This is interesting:
So I've been thinking about the power supplies and some of Tarkin's thoughts. He discovered that the supplies have different programming and I was wondering why. Why are the supplies programmed differently and what can you do with that?
Out came the scope to take a look at the phase varience on each supply. We know that two supplies should beat 180 degrees out of phase with each other, that's the master/slave relationship on each side. Provides smooth power for the chip with less capacitors required.
What I see now is that they had a master clock signal coming from one of the supplies, and the 4 pairs adjusted their phase to be 180 degrees from their pair as well as 45 degrees offset from each other pair.
In other words, the first set fired at 0 and 180. The second set fire at 45 and 225. The third set fire at 90 and 270, and the fourth fire at 135 and 315. That way the 12 volt rai sees an 8 phase power supply pulling from it, which once again makes for a smooth ride.
It also means you can't swap supplies without knowing the exact position, and why new supplies don't work right. Fascinating.
Meantime I fixed some more boards, most complex was one that had two supplies blown and 3 of the 4 drivers. Interesting.
C
Yeahp, I get that same feeling when looking at the register configurations on each DCDC. It seems they have each pair of DCDC's for a die set to be a current sharing group and auto phase control is turned on for the pair. Furthermore, each pair has its own unique group identifier and unique interleave setting ... which may be the phase variance between each group you mention. If not then theres some other group registers that seem to be configured similarily.
I imagine if someone wanted to swap the 40A ones for the 50A DCDC variant, all they would have to do is ensure the configuration of each is copied over.
When I spoke w/ Ericsson, they have a usb-pmbus adapter which you could use to hook up to the 10pin connector on each cube, this can be used to run their free gui software suite for configuring the DCDC's. Fascinating stuff to say the least.
Also, Im happy to report my Titan is now going on 203+ hrs of uptime w/o any dies going to sleep ... think this is a record.
Been like this since I changed to 200khz switching freq, could just be coincidence tho w/ some cable rearrangement on the power side of things haha! Also, DCDC's are still bout 2-4C cooler each.
One thing thats also different which I didnt think of before is ... since I updated to 200khz and RESAVED the USER_STORE_ALL values for each DCDC ... that means my custom voltages are now set in the USER_STORE_ALL ... so from the very first power on of the DCDC's they are supplying the specified voltage to the dies(not default voltage which I think is .85v) and not waiting till the rpi boots up fully to run waas to set voltage values from the advanced.conf file.
I have no idea if that would really affect anything.