- Joined
- Jan 19, 2006
- Location
- Florida
I asked "B" to do a simple test run and post it as well 
maybe we get a look see before I get it in my hands haha
maybe we get a look see before I get it in my hands haha
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
7970 Reference Vmod Log
- Thread starter Bobnova
- Start date
OP
- Joined
- May 10, 2009
- Thread Starter
- #23
I've done some testing on the OCP mod, as a number of people have reported SMDS (Sudden MOSFET Death Syndrome) with it enabled.
IMO I've found the cure: Larger resistors.
The controller really needs to see at least some of the current headed to the core in order to function properly, my guess is that it activates different numbers of phases depending on current draw.
The "stock" mod uses a 47kOhm resistor between phase current feedback and ground, with that enabled the current sensed drops to 0.3a even if the core is drawing 60a!
It also results in Bad Noises from the power section.
I played with adding resistance between the common pin on the bus resistor and GND and found that ~440k ohms knocks 30% of the sensed current, that means the OCP point is 30% higher but the card will still sense most of it and react correctly.
Future versions of this mod will have a socket for the bus resistor so it can be swapped out with different resistances, for this card I'm swapping the enable/disable pins (the pins by the power connector) with a 2p socket for sticking resistors in. That makes it tunable, but not as clean or "correct" as the bus resistor being swappable would be.
Super Quick Simple Test Run A gave X2150 or so in Heaven, I'm going to wind the CPU up and swap AB versions and see if the ram is interested in going higher now. Results in a bit!
Then it'll be on to boring review work to clear my plate to work on card2
To prospective modders:
The #1 hardest part for the majority of mods is the soldering itself, grab yourself a Weller (accept no substitutes!) 30-40w iron with a fine tip and start playing with $10 GPUs and dead motherboards.
Bonus: I am always happy to answer questions and offer guidance if I can.
IMO I've found the cure: Larger resistors.
The controller really needs to see at least some of the current headed to the core in order to function properly, my guess is that it activates different numbers of phases depending on current draw.
The "stock" mod uses a 47kOhm resistor between phase current feedback and ground, with that enabled the current sensed drops to 0.3a even if the core is drawing 60a!
It also results in Bad Noises from the power section.
I played with adding resistance between the common pin on the bus resistor and GND and found that ~440k ohms knocks 30% of the sensed current, that means the OCP point is 30% higher but the card will still sense most of it and react correctly.
Future versions of this mod will have a socket for the bus resistor so it can be swapped out with different resistances, for this card I'm swapping the enable/disable pins (the pins by the power connector) with a 2p socket for sticking resistors in. That makes it tunable, but not as clean or "correct" as the bus resistor being swappable would be.
Super Quick Simple Test Run A gave X2150 or so in Heaven, I'm going to wind the CPU up and swap AB versions and see if the ram is interested in going higher now. Results in a bit!
Then it'll be on to boring review work to clear my plate to work on card2
To prospective modders:
The #1 hardest part for the majority of mods is the soldering itself, grab yourself a Weller (accept no substitutes!) 30-40w iron with a fine tip and start playing with $10 GPUs and dead motherboards.
Bonus: I am always happy to answer questions and offer guidance if I can.
OP
- Joined
- May 10, 2009
- Thread Starter
- #24
I'll have to take this result down when I send the card back of course, but for now here's the result of some fairly quick air benching. Stock cooler.
3DMark11 - 11572 marks.
Card wasn't interested in >1200 core regardless of voltage (1.25v and up), I suspect lousy contact between core and heatsink as the cause.
Memory slider maxed out even in AB-X, T&B's going to have to figure out something to do about that, it has lots of volts and clocks left in the tank I think.
3DMark11 - 11572 marks.
Card wasn't interested in >1200 core regardless of voltage (1.25v and up), I suspect lousy contact between core and heatsink as the cause.
Memory slider maxed out even in AB-X, T&B's going to have to figure out something to do about that, it has lots of volts and clocks left in the tank I think.
- Joined
- Jan 19, 2006
- Location
- Florida
Mem 1790 easy oh yeah zoom zoom zoom Ill get the GPU up! I think this is the max AB I have for competition was sent to me MSI will have to see if theres anything more I can get.
NICE WORK "B"
- Joined
- Mar 27, 2011
- Location
- Soviet Mexico
that is some impressive memory speed, Will be interesting to see if core temp is possibly limiting that from going much higher . If anything. Makes we want to do some ram swapping from a 7970 to a 680
. If anything. Makes we want to do some ram swapping from a 7970 to a 680
OP
- Joined
- May 10, 2009
- Thread Starter
- #27
Further OCP testing has shown that there is a jump in vcore when OCP mod is enabled.
The lower the resistance on the OCP mod the higher the jump, also the higher the load the higher the jump.
At 1200MHz and 1.25v enabling the mod with a total of 487kOhms adds 0.04v.
Same clocks/volts/load (paused Heaven dragon scene) enabling the mod with a total of 267kOhms gives a vcore jump of 0.11v!
The lower the resistance on the OCP mod the higher the jump, also the higher the load the higher the jump.
At 1200MHz and 1.25v enabling the mod with a total of 487kOhms adds 0.04v.
Same clocks/volts/load (paused Heaven dragon scene) enabling the mod with a total of 267kOhms gives a vcore jump of 0.11v!
OP
- Joined
- May 10, 2009
- Thread Starter
- #29
Found a partial datasheet a while ago, while re-reading it and arming myself for attacking the GBT 7970OC that is next on my mod-plate (it has all seven (6+1? 5+2?) phase slots occupied, reference only uses six (5+1), I ran across this tidbit in the bragging section:
That'd be why the standard OCP mod fries MOSFETs, it removes almost all to all current sensing, 100 amps looks like <10 amps and the thing switches to single phase mode. POOF goes the poor MOSFET that has to switch for it.CHiL said:
- Joined
- Jan 19, 2006
- Location
- Florida
Is there a easy work around on that?sounds like thats the answer to the big cook out!
- Joined
- Mar 27, 2011
- Location
- Soviet Mexico
That is some impressive work there, and some really good RAM speeds. TJ, do you think you would be able to swap the memory like that?
Yes its entirely possible to swap the ram, I would needed to get a Better stencil for ddr3 cause the one i have is garbage. It takes around 15-25m a chip for me right now to reball and mount.
Its far easier to desolder and reball/mount the gpu than it is all the memory chips. But thats not really an option if your going from a 7970 to a 680. I'll have to see if I can get my hands on the stencils for the 7970 and 680, theres a good chance I can.
OP
- Joined
- May 10, 2009
- Thread Starter
- #33
Yup, a socket to swap resistors out so that you can adjust how much current it's seeing.
Then you can keep it under the OCP limit, but over where it starts dropping phases.
It's already installed
- Joined
- May 17, 2011
- Location
- Philly Suburbs
Bob, you think lightning mods would be the same? I like this red card
- Joined
- May 29, 2005
OP
- Joined
- May 10, 2009
- Thread Starter
- #38
One allows control over the PLL/VTT/VDDCI/whatever voltage, which should allow the card to clock higher and/or be cooled closer to "full pot" LN2 temps of 196c below zero. Colder temps mean slower death and higher max clock speeds.
The second grey boxy thing allows control over memory voltage. Putting the mod on in the first place bumps memory volts from 1.65v to 1.75v, it can be raised from there with the trimmer (grey box).
The complicated and messy bits are for disabling/adjusting the OCP trip point on the core power bits. Without the mod the card shuts down at around 1500MHz. That's fast, but we want 1700+, so OCP has to be adjusted.
It's worth nothing that the expected lifespan of this card has gone down a touch, as it's been transformed from grandma's sporty civic to a Serious Racing Car.
Civic'll do 200k-300k miles easy.
S.R.C. will do a couple thousand.
Don't do this to your 24/7 card
(It may actually last for quite a while, some cards love LN2 time and go for a long time. Others, especially with the "standard" OCP mod (complete removal of current sensing) die rather quickly. We'll be happy if it lives long enough to get some topish results in the various benchmarks.)
Also went back with rested eyes and brains and looked at things, the new improved memory performance is explained!
Due to how the voltage feedback works the increased resistance raises the voltage. That 0.8-0.1v extra brought the ram from 1500MHz or so to the slider cap of 1790MHz. Future mods will have a lower value resistor and hence a lower minimum jump. I prefer to mod things so they can be disabled entirely, but that is excessively complicated on modern CHiL bits.
The second grey boxy thing allows control over memory voltage. Putting the mod on in the first place bumps memory volts from 1.65v to 1.75v, it can be raised from there with the trimmer (grey box).
The complicated and messy bits are for disabling/adjusting the OCP trip point on the core power bits. Without the mod the card shuts down at around 1500MHz. That's fast, but we want 1700+, so OCP has to be adjusted.
It's worth nothing that the expected lifespan of this card has gone down a touch, as it's been transformed from grandma's sporty civic to a Serious Racing Car.
Civic'll do 200k-300k miles easy.
S.R.C. will do a couple thousand.
Don't do this to your 24/7 card
(It may actually last for quite a while, some cards love LN2 time and go for a long time. Others, especially with the "standard" OCP mod (complete removal of current sensing) die rather quickly. We'll be happy if it lives long enough to get some topish results in the various benchmarks.)
Also went back with rested eyes and brains and looked at things, the new improved memory performance is explained!
Due to how the voltage feedback works the increased resistance raises the voltage. That 0.8-0.1v extra brought the ram from 1500MHz or so to the slider cap of 1790MHz. Future mods will have a lower value resistor and hence a lower minimum jump. I prefer to mod things so they can be disabled entirely, but that is excessively complicated on modern CHiL bits.
- Joined
- Jan 19, 2006
- Location
- Florida
set up for tomorrow to cook the super mod on the 7970
Similar threads
