SweClockers: Geforce GTX 590 burns @ 772MHz & 1.025V

[GaiaHunter]

so is there a formula to measure time-to-death ? It maybe handy, because you will know the limit, when overclocking

There is no single formula that "fits all sizes". It will have to be done on an individual component basis.

In the case of GPUs, CPUs, etc, it is enough to say that most will last longer than their real useful life (read before becoming obsolete). On the other hand if you plan on keeping the same GPU or CPU working 24/7 for 10+ years, then it might be better to not subject them to additional stress by OCing.

 

[Idontcare]

so is there a formula to measure time-to-death ? It maybe handy, because you will know the limit, when overclocking

There is, the formula is that of a Weibull distribution.

The problem, for us laypeople, is that we don't have the lifetime reliability data which is necessary in determining the specific Weibull parameters that correctly characterize the time-dependent reliability of the specific VRM's in question here.

All we can say is that there is such a distribution, its parameters do exist (albeit unknown to us laypeople), and the failure rates and lifetime reliability expectations can be adversely impacted via the traditional degradation pathways (temp, amps, volts, etc).

We don't need to know the specifics of the equation and its parameters to be able to justifiably conclude that anything we do which results in elevated amps or elevated operating temps will have a deleterious effect on the operating lifetime of the product.

What we can't speak to is the absolute magnitude of the reduction on operating lifetime. To speak to that we do need the Weibull parameters for these VRM's.

 

[Phynaz]

Voltage Controller (PWM) = CHil CHL8266
http://www.chilsemi.com/products/multiphase-digital-power-controllers/chl8266/

Inductor = A101 1050F no specs found

MOSFET (DrMos) = Infineon TDA21211
http://www.infineon.com/dgdl/PB-TDA...e6cb4&fileId=db3a304323b87bc201240b3c2da847b1

Photos
http://translate.google.gr/translat...dia-geforce-gtx-590-studiu-de-overclocking/12

My analysis of the power MOSFETs:

They are rated for a absolute maximum current of 35A, and all the measurement data only runs to 30A. This would lead one to assume that a normal operating condition for these parts is 30A or less.

Now for a little math - using the figures of 0.910 and 0.963 volts mentioned earlier in this thread.

30 x .910 x 10 = 273w nominal
30 x .963 x 10 = 289w nominal

35 x .910 x 10 = 318w max
35 x .963 x 10 = 337w max


If the GTX590 is a 350w+ card as Anand claims, then the power supply is being run out of spec.

 

[Morg.]

I think I'll stop posting about how retarded the reference design for the 590 is.
But seriously, does anyone have a clue as to why nVidia found it funny to include the same circuitry as is required by a 580 ?

 

[Dark Shroud]

does anyone have a clue as to why nVidia found it funny to include the same circuitry as is required by a 580 ?

I'd like to know why Nvidia even tried to make this card? They should know the limits of their hardware

 

[AtenRa]

My analysis of the power MOSFETs:

They are rated for a absolute maximum current of 35A, and all the measurement data only runs to 30A. This would lead one to assume that a normal operating condition for these parts is 30A or less.

Now for a little math - using the figures of 0.910 and 0.963 volts mentioned earlier in this thread.

30 x .910 x 10 = 273w nominal
30 x .963 x 10 = 289w nominal

35 x .910 x 10 = 318w max
35 x .963 x 10 = 337w max


If the GTX590 is a 350w+ card as Anand claims, then the power supply is being run out of spec.

The wattage you calculated is for the two GPU chips (GF110) only. There are two more VRMs for memory per GPU chip (4 in total) but i haven't found specs yet.

 

[Phynaz]

The wattage you calculated is for the two GPU chips (GF110) only. There are two more VRMs for memory per GPU chip (4 in total) but i haven't found specs yet.

Did you hear that flapping noise?

That was my theory flying out the window.

 

[Seero]

Doc: "grats, it is a baby boy."
Mom is filled with joy.
Doc:"but he will die."
Mom bursted in tears and ask: "Why?"
Doc: "He will die, just like all living things."
Dad:"When?"
Doc:"I don't know."
Eventually, Dad beat the crap out of Doc, and baby boy died.

A piece of hardware dies due to improper usage doesn't mean it will die sooner under normal usage. By normal, I mean running it under its specification. As long as it is being ran under its specification, then it is being used correctly.
Just because 6990 doesn't fry at 1.2v doesn't mean it will last longer than a GTX590 when both are being running at stock settings. In fact, video card don't usually dies due to the death of its VRM. The life of the fan is probably shorter than the life of a VRM. In fact, the joins between chips and the circuit board probably dies due to heat circle before VRM dies.

Radeon 6990 Thermal During Load

The thermal graphs linked by OP indicate lower readings on the 6990, but it also stated in french:

Cette Radeon HD 6990 entraine des températures extrêmes en charge tant pour le CPU, que pour le chipset ou que pour le disque dur. Contrairement à ce qui se fait sur de nombreux systèmes, il sera préférable de placer un ventilateur en extraction à l’avant du boitier, de manière à éviter que l’air chaud expulsé par la Radeon HD 6990 ne monte dans le boîtier.

Translated through google.

The Radeon HD 6990 causes extreme temperatures in both the CPU load, as for the chipset or only for the hard disk. Contrary to what happens on many systems, it is preferable to place an extraction fan at the front of the box, so as to prevent the hot air expelled by the Radeon HD 6990 rises in the housing.

New 590 IR graph from hardware.fr
http://www.hardware.fr/medias/photos_news/00/31/IMG0031587.png
It is still hot, but not as red and we can see where is the hot spot, the ends are a bit cooler and the part where those 8 pin connectors are hotter than the other end.
Look at the 6990 again, the 2 ends are as cool. Why?

Now, GTX 590 draws around 350Watt, HD6990 draws 375Watt, why is 6990 cooler than 590 by 15 degree? Is it due to the heatsink?

 

[kevinsbane]

Now, GTX 590 draws around 350Watt, HD6990 draws 375Watt, why is 6990 cooler than 590 by 15 degree? Is it due to the heatsink?

Probably because of this.

 

[Seero]

Probably because of this.

Very possible.

So turning up the fan should fix it right?

Edit: Wait. Suppose it is the fan, meaning the amount of heat removed from 590 is slower than 6990, then why is the case temp as hot? Both HS vent hot air into the case, shouldn't the case temp be hotter with 6990? That was not shown on the IR graph.

Edit 2: By comparing the 2 IR graphs in OP's post, the immediate environment temp of 590 is also higher than 6990.

 

[Outrage]

just look at the backpanel of the 2 cards.

 

[bryanW1995]

There is, the formula is that of a Weibull distribution.

The problem, for us laypeople, is that we don't have the lifetime reliability data which is necessary in determining the specific Weibull parameters that correctly characterize the time-dependent reliability of the specific VRM's in question here.

All we can say is that there is such a distribution, its parameters do exist (albeit unknown to us laypeople), and the failure rates and lifetime reliability expectations can be adversely impacted via the traditional degradation pathways (temp, amps, volts, etc).

We don't need to know the specifics of the equation and its parameters to be able to justifiably conclude that anything we do which results in elevated amps or elevated operating temps will have a deleterious effect on the operating lifetime of the product.

What we can't speak to is the absolute magnitude of the reduction on operating lifetime. To speak to that we do need the Weibull parameters for these VRM's.

So what you're saying is that intel cpu's give good weibull? I like it!

 

[3DVagabond]

Radeon 6990 Thermal During Load

the problem is not the 90c GPU...

...it's the 112c mosfets in the center of the card.

I think that nVidia thought that using a fan instead of a blower would keep the VRM cool enough because it directed air downward, allowing them to use a lower cost power stage. It would also be quieter than a blower, which was very desirable too. Problem is they believed their own TDP specs. We all know how optimistic they have tended to be since the 480 was released.

There was an article at Hardware.fr some time ago regarding similar probs with the reference 460 when O/C'd above ~800MHz. When they questioned the AIB the stability tests they do are using games, not actual stress tests, like Furmark or 3DMark, etc... They stated that the cards were fine during their testing, with these games and that is all nVidia requires them to do before release.

The reviewers will often use more stressful tests than just playing games, as we all know. As so will many consumers. Also many people who buy these products don't have ideal cases. The ambient temp is not always going to be 20c. Dust does get into the cooler.

Then when reviewers get them and they start to blow up, the manufacturer starts blaming everyone else. The reviewers are doing their job when they stress cards. That's where we want them to break. Not in our machines. Once they've shown a weakness in a design, and the manufacturer turns a blind eye, why would anyone still buy it?

Speaking of buying one, are there any available? In Nz, there's one store in the country listing it. It's still listed as "pre-order". So there aren't any here. Just to give a comparison, 4 stores show the 6990 in stock with multiple brands available. One of the 4 is the same store that has the 590 listed for pre-order.

 

[badb0y]

I am going to echo the other people's theory and say nVidia didn't expect ATi to go balls out on power consumption like that.

 

[Silverforce11]

Now, GTX 590 draws around 350Watt, HD6990 draws 375Watt, why is 6990 cooler than 590 by 15 degree? Is it due to the heatsink?

False. The 590 draws 30-50W more than a 6990 in crysis and 3dmark. Using Furmark to measure load power use does not apply to the 590 as NV throttles it to 50% GPU usage.

In gaming load, the 6990 in normal mode uses ~320W. Do not spread fud that 590 uses less power or is a lower TDP card.

As for the PCB temps, NV has a much shorter PCB with components crammed in there. A higher TDP card on a smaller PCB will result in hotter PCB and components. The 6990 is noisier because of the blower fan, not because it has better airflow. I hate those fans.

 

[Morg.]

I believe it's time to stop talking about the 590, everyone agrees it's an epic fail and nVidia should not release pre-alpha versions of PCB's designed by underpaid indian construction workers.

 

[pandemonium]

I believe it's time to stop talking about the 590, everyone agrees it's an epic fail and nVidia should not release pre-alpha versions of PCB's designed by underpaid indian construction workers.

:thumbsup:

Everyone needs to get over it.

 

[Seero]

False. The 590 draws 30-50W more than a 6990 in crysis and 3dmark. Using Furmark to measure load power use does not apply to the 590 as NV throttles it to 50% GPU usage.

In gaming load, the 6990 in normal mode uses ~320W. Do not spread fud that 590 uses less power or is a lower TDP card.

As for the PCB temps, NV has a much shorter PCB with components crammed in there. A higher TDP card on a smaller PCB will result in hotter PCB and components. The 6990 is noisier because of the blower fan, not because it has better airflow. I hate those fans.

The IR thermal graph from OP on 590 measure temps off 590's VRM, while 6990 is taken on the core. Hardware.fr corrected the way temps are measured and showed that the core temp of 590 is 80c and 84c. Plus the fact that there is actually an EXTRA FAN near the lowest HDD bay when running with 6990 indicated that the comparison is biased.
Infrared thermal graph of 590
Infrared thermal graph of 6990, with an extra fan at the bottom front of the case!
This is INTERNET! Simply compare the side view of the IR graphs, focus your attention on the bottom right corner and you will see what I am talking about.

 

[3DVagabond]

The IR thermal graph from OP on 590 measure temps off 590's VRM, while 6990 is taken on the core. Hardware.fr corrected the way temps are measured and showed that the core temp of 590 is 80c and 84c. Plus the fact that there is actually an EXTRA FAN near the lowest HDD bay when running with 6990 indicated that the comparison is biased.
Infrared thermal graph of 590
Infrared thermal graph of 6990, with an extra fan at the bottom front of the case!
This is INTERNET! Simply compare the side view of the IR graphs, focus your attention on the bottom right corner and you will see what I am talking about.

Sorry, but I can't see it. Maybe I'm blind, but they look like the identical PC's. Just with the cards swapped out.

 

[AtenRa]

It is clear that there mast be a fan working in the front of the case in the HD6990 review. If im not mistaken the case is an Antec Sonata (I had one) and it has an 120mm fan in the front. The temperatures in the front department of the case (where the cables are) in front of the HD6990 is lower at idle and on full load and it influence the HDD temp.

HD6990

GTX590

 

[krumme]

I'd like to know why Nvidia even tried to make this card? They should know the limits of their hardware

Business is about earning money, a dicipline NV have been much better at than AMD.

On the consumer side the profit for NV comes from the volume part, ie the lower segments <96 cores.

In that segment perciewed performance more than brute numbers rule the decisions.

NV contrary to AMD knows they dont sell gfx with numbers, they know they sell entertainment and feelings.

Here brand value comes in. NV brand value is excellent, and their most valuable asset. Nothing is more valuable than a brand. It affect your cognitive perception of the reality - it shapes reality. Thats why a real Coca Cola tastes far better than all the no name competitors.

One of the points that strengtens a brand is to have the fastest card in the world. Just look at have NV markets the card.

Thats the purpose. So the seller at retail can say "its from NV the maker of the fastest gfx in the world."

Did NV succes in doing so. Yes. Here is why.

When you produce 1000 pcx, its not from a professional point a view manufactoring. Its a sample. The number of cards, i can tell you is about the sample size you use for development of high tech parts for tool development.

In the strictest sense. Its therefore not a card. With 1000pcx, you can not say it exsist. But its a marketing tool. For earning money - the purpose of the company and what NV is good at.

Now when does this marketing tools succeed?

Here comes the intereshting:
NV succeed the moment the reviewers compared it to the 6990

Its simply the very professional marketing and PR at NV doing their job well as they have done before. Add relation building over time and professional reviewers guide and you have the recipy.

Now there is perhaps some issues with the card. But it doesnt matter in this context. Remember NV can claim they have the fastest card. And they can shout far louder and better than AMD. Thats also the benefit of the brand.

I can tell from a high tech, high brand company, that we are someone that know we are hiding behind the brand. And i bet someone at NV knows the same.

But when the market turns red, its about to be creative, because it can not last. You can not continue forever to sell warm air. That why we see NV and Intel trying to expand into other markets. AMD just dont have the financial and ressources to do it, and they have traditionally always lagged a little bit on the business side of thinking.

 

[thilanliyan]

Sorry, but I can't see it. Maybe I'm blind, but they look like the identical PC's. Just with the cards swapped out.

I also don't see it. Seero, are you sure?

 

[cusideabelincoln]

deleterious

I must say I did not know this word existed. I have never seen it before - probably because of the popularity of detrimental. But I like it.

 

[happy medium]

I also don't see it. Seero, are you sure?

Post # 445, I see what he means now. You can see the blue in the bottom right hand corner of the 6990 system from the fan working and it seems to be off on the gtx590 picture.
That does look rather fishy.

 

[Vesku]

The 590 dumps more hot air into the case then the 6990, correct? That can easily explain the difference in front case thermals.