Real Power Consumption - 4870 X2 & GTX295 out of Spec!

[Eastcoasthandle]

I think the car anology isn't all that great and here is why. Sure, you wouldn't ask a corvette to pull a boat but you could easily go find a vehicle (truck, suv) that CAN pull the boat. So, there is a way to take that massive load and pull it within specs of a vehicle. Can the same be said of furmark? Can you find a Graphics card (maybe the Quadro) that is capable of that load like you can find a car that is capable of the load as well?

I think its funny that we can find programs that stress CPU's, 8 threads 100% max but once you find a GPU program that 100% max stresses the GPU that all of a sudden thats a bad program and no one should use it because it could harm the card... weird.

Nah, that car analogy is spot on. However, it depends on your understanding of it. What's being implied here is what the device's intended and expected use is (IE: the intended or expected use of a corvette). Once you understand that should it be clear why it was used in that context.

Edit:
Oh BTW, it's not programs that initially stress CPUs beyond spec that determines stability. It''s the applications and games you use that will determine stability. As that is (or should be) the focus of why you are trying to OC to begin with. And, those everyday applications and games are usually the real reason why you decided to overclock to begin with .

[.Tret]

I must be in the wrong place. I didn't know this was an automotive forum.

[adamsleath]

my system with an e8400@4.2GHz and a gtx 295 runs 180W (from the wall) at idle
and i have gtx 295 @ 640 core.
i think i get >80% efficiency with the antec sg-850W

@ load i see the power meter reading at around 380W

i expect load reductions with 40nm gpu's and perhaps a 65W quad core

[ZOMGVTEK]

Any GPU that is forced to create a high load will produce what you suggest. So, I don't really understand your question. Any other time or set of circumstances would solely depend on your tolerance for heat.

I was not aware that the 4870x2 would consume 400 watts under a heavy load at stock clocks. So overvolted and overclocked is going to be pumping some seriously insane levels of heat. One could likely near 500w of draw with a modded card, and that is getting a little crazy.

[Beefy22]

Oh BTW, it's not programs that initially stress CPUs beyond spec that determines stability. It''s the applications and games you use that will determine stability. As that is (or should be) the focus of why you are trying to OC to begin with. And, those everyday applications and games are usually the real reason why you decided to overclock to begin with .

IMO your interpretation is completely off(not trying to be arse or troll either). A stress test/program often times represents the max theoretical load any given component will be subjected to...

EX) I could game with my 8800GTS @ 837mhz core/2106mhz shader all day long(assuming I wished to subject myself to 100% fan speed) but could only fold @ 821 mhz core/2052mhz shader. If I had sold my card claiming 837/2106 stable to someone who wanted to fold that would be a lie, period. Which is why you test for stability with highest possible load scenario, you can only say 100% stable if it can pass a ~100% load scenario.

[Eastcoasthandle]

IMO your interpretation is completely off(not trying to be arse or troll either). A stress test/program often times represents the max theoretical load any given component will be subjected to...

EX) I could game with my 8800GTS @ 837mhz core/2106mhz shader all day long(assuming I wished to subject myself to 100% fan speed) but could only fold @ 821 mhz core/2052mhz shader. If I had sold my card claiming 837/2106 stable to someone who wanted to fold that would be a lie, period. Which is why you test for stability with highest possible load scenario, you can only say 100% stable if it can pass a ~100% load scenario.

Why is it off? You really didn't explain so in this post. If the actually intent is to validate your OC for games or everyday applications (for example) then any additional load created by another program is invalid.

100% stable isn't define by exceeding the load of the device in question. It's the ability to run your programs and games without issue. This is were we disagree. This is why I don't see the relation between posts.

I was not aware that the 4870x2 would consume 400 watts under a heavy load at stock clocks. So overvolted and overclocked is going to be pumping some seriously insane levels of heat. One could likely near 500w of draw with a modded card, and that is getting a little crazy.

Maybe I don't understand. However, I don't see the reasoning of this post to the OP nor why you quoted my post with this response. Is there a game that draws 400 watts from this GPU under load?

[Solus Corvus]

I don't see what the problem is. I have run a renamed furmark for hours on the stock cooler and now with water cooling and have never had a problem with my 4870x2. It's a great program for testing overclocks though maybe a little overkill if you only care about game stability.

Sure, while it is running furmark it is probably drawing more power then specification, but my power supply can handle it no problem. In spec or out of spec doesn't matter to me much as long as it's stable.

[SNiiPE_DoGG]

If folding is the load you are overclocking for then you need to be stable for that, if your gaming is the load you put on your GPU then you overclock for that. its very simple.

[Eastcoasthandle]

If folding is the load you are overclocking for then you need to be stable for that, if your gaming is the load you put on your GPU then you overclock for that. its very simple.

Agreed, you base your stability on the applications and games you use daily. Stress test/burn test applications don't necessitate stability for other applications and programs if that amount of load is not required.

[Beefy22]

Why is it off? You really didn't explain so in this post. If the actually intent is to validate your OC for games or everyday applications (for example) then any additional load created by the program is invalid.

100% stable isn't define by exceeding the load of the device in question. It's the ability to run your programs and games without issue. This is were we disagree. This is why I don't see the relation between posts.

Semantics... We disagree and that is how the posts relate.

I guess I could be persuaded to see your side if it was phrased as:

100% gaming/app stable and then listed the games/apps, because there is no guarantee if it's unstable under max load(stress test) that it will be stable on all games/apps. Your definition of 100% stable is not an absolute, therefore stability is relative and it's interpretation subjective.

[Eastcoasthandle]

Semantics... We disagree and that is how the posts relate.

I guess I could be persuaded to see your side if it was phrased as:

100% gaming/app stable and then listed the games/apps, because there is no guarantee if it's unstable under max load(stress test) that it will be stable on all games/apps. Your definition of 100% stable is not an absolute, therefore stability is relative and it's interpretation subjective.

No, it's not semantics. Semantics is arguing about the wording of my context when what's implied is the same.

In the end, stability should never be based on just stress test/burn test applications that can place a load on your PC component higher then what's normally used in the daily applications and games you use. It should be based on those applications and games.

[SNiiPE_DoGG]

you dont see extreme overclockers saying "oh, well that 6ghz cpu clock isn't 3d stable so its not a valid superPI score" thats ridiculous, you clock for your application, not the most extreme one.

[Beefy22]

No, it's not semantics. Semantics is arguing about the wording of my context when what's implied is the same.

Not necessarily, semantics is a broad field.

In the end, stability should never be based on just stress test/burn test applications that can place a load on your PC component higher then what's normally used in the daily applications and games you use. It should be based on those applications and games.

^ I agree, I just feel a distinction is warranted.

-EDIT-

you dont see extreme overclockers saying "oh, well that 6ghz cpu clock isn't 3d stable so its not a valid superPI score" thats ridiculous, you clock for your application, not the most extreme one.

Irrelevant(straw man), they've never claimed stability.

[Eastcoasthandle]

Not necessarily, semantics is a broad field.

Based on the example given it's no longer a broad field but a very specific on. However, to each his own, what's implied remains the same.

[Kingcarcas]

Exactly why i won't buy one of those cards, though the GTX260 doesn't do too bad. Lets hope 40nm cards can do better.

[LordEC911]

40nm should help power consumption but AMD & Nvidia, and now Intel, are always fitting the most performance possible into a certain TDP. Unfortunately, that won't be changing for quite awhile here.

[Manicdan]

does no one remember ATI modifying the drivers to not run at maximum for FuRMark? They knew it would run at a much higher power than what it was built to do. and even if it wasnt going to hurt the video card, they dont want to hear stories about people blowing up a 500W PSU cause they said it would be enough for their products. they have a list of certified PSUs, imagine if that list meant nothing.

they constantly adjust their drivers, and i bet one of the reasons is to ensure they stay in their power profile. running furmark as a different program is no different then abusing the product, and if something breaks, they wont feel sorry.

as interesting as this article is, its irrelevant if they do anything that 90% of the population woulnt do, but honestly since this is XS, it is pretty cool to see such high power consumption. now id like to see how hot the parts get, anyone got a thermal camera?

[STEvil]

Nah, that car analogy is spot on. However, it depends on your understanding of it. What's being implied here is what the device's intended and expected use is (IE: the intended or expected use of a corvette). Once you understand that should it be clear why it was used in that context.

Edit:
Oh BTW, it's not programs that initially stress CPUs beyond spec that determines stability. It''s the applications and games you use that will determine stability. As that is (or should be) the focus of why you are trying to OC to begin with. And, those everyday applications and games are usually the real reason why you decided to overclock to begin with .

Nah, the car analogy has its flaws. It has a point as well, but the car implied by the analogy is a commuter vehicle it seems, not anything meant to tow (or the ability).

A small truck vs large truck towing capacity would make for a better analogy. Both were bought with the capability to tow, but one is meant to tow a heavier load. Both can tow a heavy load but only one can do it "legally" or without sustaining damage from the work load.

That said, when I buy a piece of hardware I expect it to perform to its full advertised level when I want and for the warrantied time period. CPU's and GPU's are not sold as performing at 80% (number from the air) of their maximum performance... in fact there is no documentation provided which states that running your CPU/GPU at maximum performance could damage it and no software provided to tell you if you are going over the limit..

I dont disagree that furmark is a torture test that pushes things to their limit, but I believe it is a disservice to advertise a performance product as such if it is not capable of running at peak performance for its expected life cycle (realistically there will be down/idle time). Furmark is not the only application which can stress a GPU to full or very near full capacity.

[Eastcoasthandle]

Nah, the car analogy has its flaws. It has a point as well, but the car implied by the analogy is a commuter vehicle it seems, not anything meant to tow (or the ability).

A small truck vs large truck towing capacity would make for a better analogy. Both were bought with the capability to tow, but one is meant to tow a heavier load. Both can tow a heavy load but only one can do it "legally" or without sustaining damage from the work load.

That said, when I buy a piece of hardware I expect it to perform to its full advertised level when I want and for the warrantied time period. CPU's and GPU's are not sold as performing at 80% (number from the air) of their maximum performance... in fact there is no documentation provided which states that running your CPU/GPU at maximum performance could damage it and no software provided to tell you if you are going over the limit..

I dont disagree that furmark is a torture test that pushes things to their limit, but I believe it is a disservice to advertise a performance product as such if it is not capable of running at peak performance for its expected life cycle (realistically there will be down/idle time). Furmark is not the only application which can stress a GPU to full or very near full capacity.

We can only agree to disagree then. I still believe that the car analogy is spot on. The example I provided on top of the car analogy really drove home the point. I simply can't justify the use of a torture test program as a means to determine stability for games/programs that don't require such loads.

[saaya]

i dont know, those numbers dont make sense... compare them to whole system power consumption meassurements of several other sites and youll see it doesnt add up...
those numbers are too high, i find it hard to believe them... they must have gotten something wrong...

[SKYMTL]

Perhaps their clamp meter wasn't calibrated or not calibrated properly. Actually it makes me wonder if they were using an AC (incorrect) or DC (correct) or dual mode clamp meter and how they converted the reading (mA or A) into Watts. That would explain the extremely high readings they received from the PCI-E power cable.

[Barrok]

We can only agree to disagree then. I still believe that the car analogy is spot on. The example I provided on top of the car analogy really drove home the point. I simply can't justify the use of a torture test program as a means to determine stability for games/programs that don't require such loads.

See, that is the fault with your analogy. Truck companies TEST their trucks for maximum payload. Who cares if 98% of the people never reach that payload, or if the "regular" payload is alot less. The fact is the truck companies give you the maximum payload and the truck is 100% capable of that. Graphics card companies, according to this article, seem to give you a maximum wattage but not when the card is being pushed at 100%.

[STEvil]

And if a game came out that ran your GPU to 100% load you'd be making your GPU work too hard.

Maybe thats why Crysis plays like crap, it has a cap.


SKYMTL - makes sense. Could probably get a PCI-E riser card and cut the power supply lines with a dremel or something then solder in some wires in a loop to stick the clamp through, might get more reliable readings.

[Kylian]

After many days of registration-process, i finally got it to post here . Firstly i want to say, that due to the many request of non-german-readers, we decided to translate our article on graphics card power consumption into english. I think this will be interessting for some people here:

Power consumption of current graphic cards (English Version)

At this point i don´t want to go to deep into the discussion about FurMark. I think our viewpoint get´s pretty clear if you read the whole linked article. We know about the fact, that FurMark is a worst-case-scenario, but we treat the tool also as such (you´ll get what i mean, if you read the article).

Perhaps their clamp meter wasn't calibrated or not calibrated properly. Actually it makes me wonder if they were using an AC (incorrect) or DC (correct) or dual mode clamp meter and how they converted the reading (mA or A) into Watts. That would explain the extremely high readings they received from the PCI-E power cable.

Sorry, but that's not the proper way to go about things. I can´t blame you for not reading our test methods, because you propably don´t understand german language (Now you can, because we translated it ). But nevertheless throwing arround assumptions without any justifaction that are simply not true just to make our results to look wrong isn´t the nice way . Just as editor yourself (i assume your writing for Hardware Cannucks) you should know, that such a behaviour can be really frustrating for an author.

Quote from our article, that hopefully will clarify things:

The Temna-device is a True-RMS-clamp-meter. This is necessary because the graphics card isn´t purely ohmic but a reactance with transients. This current clamp was calibrated using a Fluke 9100 Universal Calibration System (Standard calibration like described in DIN EN ISO 9000 / 9001 referring to DIN EN ISO/ IEC 17025 (ISO Guide 25). Measuring Tolerance: up to 30A: +-0,08% +2mA, up to 105A: +-0,08% +11mA.

http://ht4u.net/reviews/2009/power_c...ics/index5.php
You see, our meassuring-devices are calibrated well. And as you can even see in the picture, we used the DC-Mode of our current-clamp.

i dont know, those numbers dont make sense... compare them to whole system power consumption meassurements of several other sites and youll see it doesnt add up...
those numbers are too high, i find it hard to believe them... they must have gotten something wrong...

One more time, our load-results are that high because we used FurMark as load. Have look at this comparison of tools. As you can see FurMark is much more demanding than 3DMark06. I think these 3DMark06-results should be pretty much what you expected, right? But you can also see, that there are many games that are more demanding than 3DMark06, not only FurMark.

as interesting as this article is, its irrelevant if they do anything that 90% of the population woulnt do, but honestly since this is XS, it is pretty cool to see such high power consumption. now id like to see how hot the parts get, anyone got a thermal camera?

We have one :P.
http://ht4u.net/reviews/2008/messtechnik/index8.php
And we are currently working on an separat article that will deal with the temperatures of the 4870 (and the high voltage regulator temperature). And yes, we´ll do some pictures with our thermal camera .

Greetings,
Leander [HT4U]

[STEvil]

Thanks for dropping in and explaining a bit Kylian. I dont think SKYMTL was trying to be rude in how he phrased what he said, just a bit direct maybe