different SKU’s." The 0.74% RMA rate is for all 7900 cards, also explained in the link.

And oops on the math. It's 74 per 10,000 as corrected above. And EVGA definitely has a much higher defect rate than that, which is why I don't think vendor returns are counted in that RMA rate.
 
digitalx0 said:
Actually, that's only 74 bad cards out of 10,000. It's really not too bad.

10,000 X .0074 = 74
Click to expand...

What's even more amazing is that if you go to the EVGA boards and count all the defective ones you're most likely to find that pretty much all of these cases have been reported on the EVGA forums. I didn't know their customers like the forum that much. In fact there might be even more than 74 reported cases on their boards, I only counted the GTX cards and those numbered around 40 when I last checked.

Hmm actually he said tenS of thousandS so not all the defective ones need to appear on the forums. Anyway, how many people with defective equipment would be expected appear on the manufacturer's forums? At the moment there are 138 posts in the "tally of defective 7900 cards" -thread, most of which are problem reports. Some with multiple cards.
 
You must log in or register to reply here.
Back
Top