Can I get away with GTX 590 quad SLI with 850W PSU?

[blibba]

That's my preference. I know people who've fried hardware because the PSU couldn't maintain +12v and some hardware ended up getting fried by virtue of running more current because of the reduced voltage and for anyone who knows Ohm's law, heat is proportional to current, not voltage. Personally I like to have some wiggle room with the PSU. Consider it provides power to every component, it's best not to take chances IMHO.

This will never happen on a PSU with OCP unless it's completely fucked.

 

[qubit]

I'd run one, measure 2D idle and 3D load with a power meter at the wall and find out the difference, and if you have that much to spare - THEN try both.

That said, i highly doubt you will have that much wattage left over so i voted no.

That's my preference. I know people who've fried hardware because the PSU couldn't maintain +12v and some hardware ended up getting fried by virtue of running more current because of the reduced voltage and for anyone who knows Ohm's law, heat is proportional to current, not voltage. Personally I like to have some wiggle room with the PSU. Consider it provides power to every component, it's best not to take chances IMHO.

Indeed people, I didn't think it would be safe to try 3D mode with both cards on my old PSU, but I thought I'd ask anyway, in case I did have a little wiggle room. I certainly won't be trying it given your collective responses though. 2D should be ok though and I'm willing to give it a go.

 

[Aquinus]

This will never happen on a PSU with OCP unless it's completely fucked.

I've seen some pretty weird things to computers. Even still, I still like to stick on the side of caution. It's one thing if we're talking about approaching what the PSU can do. My concern is that it will greatly over-draw and I wouldn't want to find out weather or not the OCP works or not, but that's me.

Also, don't PC's run full tilt for a second or two when they're first turned on? If OCP is working, just turning the machine on could trip it, but once again, I wouldn't want to find out.

 

[qubit]

Also, don't PC's run full tilt for a second or two when they're first turned on? If OCP is working, just turning the machine on could trip it, but once again, I wouldn't want to find out.

No, I'm pretty sure the GPUs don't start up at max power, especially big powerful ones and I'm confident enough of this to power up these cards in 2D mode only on my 850W PSU. The fan spinning up for a moment on some cards is separate and perhaps what's making you think this?

 

[Mussels]

No, I'm pretty sure the GPUs don't start up at max power, especially big powerful ones and I'm confident enough of this to power up these cards in 2D mode only on my 850W PSU. The fan spinning up for a moment on some cards is separate and perhaps what's making you think this?

having tested crossfire setups at the wall, the BIOS/boot stage definitely uses a lot more power than at idle - but less than full load.

 

[qubit]

having tested crossfire setups at the wall, the BIOS/boot stage definitely uses a lot more power than at idle - but less than full load.

Yes indeed, I think some of the power save features are turned off during that time.

 

[qubit]

Well, I have a surprising result people. And no, it's not that I bought TWO Zotac GTX 590s off that eBay seller so I now have three* GTX 590s... it's that I can run two of them in SLI 3D with zero problems on my old 850W PSU.

@newtekie1 Buddy, you might want to take note of the following. It surprised me too and I appreciate your cautious advice previously. Much better to miss out on some fun than blow up the PC.

It looks like those kilowatt power usages quoted by the enthusiast websites might be higher because they have other hardware connected to the PC, perhaps. All I have is a sound card, 2 HDDs and an SSD, which isn't a lot.

Anyway, I know it's ok, because I've got a power meter connected to my PC at the wall. With one card running flat out with Furmark (vsync off) the PC drew around 450-500W. With two cards at the desktop, the PC draws around 290W.

I then ran UT2004 and got power usage of around 650W with vsync off (ie more power used). Yes, it still runs all jerky with SLI, but that's not relevant here.

I then ran Furmark with both cards flat out and I got readings of 730-780W at the wall. This means that the PSU isn't maxed out at any point. Note that with Furmark, the GTX 590 throttles severely to 304MHz, that's half its clock speed, while TDP shows as 99% in MSI Afterburner. I'd be real pissed if a played a game that triggered such severe throttling as the frame rate is about half too.

With quad SLI, the NVIDIA driver gives you a lot of SLI options, so I made sure to select the 4 GPU mode for my 3D tests above, to maximise the power usage.

​

Note that I wouldn't run the system long term like this, as the PSU is outputting fairly close to its maximum and it's old, plus I think NT's advice to get a mobo with a PCI-E auxiliary connector for current carrying capacity is a good idea for long term use.

So there you have it, all that is left to do now is to run some extensive benchmarks comparing GTX 590 SLI with other graphics configurations and create a TPU thread on it. Watch this space.

*I'm nuts, yes I know, thanks. Likely I'll sell that EVGA Classified one and get more for it than the second Zotac card I bought for £110.

 

[newtekie1]

Note that with Furmark, the GTX 590 throttles severely to 304MHz, that's half its clock speed, while TDP shows as 99% in MSI Afterburner.

Isn't Furmark one of those trigger programs that cause the Fermi cards to instant throttle no matter what? Thought that was why W1z stopped using it for maximum power draw numbers in his reviews.

Edit: Just checked his 590 review. Yep, he gets a higher power draw running 3DMark than he does running Furmark and mentions the power limiting features of the 590 causing that.

 

[qubit]

Yes, it is, but the TDP was showing as 99%, so I figured the cards were sucking up maximum current anyway. I've got the latest 3DMark, so I'll give that a go. I have a feeling that this and also high video resolutions could push the power usage up a fair bit, so I'll have to carefully watch that power meter. I've only had a go at 1080p so far. I'll try 4K DSR now with Furmark and let you know if I blow up my rig, lol.

Note that I didn't know Furmark throttling would be anywhere near that severe though.

EDIT: Running Furmark at 4K actually consumes less power than at 2K, around 680-710W, likely due to the tanking of the framerate to around 16fps. I will try 3DMark and let you know how much power it uses, but not tonight.

 

[Champ]

I've read that psu is underratedly strong and can handle a 1kw+ load. I OCed my 290s and 4770k to the brink with no hiccups

 

[Norton]

@fullinfusion tried a similar load using 7970's (and R9 290's iirc) with his HX850 (see below)

I pulled over a KW from the wall with my 7970's on a corsair HX 850 and it died a slow death shortly after. Thus why I went to a G2 1300 PSU and never turned back

I wouldn't risk killing stuff and would suggest waiting on your 1200w

 

[qubit]

I've read that psu is underratedly strong and can handle a 1kw+ load. I OCed my 290s and 4770k to the brink with no hiccups

You're right, the reviews said this. That was one of the big reasons that I got it and it proved to be a superb buy.

@fullinfusion tried a similar load using 7970's (and R9 290's iirc) with his HX850 (see below)

I wouldn't risk killing stuff and would suggest waiting on your 1200w

Indeed, fullinfusion was overpowering his PSU which caused it to slowly degrade and die like that. However, the wattage at the wall for me tops out at 780W and only for brief moments, which is rather less than it's rated output to the PC, so it definitely not being overstressed. I'm guessing that at its rated maximum, it would be pulling 950-1000W at the wall.

Again, as I said before, if I intended this to be a long term installation, I would still upgrade to a more powerful PSU and better mobo with aux PCI-E connector for more headroom and reliability.

btw, I got quite a hit of nerdy happiness at seeing 4 GPUs in the driver and running these cards together tonight.

 

[newtekie1]

Yes, it is, but the TDP was showing as 99%, so I figured the cards were sucking up maximum current anyway.

Nope, the driver just reports max TDP when furmark is activated to trigger the throttling.

 

[qubit]

Saw a couple of glitches tonight: all the GPUs are stuck at the 3D clock of 607MHz in the MSI Afterburner window (608MHz in the system tray display for Afterburner for some reason). Disabling SLI and rebooting helped, but now GPU 2 (top card) is stuck at 608MHz according to Afterburner, with the others at 51MHz. I can hear the fan spinning fast and it's running hot at 65C, which is annoying. This is without anything to trigger a 3D clock as far as I can tell.

Also, the PC locked up solid about an hour ago while writing a forum post. Pressing reset did allow the mobo to reset properly though (this one doesn't always).

The PC has been stable when running games so far.

 

[qu4k3r]

I think the pc may boot and log on desktop, but when the gtx590's reach some load the psu will shut down if its protections work properly.

According to the following...
http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-590/specifications
http://www.geforce.com/Active/en_US/en_US/pdf/PSU.pdf
For gtx590 quad sli system they suggest psu's over 1200w.

Try that with a 850w psu it's like playing russian roulette with (n-1) chambers loaded.

 

[qubit]

Nope, the driver just reports max TDP when furmark is activated to trigger the throttling.

Forgot to mention: I also ran CoD: Advanced Warfare and Far Cry 4 and got similar wattage readings, so even though Furmark is treated as a special case for card protection by the driver I do think the card is still being maxed out with the TDP reading I saw, or nearly so. Also, it wasn't some fixed fixed value in Afterburner, but wobbling around, so even if that figure has been massaged a bit by the driver, it's not meaningless.

I've still got to run 3DMark to compare with Furmark and will get back to you. Now I'm late for work!

I think the pc may boot and log on desktop, but when the gtx590's reach some load the psu will shut down if its protections work properly.

According to the following...
http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-590/specifications
http://www.geforce.com/Active/en_US/en_US/pdf/PSU.pdf
For gtx590 quad sli system they suggest psu's over 1200w.

Try that with a 850w psu it's like playing russian roulette with (n-1) chambers loaded.

Sure, 1200W is better and recommended for the long term, but did you see my previous posts about power usage? Specifically post 32 onwards where I used the wattmeter to check on overall power consumption. In short, my 850W isn't being overloaded, which was a surprise to me as much as everyone else.

If anything is gonna get hurt it will be my wallet when I come to pay the electricity bill.

 

[blibba]

...which was a surprise to me as much as everyone else.

Not quite everyone, qubit

HH reported 824W peak with a 990X and two velociraptors: http://www.hardwareheaven.com/conte...tac-nvidia-geforce-gtx-590-quad-sli-review/17

I think you'll be fine as long as you don't use Furmark etc. Those Corsair PSUs will cut out before anything is damaged - they won't restrict power to the cards or deliver unsafe voltage or anything like that.

 

[BiggieShady]

... and for anyone who knows Ohm's law, heat is proportional to current, not voltage.

Actually heat is proportional to power and power is voltage multiplied with current, so heat is proportional to both current and voltage. Misconception comes from perfect resistors where power is proportional to square of the current, but the current always changes with the voltage.

 

[Aquinus]

Actually heat is proportional to power and power is voltage multiplied with current, so heat is proportional to both current and voltage. Misconception comes from perfect resistors where power is proportional to square of the current, but the current always changes with the voltage.

Yes and no, you're describing the outcome of the voltage change, not what's physically happening inside. Heat is directly proportional to the square of the current.

Q ∝ I^2 · R · t

I think what you're describing is

P = V · I

Where P is power. The problem is that V is not the full-loop circuit voltage, it's the voltage drop across a segment of circuit which is caused by the resistance, R. When load is applied to a CPU, resistance changes because of gates opening and closing but, voltage alone doesn't determine heat and this is why:

Take Ohm's law...

I = V / R

It's a reformulation of Ohm's law and can be rewritten like so:

P = I · V = I^2 · R = V^2 / R

Where V is still the voltage drop, which keeps Ohm's law intact.

So voltage has no bearing itself on heat, however, a voltage increase may also result in increased current because resistance will remain relatively constant for any given set of instructions in the CPU (or impedance because we're not strictly talking DC (aka. clock signals, etc.) but I'm keeping it simple.) However, voltage itself is not responsible in any way for the creation of heat energy, it is the side effect of current increasing. However it's entirely possible to have two circuits, one with 100v EMF and and another with 10v EMF and both could generate the same amount of heat depending on the resistance of each circuit.

So it's not true that voltage is directly proportional to heat when current is. Voltage just tends to indirect impact current which results in increased heat energy that gets released from the circuit. It's a better statement to say, "In the case of CPUs, voltage is indirectly responsible for changes in current which directly impacts the release of heat energy." It's simply a causal relationship, not a direct one.

 

[newtekie1]

Forgot to mention: I also ran CoD: Advanced Warfare and Far Cry 4 and got similar wattage readings, so even though Furmark is treated as a special case for card protection by the driver I do think the card is still being maxed out with the TDP reading I saw, or nearly so. Also, it wasn't some fixed fixed value in Afterburner, but wobbling around, so even if that figure has been massaged a bit by the driver, it's not meaningless.

Sure, but Furmark should be giving you way higher readings, not readings similar to what you get in games.

 

[BiggieShady]

It's a better statement to say, "In the case of CPUs, voltage is indirectly responsible for changes in current which directly impacts the release of heat energy." It's simply a causal relationship, not a direct one.

Nicely said, it clarified your earlier post greatly and it is what I was trying to say ... I was under the impression you meant voltage change would not affect power