Ok guys, I have some more results with my E-Power testing. This is a long story so either sit back and follow with me or just ignore it all
I got my new epower board a few months back (@sunset1
helped me get it) and I have been playing with it non stop since it arrived on my doorstep. I quickly figured out that it was a lot of work getting it mounted and making the card work with the epower device. I wanted to come up with a solution that would allow me to quickly change the epower from one card to another without taking it to work and soldering it. Mark (sunset1) and I came up with the idea to use connectors and prepare multiple cards so it could be swapped easily.
After lots of research I finally found a connector that had huge amperage ratings and would be perfect for my application. I knew that power delivery was critically important for the epower so I wanted to make sure the connectors and wiring I chose would be suitable. After lots of math that I wont go into right now I finally came up with a solution using that connector and wire combination that would give me 88W per wire at a length of 20 feet. The epwoer itself can deliver a maximum of 400W so if each wire was rated for 88W, then 5 wires would max out the epower. After my initial "modular" epower testing I was really happy that I found a working solution to make the epower completely modular. I modded several cards for modular epower and they all seemed to work well but I never did any hay OC testing, just mild volts/OC. I have several cards prepped for modular epower, but here is my HD4890 and a teammates HD4870 as examples of the design.
Once I had my ambient submissions in for this competition, I decided to mod my HD4890 (blue pcb above) in the same modular way I have been doing for the last month. All was going well until I attempted to replicate my previous results and could not. As I mentioned above, I could not achieve the same core OC with epower and I was getting 0.05v Vdroop (meaning epower volts were 0.05 more than GPU core volts). With the stock VRM and a simple voltmod I was able to pass vantatge at 1075 core. With the modular epower design, the most I could pass vantage was 1040 core, which is a loss of 3.3%.
This is counter intuitive to everything we know about external power delivery and I started to ask...WHY. Indeed, why would I get less OC with a much more advanced power delivery system. The obvious answer is that my wire gauge is not big enough or the connectors are not living up to their rating. With one wire and one side of the connector, I did some testing at work with an adjustable inductive load. I set it up for roughly 100W let it run for an hour. The wire was not hot and there was NO problem running 100W, so I know my connector/wire setup is legit. The next obvious answer, as DR4G00N pointed out, is that my wires are too long. If you look at all the external power setups out there, everyone is running short wires. I know my wiring is capable of raw current transmission up to 20feet, but I stated to get concerned about VRM switching frequency and long wires. So, obviously the next thing for me to do is setup my card just like everyone else. Super short wires, lots of ground, and some ground taken from the front of the card. Here are some pics of that setup
The first thing I noticed is that Vdroop is almost completely gone. It went from 0.05v to 0.01v difference. As soon as I saw the Vdroop result, I had the sinking feeling that my modular epower design was fail. Here is were it gets juicy... With the common epower design I was STILL not getting the same OC results as I did with stock VRM. Here is the break down of the OC results done with vantage:
Stock VRM: 1075 @1.5v)
Modular E-Power: 1040 core @1.5v (after 0.05V Droop)
Common E-Power: 1045 core @1.5v (after 0.01V Droop)
The common epower setup is really rock solid and much better than most I have seen. The wiring layout on this card can handle more power than the E-Power is capable of delivering and is about 300% overkill for this GPU.
My only conclusion is that the card was either damaged somehow with all the constant soldering or that the stock VRM was indeed better for this GPU core than the E-Power is. It seems that my modular design is not a complete failure, however, there is the matter of that large Vdroop @ 100% load. I will also conclude that the card seems a bit more stable with the common epower setup but for practical purposes they OC the same. For now, I am done tweaking the E-Power. I am going to leave it in the common setup and bench the card for this competition with some cold.
There is probably more to this story that I didn't think of, but that's enough for now...what are your thoughts?