Fixing a Bad BIOS Flash on the 4870

[NKD]

There does seem to be some correlation between flashed cards and dead cards. But it isn't anywhere near perfect and we are just guessing that is the culprit. An even stronger correlation is between an altered LED pattern and the failed card state. It would be wise to consult ATI and find out what the LEDs represent. It may very well be that in some cases the user is at fault - but we can't be sure without more information.

And even if the users are at fault blaming them doesn't really help them at all. What they need is information because that's what they need to make the decision of how to proceed. We are all aware of the risks in modifying hardware, there is no need to berate each other for trying something a little extreme or risky.

I agree, even if I am at fault I would love to know that why the bios flash acted good for two days and than my card died. and what the led pattern means.

[NKD]

I don't know where you are going with this, but you're not making sense. Of course review sites had cards prior to launch. How do you think they get the reviews ready for launch day? It takes time to do a review, why wouldn't they have cards before the offical launch? What do you think the NDA is for?

You say there is "no reason" but you can't make that claim, bad BIOS code can most certainly kill hardware, especially when it powers up/initializes.

we could argue all day, but that won't find a solution. I am getting myself another card soon if there is not a new flash tool. so lets just wait and see, all I was trying to say that the cards were finalized not long ago, so there might be some issue. thats all. there is no need to go any further in this, as it won't do any of us any good.

[NKD]

LOL, I was atleast happy to see those good temps for two days, the damn cooler is pretty good, only if ati had set the fan speed to 31% all the time, I would not have flashed, and may be it would be fine, or may be it would be dead, that is a mystery to me, other than the part that my card is dead. and Yes I am laughing at myself.

[SimBy]

LOL, I was atleast happy to see those good temps for two days, the damn cooler is pretty good, only if ati had set the fan speed to 31% all the time, I would not have flashed, and may be it would be fine, or may be it would be dead, that is a mystery to me, other than the part that my card is dead. and Yes I am laughing at myself.

For all those who want to increase the fan speed without the bios hack you could try this:

http://forum.beyond3d.com/showpost.p...&postcount=207

At least until we find out wtf is going on with these card deaths.

[SimBy]

http://www.rage3d.com/board/showpost...81&postcount=6

[W1zzard]

1. could someone describe the LED configuration before and after the card dies? what are the differences?
2. to find the bios flash i described earlier search the thread for -newbios
3. can someone with a bad card measure vgpu and vmem please?
4. can someone with a working card save their bios using winflash latest version
5. Look on the PCB for a chip PM25LV512A or PM25LV010A (8 legged IC). 512 = 512 k ROM, 010 = 1 MBit ROM

if there is not a new flash tool.

there won't be a new flash tool soon. i already posted the lastest yesterday @ tpu downloads but its changelog says no fixes for rv770

[Edin]

hey W1zzard...

1) when I first installed the cards there are 4 LED ligths that would lit up (LED's are located right next to 2x 6-Pin power connectors) and now only 3 of them will flash once just when I press the power button (starting up) or when system restarts. I can't remember exactly but before I think the lights were lit up more then once during bootup (someone with working card can double check)

2) I tried the original BIOS and tried the -newbios command and it made no difference on my end

3) On the picture you provide to measure the voltage on HD4850 there are no such solder points (card layout on HD4850 is different from HD4870) so I don't know where to measure the voltage. You can see image of HD4870 (top and bottom side) I took this morning, here: http://www.xtremesystems.org/forums/...&postcount=126

[W1zzard]

1) do you have a photo of that area by any chance? or a link to a highres pic of the card? none of the lights stay on ?
3) sorry i confused 4850 with 4870. does someone have the link to a 4870 voltmod article?

[Edin]

1) do you have a photo of that area by any chance? or a link to a highres pic of the card? none of the lights stay on?

yup, look at this pic (top/left side right next to 6-Pin connectors)
http://www.3sqn.com/members/images/H...ottom_Full.jpg

LED's are labeled D1602, D601, D1601 and D1603. The one that doesn't lit up now is labeled D601.

[DemonEyez]

This is why I always cringe when I see cards using this Quimodo crap. I remember hearing of all the issues this brand of ram caused on the 8800GT cards on release. They are just no where near as robust as Samsung branded ram. Once the 8800GT starting using Samsung ram the problems were gone. ATI should have stuck with Samsung ram I feel. Just saying.

Why do people immediately blame Qimonda for all the problem. With the 8800 GT's going bad people said it were the Qimonda chips crapping out, why was it then, that with the 8800 GTS which used the same Qimonda chips, but were connected to a beter power/vrm circuit (3-phase power on the GTS opposed to 2-phase on the GT) didn't have those problems.
Until there's some clear evidence it's all pure speculation

[r4st4m4n]

yup, look at this pic (top/left side right next to 6-Pin connectors)
http://www.3sqn.com/members/images/H...ottom_Full.jpg

LED's are labeled D1602, D601, D1601 and D1603. The one that doesn't lit up now is labeled D601.

Not sure if that helps but I noticed that in the 4850/4870 voltmod thread,one of the Vgpu mod possible point is labeled C601 ... It may indicate that the LED D601 and Vgpu are somehow related

Here are the posts about C601: http://www.xtremesystems.org/forums/...7&postcount=35
http://www.xtremesystems.org/forums/...&postcount=108

[W1zzard]

Not sure if that helps but I noticed that in the 4850/4870 voltmod thread,one of the Vgpu mod possible point is labeled C601 ... It may indicate that the LED D601 and Vgpu are somehow related

Here are the posts about C601: http://www.xtremesystems.org/forums/...7&postcount=35
http://www.xtremesystems.org/forums/...&postcount=108

smart idea. one of the 4 leds is the VDDC power good LED.
the top one and the bottom one are for the EXT power connectors afaik.
i would assume the 4th one is some temperature too hot indicator.

is D601 always lit on a working card or always dim?

[Jakalwarrior]

I have nothing to add here but im going to throw out a guess so if I'm right in a few months I can gloat or something.

I'm guessing it is possibly the error correction stuff on the RAM allowing cards to seem functional at high memory clocks meanwhile the ram is way past "stable". Perhaps they can even get past internal QA like that.
I remember reading some review that said the ram would correct errors when unstable and overclockers would have to look out for lowered scores. That leaves me thinking though that a small to moderate amount of instability would be nearly indetectable without a direct measure of the memory bandwidth.
Remember how they also said the ram auto negotiates some of the stuff with the controller on startup? maybe the ram keeps running at the higher clocks while correcting the errors, but on reboot it cant negotiate the settings with the controller?

[wiak]

I have nothing to add here but im going to throw out a guess so if I'm right in a few months I can gloat or something.

I'm guessing it is possibly the error correction stuff on the RAM allowing cards to seem functional at high memory clocks meanwhile the ram is way past "stable". Perhaps they can even get past internal QA like that.
I remember reading some review that said the ram would correct errors when unstable and overclockers would have to look out for lowered scores. That leaves me thinking though that a small to moderate amount of instability would be nearly indetectable without a direct measure of the memory bandwidth.
Remember how they also said the ram auto negotiates some of the stuff with the controller on startup? maybe the ram keeps running at the higher clocks while correcting the errors, but on reboot it cant negotiate the settings with the controller?

and the flash messed that up
might be

[jimmyz]

lights should all be off on working cards so the lights that stay on indicate a problem, unless ATI has switched it up this time, anyone with a working card please confirm.

[Mechromancer]

What brand cards are having these failures? Is it one specific brand, like HIS, or are these failures over all brands? If those of you with failures could post your card's brand that would be helpful. I'd like to determine the circumstances behind these failures and the only way to do that is to catalog all the details.

Please post this info:

GPU Make & Model:
GPU BIOS version: (if you have it)
Motherboard Make/Model/BIOS:
Display Driver version used:
Failure under Crossfire?: (yes/no)
Other Details: (any notable details before/after the failures)

Those are just a few variables I could think of that would cause this problem. I'd like us to be able to distinguish between a GPU issue like a board defect, or an outside cause in these particular cases.

[Logos]

What brand cards are having these failures? Is it one specific brand, like HIS, or are these failures over all brands? If those of you with failures could post your card's brand that would be helpful. I'd like to determine the circumstances behind these failures and the only way to do that is to catalog all the details.

Please post this info:

GPU Make & Model:
GPU BIOS version: (if you have it)
Motherboard Make/Model/BIOS:
Display Driver version used:
Failure under Crossfire?: (yes/no)
Other Details: (any notable details before/after the failures)

Those are just a few variables I could think of that would cause this problem. I'd like us to be able to distinguish between a GPU issue like a board defect, or an outside cause in these particular cases.

correct me if I'm wrong but I thought all brands provide reference boards at the moment, so isolating one brand in case of failure doesn't really make sense...other parameters, the one you mentioned yourself, are all valid and interesting.

[Marvin_The_Martian]

Both the people on the Folding Forums and offcourse [H] think each and every one off is lying and your cards all deid because you overclocked them to much or flashed them badly, and you know what proofs it? Because this is XS

But one off them had a point, they tested folding on those cards before they were released and I need to say 24/7 folding is heavier on a card I think then what 80% off this forum ever puts their cards through ( no wait it is xs so make it 50% ). They had no issues with them dying, what changed? I posted on the folding forum that instead of blaming it on overclocking without a second thought they should try to help Wizzard vind out what happend. If it's not a hw failure but user caused then that will come out to, but what I see is people claiming cards died without flashing or overclocking and that doesn't sound like a user mistake.

On the other hand, not to many threads about them yet. If people who don't have issues far outweigh those who do, it should be clear to because I'm not sure if it's because their still hard to get in some places or because it's just so few people with problems. So maybe we also should do a count off good vs dead cards here?

Edit: nm on that last thing, http://www.xtremesystems.org/forums/...d.php?t=192906

Edit2: Perkam wrote everyone who has a dead card flashed it. So, who's lying here? If you flashed and leid you should be banned imho, abusing a forum to try and get an rma

[mascaras]

HD4000 séries FAN FIX ( without flashing Bios ) >> http://www.xtremesystems.org/forums/...89#post3100889

[Mechromancer]

HD4000 séries FAN FIX ( without flashing Bios ) >> http://www.xtremesystems.org/forums/...postcount=4045

Very nice!

I'm starting to have doubts as to the quality of this Qimonda GDDR5. Apparently the 8800GT had problems due to their memory as well. I've also read a bit about how GDDR5 works and its NOTHING like the previous generations of memory. It varies clockspeed, voltage, and other parameters automatically to resolve signaling issues. Screwing with the BIOS is probably messing with this function in some way.

GDDR5 Articles:
http://www.extremetech.com/article2/...2310000,00.asp
http://www.eetasia.com/ART_880050357...P_74d933d1.HTM

Read up on how "signal training" works for yourself and lets discuss it. I'm no memory expert so I won't pretend to know how it all works. It just seems like these BIOS mods are the culprit.

[C'DaleRider]

HD4000 séries FAN FIX ( without flashing Bios ) >> http://www.xtremesystems.org/forums/...89#post3100889

Good link.....that's almost exactly like the profile edit that was used on the 3870's to modify the fan speeds, among other things. It's what I used on my 3870 to spin up the fan....and worked like a charm.

Like was said, easiest and safest way to modify your fan speeds without touching your BIOS, which seems to be a very dicey and potentially expensive proposition right now.

[WiCKeD]

Qimonda = Infineon. It's been used on ATi cards forever.

So far I have only seen 1 user beside Theta report dead cards without a bios flash and that also happened when he entered Crossfire.

[Solus Corvus]

I get a HIS 4870 monday. Thank you SimBy and Mascaras for the profile hack. Seeing these problems I won't be overclocking or flashing the card right away ( not until there are proper tools and full cover water blocks, down the line ), but I will be using the fan speed hack to keep it cool at least.

With my luck it would be the high power draw fan that is killing the cards.

[Theta]

Little more detail about our 4 dead 4870s:

1) Lived for just over 48 hours from arrival.
2) Were retail boxed / non-review / HIS brand.
3) Lived for 2 days of testing, heat levels ~ 90C @ load.
4) Died approx. 2 hours after enabling CrossfireX.
5) Died after 3 reboots each between XP 32, XP 64, and Vista x64.
6) No BIOS mods, no fan mods, SmartDoctor not used.

Background information:

1) Still have ~10 working 4870s, none put into CrossfireX yet, due to symptomatic failure of others. Waiting to see...
2) No problems with any 4850s in any configuration - Tri and Quad CrossfireX tested.


Looked back at the packing slip, and the HIS lot was received from the same origin... Distributor is thinking that a bad batch may have leaked, due to a large number of HIS already coming back. I do know that these are reference cards, but it is conceivable that a bad batch slipped into production.

Lastly, I am intrigued by one thing here - flashers and non-flashers seem to be reporting the same events, down to a gnat's ass. Same no-post, same LED, same symptoms. Perhaps the flash brought about something that can die, or has the potential to kick in. I'll leave it in the hands of you all to figure out, just wanted to add a few last details.

These are getting shipped out on Monday, and we'll be hoping for the best on the others that we have so far. Good luck to everyone.

[OndraSter]

Can you download BIOS from dead HD4870?

If yes, try to flash the original BIOS and then download it again to check, if flash went OK or bad and get known if it is because of BIOS or damaged card.

If it's bad idea, sorry for this.

// edit: at least, that this isn't doing HD4850, because I want to buy one of them.