[EE] Untouchables Rev0.0 : usage
I hope Shammy would not mind if this thread appear. If any issues with it - feel free to delete, I would not publish any non-public data about these stuff on purpose, thats for sure.
So I've received my sample after month of customs clearing, and now doing first steps with it. So most things described here by me are lame, stupid and beginner trash. Hope things will clarify in future and i will get new knowledge later
I've assembled next rig:
i7 920 stock, box cooler
ASUS P6T Dlx reflashed to P6T6
Reference GeForce GTX 285
Shammy's Untouchables, set to initial voltage 1.2V & 2.37V
FIVE "Pci-express" power connectors for VGA hooked to Enermax Revo 1250W
Kingpincooling.com Tek9 6.66 LN2 pot
As for tools: I've used:
Fluke 87V DMM
Tektronix 2246-Y 100MHz scope
xDevs.com ES HC1 device prototype for digital control to Untouchable
xDevs.com VARM 2-channel voltage meter
Please take a notice, that I've previously used external VRMs on VGA's, so
had some positive and successful experience from that kind of things.
I've tried and benched:
1st: 7800GTX + 2-phase + 1-phase uPI-based VRM from radeon 4850
2nd: 8600GT + same radeon 4850 VRM
3rd: 9500GT + same radeon 4850 VRM
4th: GTX285 + GPU powered from mixed Volterra 1165MF 3-phase VRMs from 4870X2
I think k|ngp|n should now know how real Tek9 6.66 look like :-D
Better is only FAT pot with untouchabled GTX480 maybe
I've hooked extra VRM with usual technique, removing stock chokes and putting volts from external VRM into power plane inputs. Now I used 7 solid copper wires for VGPU power and 3 same wires for Vmemory power.
I've never used such type solid copper wires, only thick "audiophile-type" multiwired cables with thick insulation.
Also 10 GND wires was used as interconnect. I've hooked 2-channel voltage meter to out pads on EVGA card.
Videocard have added tantalum capacitors on memory and core VRM, just in points where wires joints.
Ripple and noise was measured by oscilloscope probe on these capacitors pads.
The untouchables was cooled by high-cfm 80mm fan, and VGA NVIO area was cooled by 120mm fan. I've used LN2 pot with temps -20 to -70°C, because it's the only cooler i have now, that fits this card. Further testing will be done with chiller unit.
All PCI-E power connectors was populated with cables from single Enermax Revolution 1250W PSU. TWO was hooked to VGA itself and three for EVGA VRM.
Total look w/o fans:
Before going windows with this rig, i've applied 1450mV voltage for main 10-channel power VRM.
It was done digitally from my ES device, I think same way as EVbot does.
So, i've booted and run 3dmark03 nature, on stock clocks for VGA and 1
Drop of voltage is huge, almost 100mV. I think I should use much more wires for such type of VGAs. And maybe better use multiwire cables, to reduce impedance and skin-effect?
During test and transients I've noticed high-pitched noise from VRM (usual stuff) and heavy horisontal bars flickering fast on screen. Tried to capture that with next photo. Horisontal lines more visible with heavy loads, and less with lighter, but present always in any 3D test and appear even during 3dmark03 loading in windows.
With lower volts (stock for GTX285 are near 1.1V), e.g. 1.25V card could not even do GT1 of 3d03.
It halts like this:
Then I powered on my scope to see what happen on power lines.
During idle everything was very nice.
20mV peak-peak ripple, it's very good. My stock GeForce 8800GTX have 70mV peak-peak idle noise.
Then I've run Nature from 3d03. On stock clocks, and 1.45vgpu and 2.3Vmem applied from EVGA The Untouchables.
Remember, these signals was aquired directly from bulk tantalum capacitors (total capacitance >3200uF VGPU, >1000uF Vmem) placed nearby stock VRM chokes location on VGA PCB.
OMG, Is what i say during test. 0.6V peak-peak noise is more than enough to forget about any overclocking completely
Memory was not better, 336mV peak-peak noise.
So looks like i have new experience.
Ideas to do next time:
1. Add large bulk electrolyte capacitors as close as possible to memory chips and gpu.
2. Triple amount of wires from VRM to VGA PCB.
3. Use multiwire cables for interconnect, not solid copper ones.
4. Add oscilloscope probe connection input directly below GPU on bottom side of VGA PCB, for most accurate measurement.
If nothing help then
5. Ask Shammy for help stupid Tinny
When I've used Radeon 4870X2 Volterra VT1165MF-based VRMs image from this GTX285 didn't flicker, and there was no such big noise. But I've used my own wires during that tests:
Also there was more caps (extra electrolytic 4x1000uF 4V and 2x680 6.3V's)
P.s. thanks for watching for those who saw my livestream session with this stuff.
Last edited by TiN; 05-16-2010 at 06:44 PM.
due to the distance away there will definitely be droop , you have to tune volts to the load volts.
connect vsense + and vsense - to the actual gpu +/- behind the gpu core will also help.
i prefer the solid wires, if all your current on MB / VGA uses this to flow to the core in your inductors, they work good and always have
the solder joints doesnt seem thorough, you did scrape off insulative layer?, maybe front and end both add solder?
Last edited by Shammy; 05-16-2010 at 08:28 PM.
are the solid cables insulated?
extra low esr polymer caps around 5mohms might help...
what tantalum caps are you using? looks like kemet stuff... got specs of it?
The cable is polyestermide/polyurethane insulated - so needs scraping slightly. Maybe bad joints. Multi strand wire won't help really (i don't think skin-effect is causing any kind of problem). I have my UT half connected up to a board here, when i get time I'll complete the mod and see what happens.
Last edited by Raja@ASUS; 05-17-2010 at 01:19 AM.
I've scraped both ends, but now see I could do work better. Anyway, will add more multistrand wires (thanks for word, Raja, I speak english not natively for now )
Added more wires. Situation improved twice.
Now addin even more
Sham, what about my #5 post? Hook feedback line to that points (where VR suggested)?
I DMMed them, one seems Vcore, second GND, but want to be sure with these volterras.
Ripple on memory voltage (measurement same point, VRM out capacitor on VGA PCB) reduced from ~300+mV to near 90-120mV. It's now acceptable (as stock VRM on such cards).
Ripple on core voltage improved from ~600mV shi(f)t to mere 200-220mV on stock clocks and 1.4V (dropping to 1.33-1.36) and 240-260mV with 850MHz clocks and near zero GPU monitoring temp.
Horisontal banding flicker still exist under load, by looks like less?
Besides adding wires, i've also added caps, 2 x 1000uF 2.5V chemicons, 680uF chemicon, 2700uF 6.3V MFZ for core. For memory added 3300uF 6.3V KZE and 2700uF 6.3 MFZ.
Ramped Vmemory to 2.4V as my friend Vince told me.
Now card ABLE to do 850/1440MHz with near zero temp. NO WAY it was possible with stock VRM, even with -140°C GPU temp. Instant halt in 3D with such core and memory clocks together.
Will do insulation on card, hook up MAGNUM pot to my CPU (920), and see how it go. If things go up, i will test this card with some hexcore this weekend (and maybe QHEX+REX for 3d01).
Banding. These horisontal lines should not be there
Passed nature at 848/1944/1440 with +14C monitoring temp
Ripple and volts during test
Ripple with 900MHz test (frozen in middle of test, but that because temp too high, i guess, near +10C)
Last edited by TiN; 05-19-2010 at 09:11 AM.
Try some HF (MLCC bypassed with silvered mica in picofarad range) decoupling after the wires on the card side Tin - maybe the banding is due to noise pickup in the high KHz region.
Last edited by Raja@ASUS; 05-20-2010 at 04:06 PM.
|even more crazy stuff, r&d|
|Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)|
|Thread Tools||Search this Thread|