[Guide] Homebrew powered risers.
-
1. No need to cut the cable. Those wires are better to keep. Most of the current flows through the molex connector.
2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.
[quote]An 8 pin PCI-e connector is rated to supply 150w
A 6 pin PCI-e connector is rated to supply 75w[/quote]That’s only in theory. In practice, the difference is only one earth pin. The other earth pin added is used for cable detect only. Actual limit is 6A per pair of +12V and earth pins, that’s 18A or 216W per connector. Pay attention to cable length and thickness though. AWG20 is no good unless very short (1 foot max.), AWG18 is all right at up to 2 feet, AWG16 performs well at any reasonable length.
-
[quote]
1. No need to cut the cable. Those wires are better to keep. Most of the current flows through the molex connector.
[/quote]I disagree. If i only power 1 card with the riser, the other 3 cards can pull the power back down the ribbon cable, through the pci-e slot, across the board and in to the other cards. This has the potential of burning the track on the mobo and/or melting the ribbon/yellow cable. As i said in the original post it’s OK if you are using powered risers on ALL cards.
[quote]2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.[/quote]
Please elaborate why a cap would help? Isn’t the power from the PSU clean enough? The system will simply use the existing ground on the slot. Capacitors seem to blow once the power steps up which will happen if the wires aren’t cut.
[quote]That’s only in theory. In practice, the difference is only one earth pin. The other earth pin added is used for cable detect only. Actual limit is 6A per pair of +12V and earth pins, that’s 18A or 216W per connector. Pay attention to cable length and thickness though. AWG20 is no good unless very short (1 foot max.), AWG18 is all right at up to 2 feet, AWG16 performs well at any reasonable length.[/quote]
Agreed but the SOCKET on the cards seems to be limited and won’t actually pull more than spec. Guessing it’s to protect the VRMs but I have no concrete evidence though others seem to report it and it does explain a few things.
-
Check, check, check and thanks for guide and also the dedication=P
Your resonation regarding the things happening concur with what i have experienced.
3 mb ruined, 1 at 24pin socket (completly burned out 2 of the holes), and 2 cards that got burned to crisp below the pci-e area at bottom of card.
First one was with 6 cards mixed at once stock.
second times where 4x7950 (which worked like a charm), until i tweaked the gnu/memclock, then all hell broke lose=)Will start building powered extenders very soon and use it on ALL my connection, i take no more chances! its not broken hw that bothers me, its downtime on those gpu investments!. ordered me an extra mb now so i have one to spare=P
-
[quote author=Nutnut]
[quote author=ghostlander]
1. No need to cut the cable. Those wires are better to keep. Most of the current flows through the molex connector.
[/quote]I disagree. If i only power 1 card with the riser, the other 3 cards can pull the power back down the ribbon cable, through the pci-e slot, across the board and in to the other cards. This has the potential of burning the track on the mobo and/or melting the ribbon/yellow cable.[/quote]
Ohm’s law disagrees with you. Path from power connectors on mainboard to other PCIe slots has much lower impedance.
[quote author=Nutnut]
[quote author=ghostlander]
2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.[/quote]Please elaborate why a cap would help? Isn’t the power from the PSU clean enough? The system will simply use the existing ground on the slot. Capacitors seem to blow once the power steps up which will happen if the wires aren’t cut.[/quote]
It isn’t clean enough. Bulk capacitors inside PSU are capable of basic low frequency filtering only. Additional capacitors at loads are required. PCIe ribbon extenders (these are not risers) add considerable amounts of impedance to power rails. That includes earth. Good quality capacitors don’t bulge.
[quote author=Nutnut]
[quote author=ghostlander]That’s only in theory. In practice, the difference is only one earth pin. The other earth pin added is used for cable detect only. Actual limit is 6A per pair of +12V and earth pins, that’s 18A or 216W per connector. Pay attention to cable length and thickness though. AWG20 is no good unless very short (1 foot max.), AWG18 is all right at up to 2 feet, AWG16 performs well at any reasonable length.[/quote]Agreed but the SOCKET on the cards seems to be limited and won’t actually pull more than spec. Guessing it’s to protect the VRMs but I have no concrete evidence though others seem to report it and it does explain a few things.
[/quote]5A per pin pair is regular socket’s capability. Good quality gold plated ones can do in excess of 6A. 6/8-pin design isn’t to protect VRMs, it’s to prevent users from connecting old 6-pin only PSUs to modern power hungry cards while maintaining backward compatibility. Can be worked around with ease though.
-
Hmm what about pci-e 1x to 16x extenders?
-
[quote name=“svennand” post=“17392” timestamp=“1372088576”]
Hmm what about pci-e 1x to 16x extenders?
[/quote]Exactly the same.
[quote author=ghostlander link=topic=2193.msg17391#msg17391 date=1372088537]
[quote author=Nutnut]
[quote author=ghostlander]
1. No need to cut the cable. Those wires are better to keep. Most of the current flows through the molex connector.
[/quote]I disagree. If i only power 1 card with the riser, the other 3 cards can pull the power back down the ribbon cable, through the pci-e slot, across the board and in to the other cards. This has the potential of burning the track on the mobo and/or melting the ribbon/yellow cable.[/quote]
Ohm’s law disagrees with you. Path from power connectors on mainboard to other PCIe slots has much lower impedance.
[quote author=Nutnut]
[quote author=ghostlander]
2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.[/quote]Please elaborate why a cap would help? Isn’t the power from the PSU clean enough? The system will simply use the existing ground on the slot. Capacitors seem to blow once the power steps up which will happen if the wires aren’t cut.[/quote]
It isn’t clean enough. Bulk capacitors inside PSU are capable of basic low frequency filtering only. Additional capacitors at loads are required. PCIe ribbon extenders (these are not risers) add considerable amounts of impedance to power rails. That includes earth. Good quality capacitors don’t bulge.
[quote author=Nutnut]
[quote author=ghostlander]That’s only in theory. In practice, the difference is only one earth pin. The other earth pin added is used for cable detect only. Actual limit is 6A per pair of +12V and earth pins, that’s 18A or 216W per connector. Pay attention to cable length and thickness though. AWG20 is no good unless very short (1 foot max.), AWG18 is all right at up to 2 feet, AWG16 performs well at any reasonable length.[/quote]Agreed but the SOCKET on the cards seems to be limited and won’t actually pull more than spec. Guessing it’s to protect the VRMs but I have no concrete evidence though others seem to report it and it does explain a few things.
[/quote]5A per pin pair is regular socket’s capability. Good quality gold plated ones can do in excess of 6A. 6/8-pin design isn’t to protect VRMs, it’s to prevent users from connecting old 6-pin only PSUs to modern power hungry cards while maintaining backward compatibility. Can be worked around with ease though.
[/quote]
So I didn’t think there was any additional filtering between the 24pin +12v and the pcie slots. Is there any filtering between the 24pin socket and the pcie slots? If yes then the impedance would higher via the 24pin therefor current would flow back down the ribbon cable. If there’s no filtering then there’s no difference filtering wise between the powered riser/extender and the 24pin therefor a cap is not needed. Also, the graphics cards themselves will have the filters in their own power circuits. Then there’s the issue of multiple psu’s, if you are running multiple power supplies you don’t want to be mixing the +12v so you MUST cut the 12v.
The “power filters” (capacitors) that have started to appear IMHO are nothing more than a marketing gimmick designed to double the price. I never even saw one on an extender until people started complaining that they were exploding! Nearly ALL powered extenders out there are +12v only. No ground and no cap. Granted they don’t sever the wires but then they say that you MUST plug the riser in to the primary psu because they don’t cut the power lines.
Again i’m not an electrical engineer and people can add caps in-line if they wish. They can also leave the the ribbon un-severed if they wish (prided they are single PSU. This is how i built mine and they’re working lovely. :D
-
[quote]So I didn’t think there was any additional filtering between the 24pin +12v and the pcie slots. Is there any filtering between the 24pin socket and the pcie slots? If yes then the impedance would higher via the 24pin therefor current would flow back down the ribbon cable.[/quote]
How come additional filtering implies higher DC impedance? Resistance actually, Z = R + jX, and the latter is zero for DC. It’s defined basically by physical characteristics of PCB copper layers and connector quality. There are additional bulk capacitors on mainboards near loads to reduce transient voltage drops and improve high frequency response. VRMs operate at clock speeds up to 1MHz and above while switching PSUs work at 25kHz to 100kHz usually. We’re talking about basics now. Advanced noise filtering is much more complicated.
-
[quote name=“ghostlander” post=“17406” timestamp=“1372093311”]
[quote]So I didn’t think there was any additional filtering between the 24pin +12v and the pcie slots. Is there any filtering between the 24pin socket and the pcie slots? If yes then the impedance would higher via the 24pin therefor current would flow back down the ribbon cable.[/quote]How come additional filtering implies higher DC impedance? Resistance actually, Z = R + jX, and the latter is zero for DC. It’s defined basically by physical characteristics of PCB copper layers and connector quality. There are additional bulk capacitors on mainboards near loads to reduce transient voltage drops and improve high frequency response. VRMs operate at clock speeds up to 1MHz and above while switching PSUs work at 25kHz to 100kHz usually. We’re talking about basics now. Advanced noise filtering is much more complicated.
[/quote]So, being a man of action, i went to the shed armed with a meter, stuck it inline with the riser (no connection back to the motherboard) and here’s the results
1 card is pulling 3.1amp @ 12v through the pcie riser
Now i stuck the meter in line on the riser (creating the connection back to the motherboard). The rig was now running 3 cards and it was pulling 3 amp through the riser and 3.1 amp through each of the 24pin 12v lines for 9a total.
Result? The current draw was pretty much exactly divided between the riser and the 2x12v connectors on the 24pin so there IS power being pulled back down through the riser. I’ll keep the lines cut.
As for “power filters”… Nothing like algebra to send me to sleep but regardless, I’m SURE that the power filtering on the graphics cards will out perform and negate the need for a capacitor slapped on the riser.
-
hehe can you guys agree soon? ;D
Waiting patiently for final solution=P
Since im ALL out of luck my mainrig also went down.
This time it probably was my own fault, had issues with computer died after about an hour mining.
Since everything allready was fubar i desided to do some spring cleaning in the computer, then all of a sudden i saw a 6pin connector on the motherboard (Asus Extreme V), nicely hidden close to the CPU, after
confronting the manual about this highly exiting find i found out that the connector was for supplying the pci-e sockets with extra juice for multicard setup (thaaaats me!).
When rearranging the 7970 cards (have two of them in that computer) i saw something not that fun (another cable has meltet! 6+2pins from my seasonic x1250!)
[img]http://s22.postimg.org/6uxwxvqep/bilde_3.jpg[/img]Replaced this cable and hopefully it will soon be fixed again.
Within one week i have fried on 24pin connector, 2 motherboards burned to crisp, and now a 6+2pins connector… Ill go all in for the powered extenders! i swear :oNo wonder that computer didnt like it when i pushed 4x 7970 in it and cranked it up=P (Gigabyte 7970 Ghz edition 1100/1500 stock).
-
What on earth are you doing man?? How have you burnt an 8pin? There must be sommat up coz that ain’t right! :o
In a panic (i never thought to check them) i just looked at all of my connectors and they’re cool to the touch. I’m using seasoninc x1250 with 4x gv-797oc-gd3 (not the OC edition but still cranked to 1080/1500) and they’re fine. As are the other 2 rigs with the AX1200i. ???
Speechless.
-
[quote name=“Nutnut” post=“17429” timestamp=“1372104613”]
What on earth are you doing man?? How have you burnt an 8pin? There must be sommat up coz that ain’t right! :oIn a panic (i never thought to check them) i just looked at all of my connectors and they’re cool to the touch. I’m using seasoninc x1250 with 4x gv-797oc-gd3 (not the OC edition but still cranked to 1080/1500) and they’re fine. As are the other 2 rigs with the AX1200i. ???
Speechless.
[/quote]I have no idea how it has happend:P seems pretty far out. But nothing surprices me anymore :-\
-
[quote name=“Nutnut” post=“17409” timestamp=“1372095671”]
So, being a man of action, i went to the shed armed with a meter, stuck it inline with the riser (no connection back to the motherboard) and here’s the results1 card is pulling 3.1amp @ 12v through the pcie riser
Now i stuck the meter in line on the riser (creating the connection back to the motherboard). The rig was now running 3 cards and it was pulling 3 amp through the riser and 3.1 amp through each of the 24pin 12v lines for 9a total.
Result? The current draw was pretty much exactly divided between the riser and the 2x12v connectors on the 24pin so there IS power being pulled back down through the riser. I’ll keep the lines cut.
[/quote]3A per card is very much within the PCIe specification. That’s all right if you have one or two cards installed as two +12V pins of 24-pin EATX connector can forward enough current plus feed some ICs on mainboard. Although many Crossfire/SLI enabled mainboards come with additional Molex 8981 style connector to make sure you never exceed allowable current. Both ASUS A8N-SLI Premium and ABIT AT8 32X in my oldest rigs have one. If you set up three cards and have no Molex, it could be a problem. If there are four cards or more, expect a big problem to come.
Now the situation. You have four cards in a rig, each at 3A through a PCIe connector. One receives power through a powered riser, the other three consume 9A total through the 24-pin EATX connector on mainboard. That’s not good. Need to off-load some current by powering another riser or by connecting the existing one to the mainboard, so up to 2A may come through. I think the latter is better.
-
[quote author=ghostlander link=action=profile;u=448 date=1386650016]
[quote author=lbr link=action=profile;u=24747 date=1386624022]
Hello,For some reason I can’t post in the thread(prly cause I’m just registered).
[url=http://forum.feathercoin.com/index.php/topic,2193.msg17391.html#msg17391]http://forum.feathercoin.com/index.php/topic,2193.msg17391.html#msg17391[/url]
[quote]Ohm’s law disagrees with you. Path from power connectors on mainboard to other PCIe slots has much lower impedance.
[/quote]
Plz define “much lower”?
I mean… ok, voltage drop will occur, so what if it’s not too high? Right?
And for multiple PSU usage… cutting is absolutely must, right?[quote]It isn’t clean enough. Bulk capacitors inside PSU are capable of basic low frequency filtering only. Additional capacitors at loads are required. PCIe ribbon extenders (these are not risers) add considerable amounts of impedance to power rails. That includes earth. Good quality capacitors don’t bulge.
[/quote]F.e. CoolerMaster-SIlentProM 1000W PSU, second line caps are 16v 1600uf, low esr ones.
Imo they are good enough for not only low frequency filtering…
or… there should be some low capacitance film caps as well?If caps should be added on riser, which ones?
I assume some solid 16v?Also, why exactly caps at loads are required?
Thanks
[/quote]Make one post in the forum for newbies first. Much lowe is significantly lower. It depends, but the difference may be high. Multiple PSUs are fine as long as you synchronise them properly. 1600uF/16V are not good for HF as they’re liquid electrolytes. Solid or polymer tantalum are good. Watch the polarity as tantalums explode if reversed.
[/quote]
Thanks for your reply.
Sorry for asking so much questions, but you are the first person I’ve seen on digital currencies forums who actually has some technical understanding of electricity ; )Ok, how do I measure impedance?
I can use digital testers, analog oscilloscope.
Impedance is resistance under voltage applied, right?
What do you mean by “synchronise them properly” on/off sync?I always cut my risers in fear that +12v PSUs lines will “meet”… and nothing good happens ; )
Was my assumption right?Hm, what if I took two oscilloscope readings - from PSU +12v line and from motherboard +12v line, would it help to determine which caps exactly are needed to be soldered on the riser?
-
[quote name=“ghostlander” post=“17356” timestamp=“1372081622”]
2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.
[/quote]so… riser should also have thick-enough earth wire connected to PSU earth?
-
It’s a bit complicated to measure very low impedance with consumer grade digital multimeters. We’re talking about DC, so R=U/I. The PSUs are synchronised if they can be powered up by a single button and don’t interfere with each other. It’s not good to have rails crossed of the same PSU. It’s even worse to have them crossed of different PSUs. A good scope is what you need to check for power supply quality. It’s a matter of thumb to place capacitors as close as possible to the point of load. Small HF capable capacitors come first, bulk electrolytes come last.
[quote name=“lbr” post=“42649” timestamp=“1386704212”]
[quote author=ghostlander link=topic=2193.msg17356#msg17356 date=1372081622]
2. Attached +12V but no earth. Bad. Additional capacitor or even two can be also useful.
[/quote]so… riser should also have thick-enough earth wire connected to PSU earth?
[/quote]Yes. All power wires should be as thick as you can afford. We are not into a mass production, so the extra cost is negligible.
-
[s]So for 1x-1x riser scenario is absolutely the same, right?
Mission accomplished?
[url=http://www.ebay.com/itm/25CM-PCI-E-1x-To-16x-Riser-Card-Ribbon-Extender-Extension-Cable-With-Molex-Power-/201002973530?pt=LH_DefaultDomain_0&hash=item2eccb5f95a]http://www.ebay.com/itm/25CM-PCI-E-1x-To-16x-Riser-Card-Ribbon-Extender-Extension-Cable-With-Molex-Power-/201002973530?pt=LH_DefaultDomain_0&hash=item2eccb5f95a[/url]And the one with capacitor? Now after some time do you think the caps are needed?
[url=http://www.ebay.com/itm/4PIN-Molex-Power-Filter-PCI-E-Extension-Cable-1x-16x-Extender-Riser-Card-Adapter-/181151058265?pt=LH_DefaultDomain_0&hash=item2a2d71c959]http://www.ebay.com/itm/4PIN-Molex-Power-Filter-PCI-E-Extension-Cable-1x-16x-Extender-Riser-Card-Adapter-/181151058265?pt=LH_DefaultDomain_0&hash=item2a2d71c959[/url][/s]Screw this… I go for 16x-16x riser (4USD) and will fully follow Nut’s instructions.
-
[s]nutnut, will you update your guide with how to add eathing to the risers?[/s]
edit:
So after my serious brain fart and meltdown, I figured out the right sentence structure;
Nutnut, will you be updating your guide with how to add grounding to the home-made risers? -
I’ve order some ‘normal’ risers so i am going to read this topic carefully soon!
Added the guide to:
[url=http://forum.feathercoin.com/index.php/topic,685]http://forum.feathercoin.com/index.php/topic,685[/url] -
Any chance this could be compressed into one page, I’ll edit into the pinned guide in mining, if required?
I found it confusing what was decided safe at the end. Possibly room for someone to do a Youtube / Blue Peter “video how to”.
I just use a riser, but I like to see steam punk solutions.
-
Anyone has any objections to the following -
Ideal riser should be
0*) PCIEx16 -> PCIEx1, cause mining does not require bandwidth and x1 link theoretically puts lower strain on the mg chipset/CPU.
Tho some motherboards will require presence shorting if x1 risers are used in x16 slots.-
“cut” - to prevent current flow from/to motherboard voltage regulatorsPSU and to enable use of multiple PSUs
-
have thick enough wires on +12v lines, so they won’t burn, but not thick enough to cause significant voltage drop
-
all +12v lines shorted on the riser - to not overcurrent one pin only
-
wires soldered directly to slot
5*) +5v/+3v lines also going to PSU --> “ideally”
-
capacitors on +12vground, closer to load - better, HF fast capacitors, tantalum(?), solid(?), rating(?)
-
ground line also soldered and connected to PSU
Ideal multi-PSU
- different PSUs +12v rails separated from each other
- different PSUs +5/+3 rails separated from each other
3*) same PSU rails not mixing with each other - Properly synced - aka turning on/off at the same time
* does not really matter or not in the scope of the subject
-
