I’m hooking up a new rig. I have some GTX480s and an Antec TPQ 1200 PSU. The PSU comes with some nice PCIe cables, 8-pin connection for to the PSU and which then splits into an 8-pin and 6-pin connector for the GPU. In the little book for the GPU it says you should use two separate outputs from the PSU to power the GPU but it just seems so natural to use these nice cables that came with the PSU -why else would they design them like that?!?!?
Yes/No/guidance/opinions appreciated :)
Cannot really give a definite opinion since I am only blessed with one 470 and yes we used 2 6pin connectors branching out from the same wire outta 450W PSU. No problems yet :)
The 8 pin and six pin will work you should have two 8 pin & two 6 pin, the eight pin could also be used as four 6 pin in sli for two 470 for example…CHEERS
Thanks for the quick replies… I realised that the PSU promises to be able to power three GTX480s and the only way it can do that is if you use the splitter cables -so that’s what I’ve done. If anything catches fire or whatever I’ll be sure to post here again so no-one else makes the same mistake.
Cheers
An interesting related question… the Toughpower 1000W PSU has an explicit warning in the manual NOT to connect two PCIE power connectors to the same card if the connectors come from different 12V rails. It even puts a sticker onto each PCIe plug so you know what rail it is.
Why is mixing two rails a bad idea?
All I can think of is that if there’s a short in one rail, the shared connection through one card would then allow one rail to spill more current into the shorted rail, compounding the problem.
But that’s not a great explanation since the cards ALSO draw power through the PCIe socket, and therefore electrically connect to the motherboard’s 12V rail, and therefore all your PCIe rails end up being connected anyway.
What’s more confusing is that the PSU (which is excellent BTW) gives two hefty rails with 6 PCIe plugs. One rail has a 6 pin and 2 8 pin PCIe plugs, the other has 2 6 pin and 1 8 pin. If you have 3 GPUs (which I do) then you HAVE to mix rails onto one card. Why warn against a situation that your design forces the user to use?
So the machine is working fine but it makes me feel annoyed that I’m doing something the PSU manual clearly says “don’t do this!” and I don’t know why or what risk I’m running.
I sometimes get the feeling that the safety engineers and the main design engineers don’t talk as much as they should. Including power splitters while telling people not to use them is but one example… :)
(Of course, it’s probably more of a calculated risk. Giving the customer more options, even if they are slightly unsafe, probably leads to higher customer satisfaction overall.)
I emailed thermaltake about the mixed rails. I asked if it was a safety or reliability issue, and if I should avoid mixing the rails.
Their response:
That’s good news. It sounds like they’d rather have one rail delivering 200W than two rails delivering 100W each. For SLI gaming, the low power-per-rail would be common when one card is idle.
But for our CUDA rigs, we’ll be pulling watts from both rails (with 3 equally loaded GPUs) and it wouldn’t matter.
1200 for two GTX-480s? The peak of each GTX-480 is 600W so the peak for two is 1200W what about the MB, CPU, HDs, DVDs and BRs??? You need at least a 1500W to be close to safe.
Alex.
Not even close to correct. The current PCI-e standard only permits an absolute maximum of 300W per device, and NVIDIA’s own specifications page says the GTX480 is 250W TDP.
These are the CUDA forums, so the behavior of the GPUs is a lot more controlled than the overclocked graphics guys see.
CUDA programs tend to use significantly less wattage than graphics applications.
At this very moment, to my right hand side, I have an AMD hexacore box running 2 GTX295s and 1 GTX480 at full throttle. The Kill-a-watt meter hooked up to this box is currently reading 740 watts from the wall socket, so it’s likely only 650 real system watts. The 1000W Toughpower PSU powering it is in no danger.
If I fired up Furmark… maybe I’d pop some fuses… it’s unlikely the PSU could handle it. But with CUDA apps, I have a huge safety margin.
To my LEFT hand side, I have a box with 2 GTX480s and 1 GT240. The two GTX480s are also running full bore as I write this. I don’t have the power meter hooked up right now but from previous measurements that box is probably only drawing 700 watts. (It also has a overclocked hexacore 980x i7 chip though.)
Finally, the worst case power use of the GTX480 is 300 watts using furmark, not 600. The PCIe power specs even rated for more than 300 watts total (75+75+150).
So, you’re right that you need to worry about PSUs and overloading them. But this worry is managed by actually measuring how many watts you actually use and not just guessing.
Well I had in mind some reviews I read a while ago:
http://www.hardocp.com/article/2010/03/26/…80_sli_review/7
And indeed the CUDA applications need less wattage. Ok.
Where does it say that a GTX480 is a 600W part in that review? The results I see for the 480 show a complete overclocked Core i7 system with a GTX480 running Furmark and pulling 480W at the wall.
Also tell me exactly where is the 300W margin for PCI-e. Before setting the margin please refer us to the relevant source. Before doing this please a take a look at and then be careful before setting the marigins. : http://forums.adobe.com/message/2834332 and update your versions.
Indeed I also made a mistake I had in mind 600W but it was for the whole systtem. Sorry about this.
Alex.
The “margin” is defined in the PCI-e version 2.0 ECM specification, available here. And just to correct myself slightly, the standard says the limit is 6.25A from the motherboard slot, 6.25A from each 6 pin connector, and 12.5A from each 8 pin connector. The supply voltage is 12V, which means 75W for cards with no external power, 150W for cards with a single 6 pin power connection, 225W for cards with two 6 pin power connections or a single 8 pin connection (like a GTX275 or GTX470), 300W for a card with a 6pin and 8pin connector (like a GTX285, GTX295 or GTX480) and 375W for a card with two 8 pin connectors (of which I don’t believe there are currently any).
Have you read the last post of the link I told you. Have you read also that with a 1200W it is 87% efficient. Why do you pretend Sit to be such an expert and waste my time.
Whether you choose to believe the official PCI express SIG standards (which every manufacturer has to comply with if they want to sell their products as PCI express compliant) or some random guy on an Adobe forum is completely up to you. Even the review you linked to completely contradicts that posting.
What does that have to do with how much power a GTX480 uses?
Ok lets see what this “random” person as you say says, be careful though how you express opinions about people you don’t know, because right now you are calling him a irrelevant.
“In that case draw the 6 pin PCIe cable off V3 and the 8 pin off V4. Or at least draw them from different rails. Under load the GTX-480 can consume more than 420 W, or use more than 35 Amps.”
I will take a look at it also since I am not an expert on the power consuption issues and I will make a comment.
Regards,
Alexander.
Great I have sound some schematics because I was certain that this person is not “random”.
http://www.playtool.com/pages/psuconnector…ctors.html#eps8
8-pin
Pinout
Pins 1 through 4 Pins 5 through 8
Description Wire color Pin number Pin number Wire color Description
ground black 1 5 yellow +12 volts (12V1)
ground black 2 6 yellow +12 volts (12V1)
ground black 3 7 yellow +12 volts (12V1 or 12V2)
ground black 4 8 yellow +12 volts (12V1 or 12V2)
Connector part numbers
Motherboard connector Cable connector Terminals Maximum current per circuit
Molex 39-28-1083 Molex 39-01-2080 Molex 39-00-0168,
Molex 44476-1111 7 amps
Unofficial cable/connector maximum wattage delivery
Voltage rail Number of lines Maximum current Maximum wattage
+12 volts 4 28 amps 336 watts
6-pin
Pinout
Pins 1 through 3 Pins 4 through 6
Description Wire color Pin number Pin number Wire color Description
+12 volts yellow 1 4 black ground
+12 volts or not connected yellow or not connected 2 5 black ground
+12 volts yellow 3 6 black ground
Connector part numbers
Video card connector Cable connector Terminals Maximum current per circuit
Molex 45558-0002 Molex 45559-0002 Molex 39-00-0168,
Molex 44476-1111 8 amps
Official cable/connector maximum wattage delivery
Voltage rail Number of lines Maximum current Maximum wattage
+12 volts 3 2.083 amps 75 watts
So is it really each…? Physics.
By the way on the 8-pin it should go like 7A. This is called in Physics the Kirchoff law to comply with the 2.083. When two electic power in parallel meet their currents the output is their sum.
So what you need to contradict is 2.083A per lane (3 not 6) in 6-pin and 7A per lane (4 not 8) in 8-pin I will probably see your post at night (Greek time) I think I have wasted my time a lot with this issue. I just need you to put me correctly now the currents with the standard.
Best,
Alexander.
Those are the maximum electrical specifications of the molex connectors, which are mostly irrelevent. The parts you conveniently neglected to post from the page you linked to is the part that 100% confirms what both myself and Steve already said:
For the six pin plug:
Official cable/connector maximum wattage delivery
Voltage rail Number of lines Maximum current Maximum wattage
+12 volts 3 2.083 amps 75 watts
For the 8 pin plug:
Official cable/connector maximum wattage delivery
Voltage rail Number of lines Maximum current Maximum wattage
+12 volts 3 4.167 150 watts
Ie. for a GTX480 with one 6 pin and one 8 pin connect, the official specification limit is 6.25A@12V for the 6 pin plug (75W) and 12.5A@12V for the 8 pin plug (150W) = 225W from the PCI-e connectors, plus a maximum of 75W from the motherboard slot = 300W.