Serious design flaw on certain 4870/4890's

[Final8y]

if your really clued up in this then you would have seen that 2 apps have appeared that use very simple graphics rendering methods and seem to be bringing these cards to their knees.
and the point i have is that there are 2 apps already out there that cause this problem to appear.

are you saying that ati cards are only for gaming only and people with them should not be using furmark or occt?

regardless of the nature of these apps, they are nothing special in much form, they look fairly bland although could make good use as screen savers, never the less id be proper PEED off if i bought a bit of hardware that konked out running any sort of software.

I have used both apps & my cards passed.
I was aware of the issue a long time ago nether did i care then nore would i care now as the intend purpose is what matters to me.

If gfx card in bought with the purpose to run furmark & OCCT then don't buy a 48xx card.

 

[Cyber-Mav]

I have used both apps & my cards passed.
I was aware of the issue a long time ago nether did i care then nore would i care now as the intend purpose is what matters to me.

If gfx card in bought with the purpose to run furmark & OCCT then don't buy a 48xx card.

you seem to have changed your tune now. amd are now also aware of the issue and they say it will be fixed in future revisions, but does that mean a recall for existing cards?

"If gfx card in bought with the purpose to run furmark & OCCT then don't buy a 48xx card"

and regardless of everything else now the truth comes out, you cant use ati cards for running certain apps will probably become a catchphrase now.

why did they make a gpu that the user is not allowed to use at 100%

 

[Lightnix]

could you take a photo of the vrm section of your card and post on here pal. would like to see what sort of cooling is on there as standard.

I'd like to but I think that voids my warranty, something I'm kind of inclined to keep given I spent about £180 on the card only in February. Unless a staff member from OcUK can guarantee that it's okay to do I'd rather not. Sorry. Looking at the back of the card I can see it's a three-phase system, though. As far as I know the XFX cards are completely reference apart from the black PCB.

EDIT: Ignore this, I did stupid whilst running the test.

Anyway, there's some test data, a screenshot of OCCT after completion and a rivatuner log for anyone who wants to look at it here:
http://lightnix.co.uk/occtthing48701gblightnix.zip

 

[layte]

you seem to have changed your tune now. amd are now also aware of the issue and they say it will be fixed in future revisions, but does that mean a recall for existing cards?

"If gfx card in bought with the purpose to run furmark & OCCT then don't buy a 48xx card"

and regardless of everything else now the truth comes out, you cant use ati cards for running certain apps will probably become a catchphrase now.

why did they make a gpu that the user is not allowed to use at 100%

Probably because the only time you actually can load the GPU up in such a fashion is using an artificial tool that has no baring on any real world situation. I hate to use a car analogy, but if I sat my car in neutral and kept my foot on the accelerator something would fail sooner rather than later, why did they make a car that the user is not allowed to use at 100%. A similar situation to what we have here.

It could be worse, the GPU's could melt themselves instead of just failing.

 

[Cyber-Mav]

did you have anistropic filtering forced on in the CCC for this test?

 

[Final8y]

you seem to have changed your tune now. amd are now also aware of the issue and they say it will be fixed in future revisions, but does that mean a recall for existing cards?

"If gfx card in bought with the purpose to run furmark & OCCT then don't buy a 48xx card"

and regardless of everything else now the truth comes out, you cant use ati cards for running certain apps will probably become a catchphrase now.

why did they make a gpu that the user is not allowed to use at 100%

I have not changed my tune at all seeing as you did not understand what is most important to most in the first place so i had to separate them for you.

The majority buy something for a use & not for the purposes of benching it.

 

[Cyber-Mav]

Probably because the only time you actually can load the GPU up in such a fashion is using an artificial tool that has no baring on any real world situation. I hate to use a car analogy, but if I sat my car in neutral and kept my foot on the accelerator something would fail sooner rather than later, why did they make a car that the user is not allowed to use at 100%. A similar situation to what we have here.

It could be worse, the GPU's could melt themselves instead of just failing.

well this shows that ati designed a power supply for the card that is incapable of providing the power for thier theoretical max 1.2tflop performance.

car analogys are very different to computers so you cant even compare the 2 things. one is mechanical other is electrical. you got an even less idea on computing hardware if your trying to compare it with a car.

even then a cars cooling system is designed for it to keep the engine from meltdown even if it is stationary, thats why there is a fan that kicks in when the engine goes above a certain temperature. if it borks thats what warranty is for. and dont forget car manufacturers have also had thier fair share of recalls too. amd could be joining them in on this.

 

[Final8y]

Probably because the only time you actually can load the GPU up in such a fashion is using an artificial tool that has no baring on any real world situation. I hate to use a car analogy, but if I sat my car in neutral and kept my foot on the accelerator something would fail sooner rather than later, why did they make a car that the user is not allowed to use at 100%. A similar situation to what we have here.

It could be worse, the GPU's could melt themselves instead of just failing.

That is the best car analogy that i have seen in a long while.

 

[Lightnix]

did you have anistropic filtering forced on in the CCC for this test?

Oh damn, yeah. Whoops. Thought I'd turned all of that off. Eh, I'll run the test again tomorrow (I don't really fancy running the test whilst I'm here to be honest, that particular test makes it sound like the card is about to take off). Oh well, at least I know what the fix for the problem is.

 

[Jono8]

makes me laugh that mot long ago a lot of people were saying stability is everything and cards should never fail at any test.

 

[Final8y]

makes me laugh that mot long ago a lot of people were saying stability is everything and cards should never fail at any test.

I would really really love to see those many quotes from the people here in this thread that were saying that before.

 

[Cyber-Mav]

I have not changed my tune at all seeing as you did not understand what is most important to most in the first place so i had to separate them for you.

The majority buy something for a use & not for the purposes of benching it.

your forgetting here or not understanding here, that benching is also considered as a "use". the argument that stands here now is that your defending a company that knew about this problem to begin with. they designed a power supply for thier cards that is good enough for practical maximum usage but not good enough for theoretical maxiumum usage.

all thats happened now is that an app as come out that go beyond practical usage and pushes the card to its theoretical maximum, something that amd was proud of with thier cards stating that they have a high 1.2teraflop compute power, but turns out that if you want to use all that power its not going to happen on the cheapo 3 phase reference power design of thier cards.


the reason im ticked off about this is that recently amd slapped palit and gainward for making non reference design cards that had better power regulation circuitry. why did they do this? turns out now we have the proof, amd didnt want a new commer to show up long standing board builders like sapphire etc who went with the reference design and got stung.

thats whats got me ticked on this.

 

[Jono8]

I would really really love to see those many quotes from the people here in this thread that were saying that before.

well i certainly can't be bothered to find them

 

[Cyber-Mav]

Oh damn, yeah. Whoops. Thought I'd turned all of that off. Eh, I'll run the test again tomorrow (I don't really fancy running the test whilst I'm here to be honest, that particular test makes it sound like the card is about to take off). Oh well, at least I know what the fix for the problem is.

iv read that entire thread at xsforums last night thats why i came across that post by the guy who had AF enabled in CCC and his card didnt have the problem.

mate i dont think it would be wise to run the occt while the pc is unattended. not sure if the vrm units will blow up or catch fire.

 

[Happy]

the reason im ticked off about this is that recently amd slapped palit and gainward for making non reference design cards that had better power regulation circuitry. why did they do this? turns out now we have the proof, amd didnt want a new commer to show up long standing board builders like sapphire etc who went with the reference design and got stung.

thats whats got me ticked on this.

I thought the reason they were annoyed with pallet et el etc was they used GDDR5 on the 4850s despite AMD requesting they removed it. Which is why they didn't receive any/limited stock on the new 4890s.

 

[Cyber-Mav]

makes me laugh that mot long ago a lot of people were saying stability is everything and cards should never fail at any test.

i still say that. cards cpu's etc shoudl not fail under any software test. you would expect the manufacturer would have done the testing on them.

 

[Final8y]

your forgetting here or not understanding here, that benching is also considered as a "use". the argument that stands here now is that your defending a company that knew about this problem to begin with. they designed a power supply for thier cards that is good enough for practical maximum usage but not good enough for theoretical maxiumum usage.

all thats happened now is that an app as come out that go beyond practical usage and pushes the card to its theoretical maximum, something that amd was proud of with thier cards stating that they have a high 1.2teraflop compute power, but turns out that if you want to use all that power its not going to happen on the cheapo 3 phase reference power design of thier cards.


the reason im ticked off about this is that recently amd slapped palit and gainward for making non reference design cards that had better power regulation circuitry. why did they do this? turns out now we have the proof, amd didnt want a new commer to show up long standing board builders like sapphire etc who went with the reference design and got stung.

thats whats got me ticked on this.

In not defending as i said don't buy a 48XX card if your main use is for benching OCCT & furmark.

Theoretical maximum usage is just that 99% of the time,Theoretical & would take an extremely naive person to take it seriously.

 

[Cyber-Mav]

I thought the reason they were annoyed with pallet et el etc was they used GDDR5 on the 4850s despite AMD requesting they removed it. Which is why they didn't receive any/limited stock on the new 4890s.

nah the article said that amd were peed cuz they wanted thier long standing partners like sapphire etc to have the first non reference cards out before letting other board partners getting the non reference stuff out.


also 4850 with ddr5 is a proper slap in the non reference design standards. thats basically a 4870 then lol. how can any other board partners 4850's compete with that?

 

[Cyber-Mav]

In not defending as i said don't buy a 48XX card if your main use is for benching OCCT & furmark.

Theoretical maximum usage is just that 99% of the time,Theoretical & would take an extremely naive person to take it seriously.

but the thing is you shouldnt have to alter your purchase because of 2 apps not being safe to run. 4870 is a decent card, i can now see how ati managed to get good prices on these, and thats by going cheapo on the power side of things.

also your wrong about saying not to use a 48XX card for occt or furmark. you can easily use a non reference design card that has 4 phase or more power regulation and never have any issues with it. its just the reference design that needs a slap.

 

[Final8y]

but the thing is you shouldnt have to alter your purchase because of 2 apps not being safe to run. 4870 is a decent card, i can now see how ati managed to get good prices on these, and thats by going cheapo on the power side of things.

also your wrong about saying not to use a 48XX card for occt or furmark. you can easily use a non reference design card that has 4 phase or more power regulation and never have any issues with it. its just the reference design that needs a slap.

Thing is i would not alter my purchase because of those 2 apps no matter what the brand.

Yes you can use non ref 48xx, but the majority of 48XX are ref.