Has anybody discovered how to undervolt Cell or RSX?

[raidriar]

Are voltage controls exposed anywhere in the system firmware or the syscon? Would be a great experiment to try and undervolt the cell or rsx especially in the 90nm units.

 

[Cruslan1]

No. Only we can overclock RSX. @zecoxao discovered both overclock RSX and CELL but only applicable RSX one. I believe undervolting in ps3 will never happen.

 

[villahed94]

It turns out that you can actually undervolt both.
As shown by @RIP-Felix in his videos about overclocking the RSX for massive performance increases, he uses the VID tables that SYSCON uses to increase voltages to increase stability at overclocked clocks.
The opposite is true too. If you keep stock clocks you can undervolt both RSX and CELL.
I have used the tables and undervolted my 90nm RSX on a CECHA-01 from its stock 1.2V VCore down to 1.00V.
This resulted in a 5-6C reduction in temps and a total system power draw reduction of ~15W (208 down to 193W! - a 7.5% delta)

Now I'm wondering what's the EEPROM address for addressing the CELL VID table.

 

[Falco1211]

It turns out that you can actually undervolt both.
As shown by @RIP-Felix in his videos about overclocking the RSX for massive performance increases, he uses the VID tables that SYSCON uses to increase voltages to increase stability at overclocked clocks.
The opposite is true too. If you keep stock clocks you can undervolt both RSX and CELL.
I have used the tables and undervolted my 90nm RSX on a CECHA-01 from its stock 1.2V VCore down to 1.00V.
This resulted in a 5-6C reduction in temps and a total system power draw reduction of ~15W (208 down to 193W! - a 7.5% delta)

Now I'm wondering what's the EEPROM address for addressing the CELL VID table.

This looks really promising, if you managed to knock down power draw by 7.5% imagine what you can do with the Cell which by far runs the hottest.

 

[villahed94]

This looks really promising, if you managed to knock down power draw by 7.5% imagine what you can do with the Cell which by far runs the hottest.

This is precisely what I'm wondering. I need to contact @M4j0r who is the real master of SYSCON for the EEPROM address to configure the buck controller for CELL.

I must emphasize, YOU CAN'T SET THE SAME VOLTAGES AS ME, each IC will have a different voltage/frequency curve. Newer ICs will have better characteristics than older ones.

For example, I just got another CECHA01, this one's earlier with an RSX datecode of 0645. It can't go below 1.10V or it'll crash. The one in this post has a datecode of 0710 and hence it can do lower voltages at stock clocks.

 

[NightLess]

This is precisely what I'm wondering. I need to contact @M4j0r who is the real master of SYSCON for the EEPROM address to configure the buck controller for CELL.

I must emphasize, YOU CAN'T SET THE SAME VOLTAGES AS ME, each IC will have a different voltage/frequency curve. Newer ICs will have better characteristics than older ones.

For example, I just got another CECHA01, this one's earlier with an RSX datecode of 0645. It can't go below 1.10V or it'll crash. The one in this post has a datecode of 0710 and hence it can do lower voltages at stock clocks.

Hello, have you found a way to lower the cell voltage?

 

[villahed94]

Hello, have you found a way to lower the cell voltage?

So far the only possible method is to hijack the output ports of the Syscon to the NCP buck controller and set the bit table manually.
Looking at a dump from my Syscon at the segment dedicated to platform configs(where the VID for RSX also resides) it's full of unset FFs.
I would assume the offset for CELL must be earlier than 0x3111 (and after 0x3100) because in most processes CELL comes before RSX.

 

[NightLess]

So far the only possible method is to hijack the output ports of the Syscon to the NCP buck controller and set the bit table manually.
Looking at a dump from my Syscon at the segment dedicated to platform configs(where the VID for RSX also resides) it's full of unset FFs.
I would assume the offset for CELL must be earlier than 0x3111 (and after 0x3100) because in most processes CELL comes before RSX.

This may be a bit difficult for me. If there is no detailed method, I don't know how to change it. Thank you for your answer.

 

[Starship2]

So far the only possible method is to hijack the output ports of the Syscon to the NCP buck controller and set the bit table manually.
Looking at a dump from my Syscon at the segment dedicated to platform configs(where the VID for RSX also resides) it's full of unset FFs.
I would assume the offset for CELL must be earlier than 0x3111 (and after 0x3100) because in most processes CELL comes before RSX.

that could be a good guess. i assume it to be in this area, too. if i got some time later i can do a little bit of research on a donor board. i'll let you know.

edit:
Im confident that we were right with our guesses. Cell VID seems to be at 0x310E.
As i dont have my multimeter right here, i cant surely confirm. On my DIA-002 there was a value of 30 what equals 1.45v if you translate it with the RSX table. Seems plausible to me or what do you guys think? Maybe someone can do quick measurements to approve. All other bytes in 0x3100 are FF, same as 0x3110 where RSX VID is located. Thats pretty suspicious

 

[villahed94]

that could be a good guess. i assume it to be in this area, too. if i got some time later i can do a little bit of research on a donor board. i'll let you know.

edit:
Im confident that we were right with our guesses. Cell VID seems to be at 0x310E.
As i dont have my multimeter right here, i cant surely confirm. On my DIA-002 there was a value of 30 what equals 1.45v if you translate it with the RSX table. Seems plausible to me or what do you guys think? Maybe someone can do quick measurements to approve. All other bytes in 0x3100 are FF, same as 0x3110 where RSX VID is located. Thats pretty suspicious

BREAKING NEWS: I FOUND THE ADDRESS.
It's 0x3110 for CELL and 3111 for RSX.
Will perform some testing later today to see how low can one go.
For the record, my CELLs are doing 1.18V whereas the schematic calls for 1.00V

 

[NightLess]

BREAKING NEWS: I FOUND THE ADDRESS.
It's 0x3110 for CELL and 3111 for RSX.
Will perform some testing later today to see how low can one go.
For the record, my CELLs are doing 1.18V whereas the schematic calls for 1.00V

I'm about the same voltage as you, I'll try it later.

 

[NightLess]

Well done villahed! I have successfully lowered the voltage of the cell, I will do more tests on it.

 

[NightLess]

I run TLOU for 6 hours at 1.023v and it was very stable, the fans stayed at 28% and the Cell was at 59 degrees, yes that's a 10 degree drop from stock voltage! The fans were 3% lower, incredible.

By comparing several power consumption records
Stock
XMB:~150w MUSIC:~158w TLOU:~177w PS2 Kingdom Hearts:~176w

Cell [email protected]
XMB:~120w MUSIC:~125w TLOU:~140w PS2 Kingdom Hearts:~137w

I ended up increasing the voltage a bit to prevent instability, the power consumption increased by ~5w.

 

[villahed94]

I'm going to check the voltage on mine from an oscilloscope. I did manage to go even lower, at a set point of 1.0375V according to the VID table, but a real voltage of 0.988V. It was surprisingly stable with not as demanding games, but TLOU triggered a RLOD of 1001 (BE Power Fail). Hence I'm going to check on an oscilloscope how the voltage looks and maybe also change the POWGOOD settings just like Sony did on the Frankie'd RSXs.

 

[NightLess]

I'm going to check the voltage on mine from an oscilloscope. I did manage to go even lower, at a set point of 1.0375V according to the VID table, but a real voltage of 0.988V. It was surprisingly stable with not as demanding games, but TLOU triggered a RLOD of 1001 (BE Power Fail). Hence I'm going to check on an oscilloscope how the voltage looks and maybe also change the POWGOOD settings just like Sony did on the Frankie'd RSXs.

Yes, the actual voltage is a bit lower, and the two I tested were not stable at 1.0v.

 

[villahed94]

I will provide scope measurements later but I think it's more likely the Power delivery system might be also finnicky.
The console I'm testing is a CECHA01 (older than the one from the photops on top) with a runtime of 182 days (!)
I did have to rescue this one with NEC TOKIN replacements for the RSX because the TOKINs were tarnished!

 

[NightLess]

I will provide scope measurements later but I think it's more likely the Power delivery system might be also finnicky.
The console I'm testing is a CECHA01 (older than the one from the photops on top) with a runtime of 182 days (!)
I did have to rescue this one with NEC TOKIN replacements because the TOKINs were tarnished!

I don't have an oscilloscope, but I have installed a Felix Tnatalizer on both machines, so the power supply system is probably OK? I'm testing my A+B Frankie, it has a stock voltage of 1.229v, and it even triggered a power failure during startup when I adjusted it to 1.0v. I've increased the voltage to 1.093v, and it seems that it needs to consume more voltage, but it runs cooler than my B00

 

[villahed94]

Yeah, it seems this one is due for TOKIN replacement on CELL as well because same 1001 error while loading TLOU even at 1.2V VID.
However, here are some scope measurements at 3 different points on XMB:
1.20V VID, 1.13V measured.

1.12V VID, 1.06V measured

1.05V VID, 0.99V measured

The ripple and pulses look very nasty, will replace at least 2 TOKINs for now.

 

[NightLess]

Yeah, it seems this one is due for TOKIN replacement on CELL as well because same 1001 error while loading TLOU even at 1.2V VID.
However, here are some scope measurements at 3 different points on XMB:
1.20V VID, 1.13V measured.

1.12V VID, 1.06V measured

1.05V VID, 0.99V measured

The ripple and pulses look very nasty, will replace at least 2 TOKINs for now.

Yes, this ripple is too high, you have to attack Tokin.

 

[LuanTeles]

@RIP-Felix