Printable View
Sources close to engineers at EVGA complained that VRMs used in the GTX 590 were not adequate, and from what I understand Nvidia chose to use driver level power restriction/throttling instead of a more expensive solution.Quote:
Originally Posted by Aten-Ra
It seems it was a decision made to compete with the 6990 on price and they believed the driver would be sufficient enough to protect the card.
Incorrect on all points. You can't use the same fan on a dual PCB card as you would on a single PCB. The heatsink design would also be different due to 2 PCB's. With the added space of a dual PCB solution, more phases of VRM per GPU would be supplying the same power as the phases do now and therefore would not be working as hard and would produce less heat and be more reliable as a result.Quote:
Originally Posted by bamtan2
As it stands now it looks like the VRM is basically at its limit at stock clocks and voltage.
That's just plain assumption right there. "Sources close to..." so what's that? Parents and siblings or just some made-up "source" that's being used to justify a forum post?Quote:
Originally Posted by Stukov
The amount of misinformation here is about on par for an NVIDIA thread but some of these posts really take the cake...
And yet the card can OV at 1.05V and OC at 900MHz (On air)Quote:
Originally Posted by CryptiK
http://lab501.ro/placi-video/nvidia-...overclocking/7
The VRM implementation of GTX590 has a limit of ~1.05Volts (AIR) and it certainly is not at its limits at default settings of 0.932V
Yeh, one miracle card. Does anyone know if it's even still alive?Quote:
Originally Posted by Aten-Ra
I know you like Nvidia and all, but surely with Nvidia lowering default volts and locking out any voltage adjustment, and with OCP killing overclocking performance by down clocking the card, you must be able to see there is an inherant weakness in some of the cards components, for Nvidia to take such drastic action to limit power consumption!... :confused:Quote:
Originally Posted by SKYMTL
He's saying that there should not be a thermal wall. Those cards are hitting a thermal wall.Quote:
Originally Posted by STEvil
Speak with chipsy on Guru3d forums.Quote:
Originally Posted by SKYMTL
This has nothing to do with liking or not liking NVIDIA. It has to do with going against all the BS that's starting to pop up yet again regardless of how much flak I take.Quote:
Originally Posted by Sherman Tank
Why must there be an inherent weakness? Under volting has been done with literally EVERY dual GPU card. As for the OCP being implemented, AMD does the same thing or do you not remember a certain HD 6970 downclocking in Metro 2033 (and likely other scenarios) at reference clocks? So I am guessing there is some weakness there too? Personally, I don't think so in either case.
What I think is that the OCP is in place to control power consumption not due to PSU insufficiency but rather to avoid drawing too much power through the PCI-E slot. We have all seen what can happen when GPUs draw excess PCI-E power on boards that aren't able to handle it.
I mean sure, the components on any given GPU can implode if pushed too hard but that doesn't mean they were under designed. It just means they are being pushed far beyond their reference spec by overclocking, over voltage, etc. There is just no way either AMD or NVIDIA can design for every eventuality.
On the flip side of the coin NVIDIA can be blamed to a certain extent since the OCP wasn't implemented in their early driver revisions for the GTX 590. That IMO was a huge mistake and did give them a black eye on this launch. I do however find it ironic that people are now complaining that the OCP is actually working the way it should. ;)
I'd argue about the "working" part. If it was really working as it should, then you should be able to overvolt how high your heart desires and then get throttled to the bone, get "driver stopped working", blue screen, or whatever, just not the magical smoke. Instead, you have your voltage locked. And with stock clocks, you cannot really overclock it that high, for the OCP to start working (in games, in furmark it will engage with stock clocks too).Quote:
Originally Posted by SKYMTL
Sticking with older non-voltage-locked drivers isnt' an option, since the newest drivers introduce multithreaded rendering for all games. What happens without it, see DA2 for an example.
Maybe nVidia can sort it out, make the OCP work better and release the voltage from it's lock. But with a software sollution, I wouldn't be holding my breath for it...
Nope, they are not hitting a thermal wall, in particular the 590.Quote:
Originally Posted by cegras
Ultimately the more power a card uses the better cooling it needs, if cards continue to get hotter, triple slot coolers may become the norm.
How ever, I guess you could say current 'dual' GPU cards are beginning to approach the thermal limits of dual slot coolers.
No, both of you said the VRM was weak.Quote:
Originally Posted by cegras
Cooling the VRM will result in it not dieing as easily. The PCB of the cards will be cooled by the LN2 (not only the GPU's get cold) and will allow the VRM's to perform FAR beyond official specifications.Quote:
Originally Posted by Aten-Ra
Because Nvidia took very bold steps to greatly reduce the overclockability of the 590 after reports of premature deaths. The only reason Nvidia would do this, is if there was actually a serious issue with the card.Quote:
Originally Posted by SKYMTL
Can you not adjust powertune to +20 to avoid this? Also, if PT wasn't there, do you really think 6970's would be blowing up?Quote:
Originally Posted by SKYMTL
Wrong, it was first introduced for Fermi to simply keep cards in spec and avoid negative press reg. power consumption, especially in Furmark.Quote:
Originally Posted by SKYMTL
It obviously came in very handy to stop 590's from blowing up also.
IMO, the card is under designed if it blows up at stock or with mild overclocks, also note it's not actually the GPU's themselves that are blowing up.Quote:
Originally Posted by SKYMTL
The fact the card has now been artificially made unclockable just adds to the disappointment and frustration that Nvidia didn't build this cards with high enough quality components to allow some headroom (which we have now all come to expect), I think we all know that if Nvidia hadn't done this, the 590 could have actually been the 'fastest card in the world'.
IMO, Nvidia are already pushing their card much further than what the reference spec 'should be', it's like they'v left ZERO safety margin, which is never a good idea, and the only reason I can think of why they'd done this, was because the 6990 caught them off guard.Quote:
Originally Posted by SKYMTL
The thing that are holding these cards back is the inadequate cooling. What tests have shown is that the fermi architecture consumes alot less power the cooler they run. Just a couple of things would have made these cards much better, the biggest i think is a better cooler with a larger vapor chamber or best yet, a huge heatpipe/ large fan solution(and a more breathable slot plate). These things should be well under 80 degree under load if given a good cooler. These things are about 170 mhz clock slower than a gtx 580 which are capable of the mid 80's. These sometimes go into the mid 90 range and I could easily see this adding 50 watts of power to the power consumption. Considering the low clocks, these things should produce less heat and use less energy. I think the high heat is also a result of the overly packed PCB, it has so much on it and the so much heat is radiating from the card over such a small surface area. It makes it difficult to cool.
I think heat is the biggest problem and the less than desirable cooling solution(cards with a die this big should not have an 11" cooler). This card should have no problem hitting gtx 580 clocks considering the regular gtx 580 regularly goes past 900.
NV needs to byte the bullet, hell make a triple slot if they have too. It is a shame that these cards don't destroy a gtx 560 SLI performance when the chip size is so much bigger.
^^^
Heat isn't the issue with the 590, although yes it's true, Fermi does use more power the hotter it get's, but the 590 is already pretty cool, a better, safer and simpler solution would of been to design the card correctly to begin with.
i.e. Add more or better power circuitry to reduce the load on each individual component.
Maybe its time for nVidia to consider a newer revision if they really want to have the fastest graphics card title - for real this time. They've revised the 295 to begin w/.
Now that we all know, no one really cares(you rly?) about 590 being shorter than 6990. Put in some quality components on an uncramped pcb;
notch the def. clocks up a bit and never worry about the power consumption; coz ppl buying them knows what a dual gpu is.:idea::D
Its hardly cool though. Most reviews have this close to 90 or past it. Running this thing cooler would have so many positive effects I am almost certain the vrm issues would go away with the reduced power consumption(them running cooler would also help them from popping). Almost as importantly, a better cooling solution would allow them to run higher clocks. This cards is slower than the 6990. Not by a huge amount but it is there and enough that AMD can take corny jabs at them.Quote:
Originally Posted by Sherman Tank
I agree better circuitry could be in order as I have said in one of my other posts, better cooling would just add to this effect.
but running cool would be loud :rolleyes:
SKYMTL - OCP does not have anything to do with power from the PCI-E slot itself.
Yes i know some guys trying hard..... but both GTX590 and HD6990 can OC without OV.
Have a look at this review with GTX590 @ 675MHz default Voltage vs HD6990 @ 900MHz default Voltage.
http://lab501.ro/placi-video/asus-ge...-of-the-titans
I don't think it has to be loud. I think I know what your implying with NV and their marketing. I think they just need to abandon this vapour cooler design and start making sinks like aftermarket ones. Although they might need to go triple slot for cooling with these kinds of setups, It would be totally worth it because you could build something with temps that the best two slot coolers get on a single GPU. I also think they should extend the heatsinks upward like how MSI does it to increase surface area. I think a dual fan would help.Quote:
Originally Posted by STEvil
The gtx 580 can be cooled properly is capable of amazing clocks.
http://tbreak.com/tech/2011/03/msi-n...ning-review/3/
The temperature at load is 55C and this at 832mhz clocks. It overclocks to 982 in this review. Compare this to the vapor chamber stuff and you will see that the stock cooling isn't that good. The lightning although not quiet, is not freakishly loud. Loud as a midrange card really.
http://www.techpowerup.com/reviews/M...htning/22.html
I think Nvidia really cheaped out on the design and the reduced vapor chamber size is already showing this.
If this card was an extra inch, increased surface area, this card could have been alot faster.
675 seems (is) nothing when you know what's under the bonnet, and what it's capable of.Quote:
Originally Posted by Aten-Ra
With my waterblocks installed, I was able to push my 6990+6970 to 1.25v, 1050/1350. No black smoke, nice and cool, whisper quiet. Everything is working fine. No throttling... Idle temp 26 celsius, max Furmark load temp 43-45 celsius. :)
You can highly OC the 6990, just like two highly OCed 6970 in Crossfire with a waterblock. But you will never be able to push the 590 to 580 SLI OC level, since you can't OC the 590 at all. :)
The OCP limits the amount of current a given design can draw. It doesn't have to directly affect the power drawn from the PCI-E slot but it has a trickle-down effect to this area. If there is no limit to the amount of current, the excess power draw has to be received from somewhere.Quote:
Originally Posted by STEvil
Most designs will first max out the PCI-E power connector draw and then fall back on the PCI-E slot's available power. Very few (if any) cards will need to fall back on that since there is an overhead built into the power connector design (ie: if a card draws 300W, an 8-pin and a 6-pin will be used, giving a buffer zone of 75W before the PCI-E slot's capacity is factored upon).
However, if the OCP is too high or not in place and the two built in connectors (in this case two 8-pins for a total of 300W) can't handle the excess draw, the PCI-E slot will be called upon to supply the remainder. Unfortunately, while the slot is RATED to supply up to 150W, many boards just can't support that since they don't have excess capacity for +12V power through the board itself. Hence why we have seen burnt power connectors, etc in some cases. So, while OCP many not directly affect the PCI-E slot, it eventually does in certain scenarios.
PCI-E slot power is usually only linked to a set component on the card which will keep it at or under PCI-E spec which is 75 watts for the slot itself. The power regulation circuity of the card should not have to worry about load balancing between the power connectors and the pci-e slot.
Actually, they took said steps BEFORE since the default voltage on the cores is very low for a Fermi-based card. So the limitation was always there. :up:Quote:
Originally Posted by Sherman Tank
You hit the nail on the head. AMD allows adjustment of their software. NVIDIA doesn't and yes, that's their mistake. As for HD 6970s blowing up, I can't really comment since AMD's utility worked from day 1 while NVIDIA's didn't so the cases aren't really the same.Quote:
Can you not adjust powertune to +20 to avoid this? Also, if PT wasn't there, do you really think 6970's would be blowing up?
The implementation is fundamentally different this time around. Instead of being just app detection, it is now hardware based as well in order to limit current before the card exceeds a certain threshold.Quote:
Wrong, it was first introduced for Fermi to simply keep cards in spec and avoid negative press reg. power consumption, especially in Furmark.
I see no indication that the cards are blowing up a stock or with minor overclocks. One POSSIBLE report of a stock card going off means nothing since we have no idea the circumstances or if the individual concerned even left it at stock.Quote:
IMO, the card is under designed if it blows up at stock or with mild overclocks, also note it's not actually the GPU's themselves that are blowing up.
Overclocking is a value-added feature but is never guaranteed. I know people love to do it but supporting it past a certain extent isn't in NVIDIA's best interest. Then again, you forget that the card CAN be overclocked but is highly limited by its default voltage; a voltage which was put in place to limit power consumption.Quote:
The fact the card has now been artificially made unclockable just adds to the disappointment and frustration that Nvidia didn't build this cards with high enough quality components to allow some headroom (which we have now all come to expect), I think we all know that if Nvidia hadn't done this, the 590 could have actually been the 'fastest card in the world'.
Assumptions. Not much more to say other than that. :up:Quote:
IMO, Nvidia are already pushing their card much further than what the reference spec 'should be', it's like they'v left ZERO safety margin, which is never a good idea, and the only reason I can think of why they'd done this, was because the 6990 caught them off guard.