BTC_ISTANBUL
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 06:15:31 AM |
|
The risers were designed for 3.9 A. When 480X series appeared they they draw 6.5 A.
So, they made like this if they increase voltage to 12 then turn it to at the riser 5 V , the wattage would remain same but the cables can carry the load safely.
|
|
|
|
Zorg33
Jr. Member
Offline
Activity: 144
Merit: 2
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 10:15:08 AM |
|
The risers were designed for 3.9 A. When 480X series appeared they they draw 6.5 A.
So, they made like this if they increase voltage to 12 then turn it to at the riser 5 V , the wattage would remain same but the cables can carry the load safely.
No, that is incorrect. The 99% of the power is fed from the 12V either way. The card draws only 0.5A from 5V (stepped down to 3.3V by the 1084 IC on the riser)
|
|
|
|
MarkAz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 12:48:10 PM |
|
No, that is incorrect. The 99% of the power is fed from the 12V either way. The card draws only 0.5A from 5V (stepped down to 3.3V by the 1084 IC on the riser)
In practice I've seen that most cards pull about 1a 5v normally, with peak demand sometimes going to 2a. This is from my test rig, which is a 6 card system (the displays are purely from the molex side power demand): ![](https://ip.bitcointalk.org/?u=http%3A%2F%2Fwww.analogx.com%2Fimages%2Fgpu%2Fdsc00072s.jpg&t=665&c=sDhb_YSQspE1rg)
|
|
|
|
asbator
Member
![*](https://bitcointalk.org/Themes/custom1/images/star.gif)
Offline
Activity: 68
Merit: 10
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 06:17:21 PM Last edit: September 29, 2016, 06:33:27 PM by asbator |
|
|
|
|
|
MarkAz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 07:13:16 PM |
|
I'm not a fan of those, I suspect the FDD pins don't handle enough constant power to make it a good idea... Even the normal 4 pin molex connector is a bit questionable with some of these high demand cards, like the reference RX 480's.
|
|
|
|
Zorg33
Jr. Member
Offline
Activity: 144
Merit: 2
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 07:30:24 PM |
|
The capacitors are important.I do not see any difference between versions.I own all the versions and they work.
there is difference between the caps, even though you don't see it. The earlier version has 100 uF caps, and the newer has 270 uF caps. In practice it does not mean much, they are just filtering caps there. I have risers from V2, V3, V5, V6 and all work perfectly. The quality and the shielding of the USB 3.0 cable is much more important. @asbator They are only good if you solder the wires directly onto the pins of the floppy connector, otherwise it's gonna burn.
|
|
|
|
charles2k
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 29, 2016, 11:38:19 PM |
|
Wow, I bow before your Google skills! ![Wink](https://bitcointalk.org/Smileys/default/wink.gif) If you actually look at the board, you can see that they've actually populated a bunch of the unused spaces on the riser, so I think it's pretty safe to say that they're doing the 12v -> 5v stepdown on the riser. I found a slightly different version that I liked a bit more and shot them an email to get some to do some testing with... I'm not certain that I like it better than sidehack's solution, but it's certainly worth evaluating. It is this version? https://www.alibaba.com/product-detail/PCI-E-x1-to-x-16_60536973971.html
|
|
|
|
Tmdz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 30, 2016, 06:11:57 AM |
|
I have all v005 and v006 on the ones I have, they all work exactly the same.
Sweet pics of that farm, that is a crazy amount of money invested there.
|
|
|
|
asbator
Member
![*](https://bitcointalk.org/Themes/custom1/images/star.gif)
Offline
Activity: 68
Merit: 10
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 30, 2016, 06:39:09 AM |
|
The capacitors are important.I do not see any difference between versions.I own all the versions and they work.
there is difference between the caps, even though you don't see it. The earlier version has 100 uF caps, and the newer has 270 uF caps. In practice it does not mean much, they are just filtering caps there. I have risers from V2, V3, V5, V6 and all work perfectly. The quality and the shielding of the USB 3.0 cable is much more important. @asbator They are only good if you solder the wires directly onto the pins of the floppy connector, otherwise it's gonna burn. Thank you Zorg. Do you mean cuting of SATA connector? My PSU has 2 FDD connectors, so i can use them directly and test it. Burning cable doesn't damage GPU ![Huh](https://bitcointalk.org/Smileys/default/huh.gif)
|
|
|
|
MarkAz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 30, 2016, 09:54:23 AM |
|
It is this version?
Yup, that's the one - I liked the fact that the USB cable comes out the back (they have all 3 orientations, and you can get whichever one you want). Just would work better for my builds... All these places say they manufacture them, but this place actually seemed like they were.
|
|
|
|
PVmining
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 30, 2016, 10:53:22 AM Last edit: September 30, 2016, 11:30:08 AM by PVmining |
|
very interesting thread!
I use a bunch of VER003 since two years and bought VER005 for some new rigs a few weeks ago. No problems so far. I would be happy if the soldering would be look more perfect, but no restriction because of that. Had only one VER003 that does not work out of the box.
Question: Does anybody know how much of the power were drawn from the PCIe Slot (and in our case through the USB risers) if using some 6 PIN GPU with max 150w? Does the GPU takes the most of the energy from the 6PIN?
@MarkAz. your rigs are looking beautiful!
|
|
|
|
asbator
Member
![*](https://bitcointalk.org/Themes/custom1/images/star.gif)
Offline
Activity: 68
Merit: 10
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
September 30, 2016, 12:05:22 PM |
|
I did some research and realized that "my" super-duper riser doesn't have 5->3,3 voltage stepdown so it must be used through sata adapter. I also found info on AMD site that excessive current draw over PCIE slot was a driver issue and was fixed. Theoretically GPU can draw up to 5,5A that way while SATA connector can withstain just 4,5A, but real draw should be smaller now (Markaz rig shows ~3A per GPU).
With 6pin cards it will be probably around 25% more (?)
|
|
|
|
Za1n (OP)
Legendary
Offline
Activity: 1078
Merit: 1011
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 02, 2016, 09:35:54 AM |
|
I did some research and realized that "my" super-duper riser doesn't have 5->3,3 voltage stepdown so it must be used through sata adapter. I also found info on AMD site that excessive current draw over PCIE slot was a driver issue and was fixed. Theoretically GPU can draw up to 5,5A that way while SATA connector can withstain just 4,5A, but real draw should be smaller now (Markaz rig shows ~3A per GPU).
With 6pin cards it will be probably around 25% more (?)
If I am interpreting the picture correctly, it looks like the cards are drawing ~3A each from the PCIe power connectors on the card from the 12V rail and another ~0.9A each from the 5V rail through the riser's Molex connectors. This seems a bit low for total card wattage power draw as these figures indicate only ~ 40 watts total draw per card.
|
|
|
|
Za1n (OP)
Legendary
Offline
Activity: 1078
Merit: 1011
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 02, 2016, 09:44:50 AM |
|
Wow, I bow before your Google skills! ![Wink](https://bitcointalk.org/Smileys/default/wink.gif) If you actually look at the board, you can see that they've actually populated a bunch of the unused spaces on the riser, so I think it's pretty safe to say that they're doing the 12v -> 5v stepdown on the riser. I found a slightly different version that I liked a bit more and shot them an email to get some to do some testing with... I'm not certain that I like it better than sidehack's solution, but it's certainly worth evaluating. It is this version? https://www.alibaba.com/product-detail/PCI-E-x1-to-x-16_60536973971.htmlThose risers do look boss. Would be interested to hear how they work out. Assuming they are the real deal, if someone were to do a bulk order and would want to sell some off in the US, I would interested in acquiring a few dozen. For now I ended up going with a narrow design that I found on ebay (seller in Texas) which has a 4-pin connector on the riser and this uses a small cable to connect to a regular molex feed. No version number on the board that I could find. I only ordered 6 to test with and installed them all before I could snap a picture. I have a second larger order coming in now that I know they work and will take a few pictures and post.
|
|
|
|
MarkAz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 02, 2016, 10:07:59 AM |
|
If I am interpreting the picture correctly, it looks like the cards are drawing ~3A each from the PCIe power connectors on the card from the 12V rail and another ~0.9A each from the 5V rail through the riser's Molex connectors. This seems a bit low for total card wattage power draw as these figures indicate only ~ 40 watts total draw per card.
You're half right, those displays are only monitoring power being drawn over the Molex connector, so your math is correct, but that's only what it's drawing over the PCIe slot - it's pulling alot more through the 1x 6-pin and 1x 8-pin on each board. And those numbers were generally accurate for all my R9's and GTX970 (they used a bit less), then the RX's used a bit more... So if you err on the side of caution, I would have more power than needed on the riser.
|
|
|
|
Za1n (OP)
Legendary
Offline
Activity: 1078
Merit: 1011
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 02, 2016, 04:22:03 PM |
|
If I am interpreting the picture correctly, it looks like the cards are drawing ~3A each from the PCIe power connectors on the card from the 12V rail and another ~0.9A each from the 5V rail through the riser's Molex connectors. This seems a bit low for total card wattage power draw as these figures indicate only ~ 40 watts total draw per card.
You're half right, those displays are only monitoring power being drawn over the Molex connector, so your math is correct, but that's only what it's drawing over the PCIe slot - it's pulling alot more through the 1x 6-pin and 1x 8-pin on each board. And those numbers were generally accurate for all my R9's and GTX970 (they used a bit less), then the RX's used a bit more... So if you err on the side of caution, I would have more power than needed on the riser. Ahh, ok thanks for clarifying. I figured it was something I was overlooking and that makes more sense. Still, only 40 watts over the PCIe slot seems quite reasonable and should be easily handled via the Molex connectors. This is good information to know as I have never been able to independently measure the different power draws (slot versus the 6 or 8-pin connectors) myself. I assume these readings are typical for most cards?
|
|
|
|
MarkAz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 02, 2016, 06:46:45 PM |
|
Ahh, ok thanks for clarifying. I figured it was something I was overlooking and that makes more sense. Still, only 40 watts over the PCIe slot seems quite reasonable and should be easily handled via the Molex connectors. This is good information to know as I have never been able to independently measure the different power draws (slot versus the 6 or 8-pin connectors) myself. I assume these readings are typical for most cards?
So I tested probably 4 makes of R9 390's and they all were around this; I tested one make of GTX970's and it was less, I tested 2 makes of RX480's and it was considerably more, and I tested 2 brands of RX 470 but I can't remember what they ended up - I'm pretty sure it was less than the 390's, either way it was nothing to be overly concerned about. I need to make one mega test box, that shows power consumption for everything broken out... My next build is probably going to be a 7 slot, so maybe I'll do that as well.
|
|
|
|
Iegion
Newbie
Offline
Activity: 39
Merit: 0
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 03, 2016, 01:16:16 AM |
|
nice farm mark, did you build that frame yourself?
|
|
|
|
|
Zorg33
Jr. Member
Offline
Activity: 144
Merit: 2
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
October 03, 2016, 02:58:31 PM |
|
I see that in Sidehack's solution the riser's 5V is also fed from 12V. Or just the same yellow wire is used for 5V...but I doubt it, because you need color coding if you run the wires in insulation tubes. Thank you Zorg. Do you mean cuting of SATA connector? My PSU has 2 FDD connectors, so i can use them directly and test it. Burning cable doesn't damage GPU ![Huh](https://bitcointalk.org/Smileys/default/huh.gif) No, I mean cutting off the floppy connector, because it's very weak for this application.
|
|
|
|
|