cha0shacker
Member
OC is always a lottery. For example, i have an early 21XX model (from 2010). OCs like arse
PS3 Project RSX Boost: Overclock your Retail PS3 RSX Speeds (ps3 cfw only)
RIP-Felix
Senior Member
Facinating! Yes, those are core voltage power faiulres, which very well could be caused by either failing tokins or increased noise/ripple from overclocking. I suspected this might be the case. And I think tantalizers would definately be able to handle more of it than the old tokins can.here you go, it is from my Debby
does this help in any way? have bricked it with 1000, 950, 900MHz memory clockCode:Firmware Version: 4.21 (build 49277) Platform ID: Deb01 Product Code: 00 81 Product Sub Code: 00 09 Hardware Config: 4610FFFF8107FFFF Syscon Fimware Version: 0E69.0001000400040001 (EEPROM: 0001000400040001) Bringup Count: 1387, Shutdown Count: 962 Runtime: 153 Days, 20 Hours, 8 Minutes, 34 Seconds Error Log 01: A0902120 Tue Sep 5 03:27:58 2006 02: A0801002 Tue Sep 5 03:27:58 2006 03: A0801002 Tue Sep 5 02:56:15 2006 04: A0801001 Tue Sep 5 02:53:07 2006 05: A0801002 Tue Sep 5 02:45:33 2006 06: A0801001 Tue Sep 5 02:44:29 2006 07: A0801001 Tue Sep 5 02:30:04 2006 08: A0801001 Tue Sep 5 02:27:44 2006 09: A0801001 Tue Sep 5 02:12:12 2006 10: A0801002 Tue Sep 5 02:08:33 2006 11: A0801001 Sun Sep 3 15:01:40 2006 12: A0801001 Sun Sep 3 14:59:48 2006 13: A0801001 Sun Sep 3 14:57:59 2006 14: A0801001 Sun Sep 3 14:56:16 2006 15: A0801001 Sun Sep 3 14:53:52 2006 16: A0801001 Fri May 19 13:06:04 2006 17: A0801001 Fri May 19 12:59:38 2006 18: A0801001 Fri May 19 12:57:56 2006 19: A0801001 Fri May 19 12:56:12 2006 20: A0801001 Fri May 19 12:55:03 2006 21: A0801001 Fri May 19 12:53:51 2006 22: A0801001 Fri May 19 12:52:57 2006 23: A0802022 Mon Feb 27 15:33:21 2006 24: A0802022 Mon Feb 27 15:25:06 2006 25: A0802022 Mon Feb 27 15:23:24 2006 26: A0802022 Mon Feb 27 15:21:04 2006 27: A0801001 Mon Feb 27 13:56:07 2006 28: A0801001 Mon Feb 20 01:26:18 2006 29: A0801001 Sun Feb 19 10:02:29 2006 30: A0802022 Fri Feb 17 10:59:40 2006 31: A0802022 Fri Feb 17 09:55:44 2006 32: FFFFFFFF Tue Feb 14 09:09:26 2006
last is reverted back to normal
@RIP-Felix
do you think the Tokins could be culprit on 65nm, they can't be pushed like 40nm?
@bguerville
will talk to you in pm
Very interesting indeed. As for the memory clock instability, I'm curious if it would produce a different error or also register as a core power failure. I'm not intentionally bricking a console to find out.
NunoRS2000
Member
I've been playing with the 700/900 overclock on my CECH 2504-A and in the games I tested, they showed amazing results. In Crysis final boss that normally would run a around 13 to 21fps in the base clock values was running at 26 to 31fps during all the fight with the overclock. Bayonetta seems to get a good enough boost to allow for those special dodges.
All looked as it should. No artifacts or any kind of problems. Everything is working nice and smooth.
Thanks for this amazing project
All looked as it should. No artifacts or any kind of problems. Everything is working nice and smooth.
Thanks for this amazing project
cha0shacker
Member
Bayonetta is still very heavily CPU bound, as is GTA IV for exampleI've been playing with the 700/900 overclock on my CECH 2504-A and in the games I tested, they showed amazing results. In Crysis final boss that normally would run a around 13 to 21fps in the base clock values was running at 26 to 31fps during all the fight with the overclock. Bayonetta seems to get a good enough boost to allow for those special dodges.
All looked as it should. No artifacts or any kind of problems. Everything is working nice and smooth.
Thanks for this amazing project
Guys I was very busy in the past couple days, it's nice to see new users trying OC.
Please can you post the ps3 model, rsx model dump from lv1( if you don't know how to find it, send my the dump via dm).
And the clock speeds thar worked and the ones that not.
here you go, it is from my Debby
does this help in any way? have bricked it with 1000, 950, 900MHz memory clockCode:Firmware Version: 4.21 (build 49277) Platform ID: Deb01 Product Code: 00 81 Product Sub Code: 00 09 Hardware Config: 4610FFFF8107FFFF Syscon Fimware Version: 0E69.0001000400040001 (EEPROM: 0001000400040001) Bringup Count: 1387, Shutdown Count: 962 Runtime: 153 Days, 20 Hours, 8 Minutes, 34 Seconds Error Log 01: A0902120 Tue Sep 5 03:27:58 2006 02: A0801002 Tue Sep 5 03:27:58 2006 03: A0801002 Tue Sep 5 02:56:15 2006 04: A0801001 Tue Sep 5 02:53:07 2006 05: A0801002 Tue Sep 5 02:45:33 2006 06: A0801001 Tue Sep 5 02:44:29 2006 07: A0801001 Tue Sep 5 02:30:04 2006 08: A0801001 Tue Sep 5 02:27:44 2006 09: A0801001 Tue Sep 5 02:12:12 2006 10: A0801002 Tue Sep 5 02:08:33 2006 11: A0801001 Sun Sep 3 15:01:40 2006 12: A0801001 Sun Sep 3 14:59:48 2006 13: A0801001 Sun Sep 3 14:57:59 2006 14: A0801001 Sun Sep 3 14:56:16 2006 15: A0801001 Sun Sep 3 14:53:52 2006 16: A0801001 Fri May 19 13:06:04 2006 17: A0801001 Fri May 19 12:59:38 2006 18: A0801001 Fri May 19 12:57:56 2006 19: A0801001 Fri May 19 12:56:12 2006 20: A0801001 Fri May 19 12:55:03 2006 21: A0801001 Fri May 19 12:53:51 2006 22: A0801001 Fri May 19 12:52:57 2006 23: A0802022 Mon Feb 27 15:33:21 2006 24: A0802022 Mon Feb 27 15:25:06 2006 25: A0802022 Mon Feb 27 15:23:24 2006 26: A0802022 Mon Feb 27 15:21:04 2006 27: A0801001 Mon Feb 27 13:56:07 2006 28: A0801001 Mon Feb 20 01:26:18 2006 29: A0801001 Sun Feb 19 10:02:29 2006 30: A0802022 Fri Feb 17 10:59:40 2006 31: A0802022 Fri Feb 17 09:55:44 2006 32: FFFFFFFF Tue Feb 14 09:09:26 2006
last is reverted back to normal
@RIP-Felix
do you think the Tokins could be culprit on 65nm, they can't be pushed like 40nm?
@bguerville
will talk to you in pm
Haxxxen, please post your model too and the rsx string.
Not yet, I was very busy, I'll build one with Haxxxen's instructions and let you know.
Last edited by a moderator:
haxxxen
Developer
I see, so must be some voltage instability. interesting that you can read from the errors. I still haven't looked into this Syscon development at all.
about the possible core failure, don't worry and no problem. I will try it as well, but I first wanted to check memory clock
@LuanTeles
it was my Debby, DECR-1400A DEB-001 with "rsx65 a03". about the serial I dunno, but maybe @vyktormvmpay25 can tell more?
vyktormvmpay25
Senior Member
I didn't test sb uart log to see entire log after fixed and assembly.
@bguerville do you think that bgtoolset could work in this model ?
We don't have any dump of IDCP-IP firmware.
Probably yes.. as long as you have a functional browser with the Flash player plugin, in theory there should be no particular obstacle to port offsets on any fw > 4.10, CEX, DEX or DECR.
Although the older the firmware the more chances there are that big differences in vsh.self imply having to replace one or more ROP gadgets..
Tanzu15
Member
Might have been just a normal crash. Happened to me even in stock.
Cool, it would be nice to dump that firmware.
Well, if you want me to start looking at it, it would require extracting vsh.self for starters, opening/analysing it in IDA 7.x with the updated self analysis plugin and sending me the ida database by pm.
That would be the first step...
After that, depending on what I see in the IDA db, there would be testing to do at your end..
Joonie
Developer
there should be 1 or 2 that are dumped. I did help one user to build a custom one out of regular dex ofw
dtcpipfw is pretty much identical to dex
@RIP-Felix Being someone who has worked for companies like Mushkin Memory being involved with R&D I can say that much like the larger scale production process the production process of the components themselves, like the RSX or CPU, can be moved to a new node but to minimise waste and cost the last of any old materials will still get used if the failure rate is calculated to be acceptable on the new node.
What's not very commonly known is that manufactures do have ways of knowing failure rates, the method varies depending on the part being made weather it be memory, CPU die, GPU die, etc, but they do know and they even know the clock frequency potential because manufacturers have access to the data on exactly where on the wafer the die has been cut from. For them the "silicon lottery" does not really exist and this process is known as "Binning".
To be clear here when I say manufacturer I'm talking about the companies involved in actually bringing a part into existence, so for example a company like Sapphire or Powercolor wouldn't automatically know the potential of the GPU or memory on their products because they are simple AIBs, they would have to request that information from AMD, who would ask the fab if they don't already have the information on the batches and where on each wafer each die was cut. Same for the memory AIBs would have to request from Hynix\Micron etc but they aren't guaranteed to be given the information on either of these counts.
There's a reason the 65nm RSX node was a short lived one for the PS3, it was overwhelmingly likely purely a cost saving move only, shrink the node to get more dies per wafer and use up the last of the unsuitable\old production materials. It's for the same reason why you see a smattering of NEC/TOKINS on the 20XX models but not the 21XX models which went into production only a handful of months later - 20XX models were just using up old components\materials. Not necessarily a bad thing, but is when the components being used aren't ideal. Engineers are rarely listened to and I know from experience the arguments gotten into over what the engineers want and what someone above them is saying for "costing down" (I'll assume you're familiar with that term). As long as the failure rate is calculated to be low enough it'll still be rolled out, that's just how manufacturing is done.
Ripple\noise is a funny thing, depending on circuit design such as where the traces are on the PCB, how long they have to be, number of layers the PCB has, voltage stability from the power source (and even mains socket), are the traces isolated on the PCB from crosstalk interference, etc, all have a large say on how much and the type of filtering you need and even how far away from the component you are trying to filter the filtering is taking place. For example a well laid out trace path that is isolated with filtering happening close to or right on top of the component you want to filter isn't going to require as much filtering as something with a long trace path. The specifications of the high and low side MOSFETs are more important here along with the VRM as a whole, assuming the level of filtering is at least adequate.
Multiple stage filtering can be advantageous if the power supply is providing dirty power (which will cause ripple on all of the voltage rails which will be passed on to everything in the system, an Oscilloscope is really the only way to see things accurately) but otherwise you shouldn't need excessive filtering, things like sound chips are different, but I digress. The more current the RSX draws the more likelyhood of transient spikes as the RSX goes from idle to active states which can lead to instabilities. You're better off just using a better power supply than bolting on a lot of filtering but a bit extra filtering would be interesting to do just to see how well Sony have designed things. It's for this very reason that with my 2503A slim I switched out the APS 270 for the 250, it's simply more capable which has numerous advantages;
- more stable power delivery
- shouldn't get as hot
- the extra amps (18 vs 16) will just help the console stay more stable when being pushed hard
- as it now is the extra amperage is useful for overclocking. Especially if someone figures out the CPU which must be taking power from the 12v rail.
You're also likely to run into more limitations if there aren't enough phases on the VRM. The cheap way around this when designing a VRM is to add phase doublers but when doing that you should really be using parts all from the same manufacturer (MOSFETs, doublers, voltage controller) and the design should be made with this in mind from the very start as otherwise at best you've given yourself an unnecessary headache getting mismatching components working even vaguely right and at worst you're going to throw everything out and make things worse.
If you're asking what needs to be looked at to determine OC viability on the PS3 the checklist would be;
- Voltage stability from the PSU and stability of the voltage being supplied to the RSX
- Number of phases for the RSX and does the design use phase doublers
- MOSFETs used
- Switching frequency of the gates, lower frequencies will keep the VRM cooler and less stressed but voltage will be more unstable. Testing would need to be done but assuming the VRM is capable of it a frequency around 625 is probably going to be close to the optimal area for voltage stability vs. heat output.
Right now on any PS3 slim model my recommendations for a long term reliable overclock would be 650MHz core 700MHz (1.4GHz effective) memory. There shouldn't be a single PS3 slim out there that can't handle these clocks taking into account the vastly superior node (65nm or 40nm) the RSX is made with compared to the G70s 110nm, and even there the reference G70 came with a clock of 430MHz. Using that as a (very) rough baseline to estimate critical points with each node shrink based on the core clocks Sony went with from the get-go; 90nm > 500MHz (+70MHz over G70) > 65nm +140MHz (theoretical safe core speed: 640MHz) > 40nm +210MHz (710MHz). These are very, very rough extrapolation numbers and make a number of assumptions such as Sony keeping the same RSX core voltage across node shrinks but it charts a rough path everybody should follow to minimise the risk of bricks.
Phew this took a long time to type, I've tried to cover everything I wanted without making anything too difficult for most people to be able to follow. Any questions from anyone just hit me with a message but the TLDR is that I need to know more about the power delivery of the PS3 and are the components used relatively consistent.
Side Note: If you can find the RSX voltage you can skew the VRM signal to it in software to pump a higher voltage but measured by the syscon (if it even bothers to do this) the VID reading would still be default as far as it is concerned only a realtime voltage reading would show the difference. How much voltage you could give the RSX would depend on what variant you have but if for example the default voltage is 1.2v, I wouldn't recommend more than 1.275v more than that thermal issues will probably start to arise but this isn't my post about thermodynamics 101 at this point :p
Side Note 2: It would require someone with the coding familiarity for PS3 to make it happen but there could be a way to resurrect "dead" PS3s from overzealous OCs. Insert some code, where or how would be down to those with the necessary familiarity with PS3, that looks for a .PUP from a storage device connected over USB before the RSX tries to be initialised. In theory this can be done before any hardware aside from the USB ports are initialised. You'd ideally specify in the code the filename to look for, something universal like "backup.pup" or something.
I've not seen this process, but I'm interested. Links?
The highest "compatible", and I use the term loosely as even doing it this way there is not a guarantee it would work because other factors are also at play, memory would be Samsung K4J52324QE - BJ1A, rated for 1Gbps @ 1.9v. IF this replacement works you're limited by the timings of the lower rated parts which will be tighter which is fine for raw efficiency per clock cycle but not in terms of optimal frequency. Ultimately though what you'd lose in frequency is made up for in equal measure by the higher bandwidth you'd get with tighter timings, it's not optimal, but maybe 95% of optimal. The last limiting factor would be voltage as the highest rated BJ1A is meant to wor at 1.9v any frequency you'd get out of them would be limited by the lower 1.8v.@ninku It's always possible there was data already corrupt on your HDD however with that memory frequency there is quite a high possibility that caused it. What a lot of people aren't aware of is that GDDR can have hundreds or even thousands of errors happening that are not produced as visual artifacts but other behaviour such as flat out crashes or lower performance vs. a lower clock. I'd lower your memory clock by 100MHz if I were you and do extended testing at 850 before nudging things higher again.
On a separate but related note does anyone have detailed data for the Delta KFB1012HE\KSB1012HE, Nidec G10C12MS2AH-56J14\G10C12MS1AH-56J14, and NMB-MAT BG1004-B045-P00? I ask because what would be interesting to know is which fan has the highest CFM while staying the quietest. If I find anything in the meantime I'll update this post.
Last edited by a moderator:
cha0shacker
Member
I need to add a little. Nvidia themselves made G70s on 90nm, the G71. It comes in up to 650MHz core (7900 GTX).@RIP-Felix Being someone who has worked for companies like Mushkin Memory being involved with R&D I can say that much like the larger scale production process the production process of the components themselves, like the RSX or CPU, can be moved to a new node but to minimise waste and cost the last of any old materials will still get used if the failure rate is calculated to be acceptable on the new node.
What's not very commonly known is that manufactures do have ways of knowing failure rates, the method varies depending on the part being made weather it be memory, CPU die, GPU die, etc, but they do know and they even know the clock frequency potential because manufacturers have access to the data on exactly where on the wafer the die has been cut from. For them the "silicon lottery" does not really exist and this process is known as "Binning".
To be clear here when I say manufacturer I'm talking about the companies involved in actually bringing a part into existence, so for example a company like Sapphire or Powercolor wouldn't automatically know the potential of the GPU or memory on their products because they are simple AIBs, they would have to request that information from AMD, who would ask the fab if they don't already have the information on the batches and where on each wafer each die was cut. Same for the memory AIBs would have to request from Hynix\Micron etc but they aren't guaranteed to be given the information on either of these counts.
There's a reason the 65nm RSX node was a short lived one for the PS3, it was overwhelmingly likely purely a cost saving move only, shrink the node to get more dies per wafer and use up the last of the unsuitable\old production materials. It's for the same reason why you see a smattering of NEC/TOKINS on the 20XX models but not the 21XX models which went into production only a handful of months later - 20XX models were just using up old components\materials. Not necessarily a bad thing, but is when the components being used aren't ideal. Engineers are rarely listened to and I know from experience the arguments gotten into over what the engineers want and what someone above them is saying for "costing down" (I'll assume you're familiar with that term). As long as the failure rate is calculated to be low enough it'll still be rolled out, that's just how manufacturing is done.
Ripple\noise is a funny thing, depending on circuit design such as where the traces are on the PCB, how long they have to be, number of layers the PCB has, voltage stability from the power source (and even mains socket), are the traces isolated on the PCB from crosstalk interference, etc, all have a large say on how much and the type of filtering you need and even how far away from the component you are trying to filter the filtering is taking place. For example a well laid out trace path that is isolated with filtering happening close to or right on top of the component you want to filter isn't going to require as much filtering as something with a long trace path. The specifications of the high and low side MOSFETs are more important here along with the VRM as a whole, assuming the level of filtering is at least adequate.
Multiple stage filtering can be advantageous if the power supply is providing dirty power (which will cause ripple on all of the voltage rails which will be passed on to everything in the system, an Oscilloscope is really the only way to see things accurately) but otherwise you shouldn't need excessive filtering, things like sound chips are different, but I digress. The more current the RSX draws the more likelyhood of transient spikes as the RSX goes from idle to active states which can lead to instabilities. You're better off just using a better power supply than bolting on a lot of filtering but a bit extra filtering would be interesting to do just to see how well Sony have designed things. It's for this very reason that with my 2503A slim I switched out the APS 270 for the 250, it's simply more capable which has numerous advantages;
- more stable power delivery
- shouldn't get as hot
- the extra amps (18 vs 16) will just help the console stay more stable when being pushed hard
- as it now is the extra amperage is useful for overclocking. Especially if someone figures out the CPU which must be taking power from the 12v rail.
You're also likely to run into more limitations if there aren't enough phases on the VRM. The cheap way around this when designing a VRM is to add phase doublers but when doing that you should really be using parts all from the same manufacturer (MOSFETs, doublers, voltage controller) and the design should be made with this in mind from the very start as otherwise at best you've given yourself an unnecessary headache getting mismatching components working even vaguely right and at worst you're going to throw everything out and make things worse.
If you're asking what needs to be looked at to determine OC viability on the PS3 the checklist would be;
- Voltage stability from the PSU and stability of the voltage being supplied to the RSX
- Number of phases for the RSX and does the design use phase doublers
- MOSFETs used
- Switching frequency of the gates, lower frequencies will keep the VRM cooler and less stressed but voltage will be more unstable. Testing would need to be done but assuming the VRM is capable of it a frequency around 625 is probably going to be close to the optimal area for voltage stability vs. heat output.
Right now on any PS3 slim model my recommendations for a long term reliable overclock would be 650MHz core 700MHz (1.4GHz effective) memory. There shouldn't be a single PS3 slim out there that can't handle these clocks taking into account the vastly superior node (65nm or 40nm) the RSX is made with compared to the G70s 110nm, and even there the reference G70 came with a clock of 430MHz. Using that as a (very) rough baseline to estimate critical points with each node shrink based on the core clocks Sony went with from the get-go; 90nm > 500MHz (+70MHz over G70) > 65nm +140MHz (theoretical safe core speed: 640MHz) > 40nm +210MHz (710MHz). These are very, very rough extrapolation numbers and make a number of assumptions such as Sony keeping the same RSX core voltage across node shrinks but it charts a rough path everybody should follow to minimise the risk of bricks.
Phew this took a long time to type, I've tried to cover everything I wanted without making anything too difficult for most people to be able to follow. Any questions from anyone just hit me with a message but the TLDR is that I need to know more about the power delivery of the PS3 and are the components used relatively consistent.
Side Note: If you can find the RSX voltage you can skew the VRM signal to it in software to pump a higher voltage but measured by the syscon (if it even bothers to do this) the VID reading would still be default as far as it is concerned only a realtime voltage reading would show the difference. How much voltage you could give the RSX would depend on what variant you have but if for example the default voltage is 1.2v, I wouldn't recommend more than 1.275v more than that thermal issues will probably start to arise but this isn't my post about thermodynamics 101 at this point :p
Side Note 2: It would require someone with the coding familiarity for PS3 to make it happen but there could be a way to resurrect "dead" PS3s from overzealous OCs. Insert some code, where or how would be down to those with the necessary familiarity with PS3, that looks for a .PUP from a storage device connected over USB before the RSX tries to be initialised. In theory this can be done before any hardware aside from the USB ports are initialised. You'd ideally specify in the code the filename to look for, something universal like "backup.pup" or something.
I've not seen this process, but I'm interested. Links?
@ninku It's always possible there was data already corrupt on your HDD however with that memory frequency there is quite a high possibility that caused it. What a lot of people aren't aware of is that GDDR can have hundreds or even thousands of errors happening that are not produced as visual artifacts but other behaviour such as flat out crashes or lower performance vs. a lower clock. I'd lower your memory clock by 100MHz if I were you and do extended testing at 850 before nudging things higher again.
Also since the RSX doesn't do power saving the switches between idle and full load shouldn't be as drastic on the VRM.
Tanzu15
Member
Bro this was an entire damn book and I read the whole thing word by word. Very insightful. My PS3 is fully stable at 750/950. Every game I test has zero issues and just pure good performance. Zero crashes and hours none stop of playing games. But that power supply bit peaked my interest. Is there a way to know which power supply I have without opening the console?@RIP-Felix Being someone who has worked for companies like Mushkin Memory being involved with R&D I can say that much like the larger scale production process the production process of the components themselves, like the RSX or CPU, can be moved to a new node but to minimise waste and cost the last of any old materials will still get used if the failure rate is calculated to be acceptable on the new node.
What's not very commonly known is that manufactures do have ways of knowing failure rates, the method varies depending on the part being made weather it be memory, CPU die, GPU die, etc, but they do know and they even know the clock frequency potential because manufacturers have access to the data on exactly where on the wafer the die has been cut from. For them the "silicon lottery" does not really exist and this process is known as "Binning".
To be clear here when I say manufacturer I'm talking about the companies involved in actually bringing a part into existence, so for example a company like Sapphire or Powercolor wouldn't automatically know the potential of the GPU or memory on their products because they are simple AIBs, they would have to request that information from AMD, who would ask the fab if they don't already have the information on the batches and where on each wafer each die was cut. Same for the memory AIBs would have to request from Hynix\Micron etc but they aren't guaranteed to be given the information on either of these counts.
There's a reason the 65nm RSX node was a short lived one for the PS3, it was overwhelmingly likely purely a cost saving move only, shrink the node to get more dies per wafer and use up the last of the unsuitable\old production materials. It's for the same reason why you see a smattering of NEC/TOKINS on the 20XX models but not the 21XX models which went into production only a handful of months later - 20XX models were just using up old components\materials. Not necessarily a bad thing, but is when the components being used aren't ideal. Engineers are rarely listened to and I know from experience the arguments gotten into over what the engineers want and what someone above them is saying for "costing down" (I'll assume you're familiar with that term). As long as the failure rate is calculated to be low enough it'll still be rolled out, that's just how manufacturing is done.
Ripple\noise is a funny thing, depending on circuit design such as where the traces are on the PCB, how long they have to be, number of layers the PCB has, voltage stability from the power source (and even mains socket), are the traces isolated on the PCB from crosstalk interference, etc, all have a large say on how much and the type of filtering you need and even how far away from the component you are trying to filter the filtering is taking place. For example a well laid out trace path that is isolated with filtering happening close to or right on top of the component you want to filter isn't going to require as much filtering as something with a long trace path. The specifications of the high and low side MOSFETs are more important here along with the VRM as a whole, assuming the level of filtering is at least adequate.
Multiple stage filtering can be advantageous if the power supply is providing dirty power (which will cause ripple on all of the voltage rails which will be passed on to everything in the system, an Oscilloscope is really the only way to see things accurately) but otherwise you shouldn't need excessive filtering, things like sound chips are different, but I digress. The more current the RSX draws the more likelyhood of transient spikes as the RSX goes from idle to active states which can lead to instabilities. You're better off just using a better power supply than bolting on a lot of filtering but a bit extra filtering would be interesting to do just to see how well Sony have designed things. It's for this very reason that with my 2503A slim I switched out the APS 270 for the 250, it's simply more capable which has numerous advantages;
- more stable power delivery
- shouldn't get as hot
- the extra amps (18 vs 16) will just help the console stay more stable when being pushed hard
- as it now is the extra amperage is useful for overclocking. Especially if someone figures out the CPU which must be taking power from the 12v rail.
You're also likely to run into more limitations if there aren't enough phases on the VRM. The cheap way around this when designing a VRM is to add phase doublers but when doing that you should really be using parts all from the same manufacturer (MOSFETs, doublers, voltage controller) and the design should be made with this in mind from the very start as otherwise at best you've given yourself an unnecessary headache getting mismatching components working even vaguely right and at worst you're going to throw everything out and make things worse.
If you're asking what needs to be looked at to determine OC viability on the PS3 the checklist would be;
- Voltage stability from the PSU and stability of the voltage being supplied to the RSX
- Number of phases for the RSX and does the design use phase doublers
- MOSFETs used
- Switching frequency of the gates, lower frequencies will keep the VRM cooler and less stressed but voltage will be more unstable. Testing would need to be done but assuming the VRM is capable of it a frequency around 625 is probably going to be close to the optimal area for voltage stability vs. heat output.
Right now on any PS3 slim model my recommendations for a long term reliable overclock would be 650MHz core 700MHz (1.4GHz effective) memory. There shouldn't be a single PS3 slim out there that can't handle these clocks taking into account the vastly superior node (65nm or 40nm) the RSX is made with compared to the G70s 110nm, and even there the reference G70 came with a clock of 430MHz. Using that as a (very) rough baseline to estimate critical points with each node shrink based on the core clocks Sony went with from the get-go; 90nm > 500MHz (+70MHz over G70) > 65nm +140MHz (theoretical safe core speed: 640MHz) > 40nm +210MHz (710MHz). These are very, very rough extrapolation numbers and make a number of assumptions such as Sony keeping the same RSX core voltage across node shrinks but it charts a rough path everybody should follow to minimise the risk of bricks.
Phew this took a long time to type, I've tried to cover everything I wanted without making anything too difficult for most people to be able to follow. Any questions from anyone just hit me with a message but the TLDR is that I need to know more about the power delivery of the PS3 and are the components used relatively consistent.
Side Note: If you can find the RSX voltage you can skew the VRM signal to it in software to pump a higher voltage but measured by the syscon (if it even bothers to do this) the VID reading would still be default as far as it is concerned only a realtime voltage reading would show the difference. How much voltage you could give the RSX would depend on what variant you have but if for example the default voltage is 1.2v, I wouldn't recommend more than 1.275v more than that thermal issues will probably start to arise but this isn't my post about thermodynamics 101 at this point :p
Side Note 2: It would require someone with the coding familiarity for PS3 to make it happen but there could be a way to resurrect "dead" PS3s from overzealous OCs. Insert some code, where or how would be down to those with the necessary familiarity with PS3, that looks for a .PUP from a storage device connected over USB before the RSX tries to be initialised. In theory this can be done before any hardware aside from the USB ports are initialised. You'd ideally specify in the code the filename to look for, something universal like "backup.pup" or something.
I've not seen this process, but I'm interested. Links?
@ninku It's always possible there was data already corrupt on your HDD however with that memory frequency there is quite a high possibility that caused it. What a lot of people aren't aware of is that GDDR can have hundreds or even thousands of errors happening that are not produced as visual artifacts but other behaviour such as flat out crashes or lower performance vs. a lower clock. I'd lower your memory clock by 100MHz if I were you and do extended testing at 850 before nudging things higher again.
Mitsu™
Senior Member
That's funny. I was just searching informations about this. My two CECH-2004 slims have different models :
Nidec G10C12MS1AH-56J14 and Delta KFB1012HE.
So I'm trying to figure which one is the best to use in my working Slim.
Some (rare) users report that the Delta model is quieter while cooling the same as the Nidec one.
I will try it and see.
@cha0shacker The RSX is a modified G70, or 71, they are both the same thing mostly, however G70\71 are not identical to the RSX so you can't (and shouldn't) make like for like comparisons. When the RSX began development the architecture it was based upon and modified from was the 7800GTX so as that is the closest fit for the RSX from a design sense that is the baseline we must use. Beyond that it has to be considered that the VRM in the PS3 is not going to be anywhere near as robust as that on a 7800GTX so expectations need to be tempered accordingly a less robust VRM means OC potential will be lower and the VRM will be more likely to give out if pushed too hard. As I said for a 65nm or 40nm RSX 650MHz on the core is a realistic frequency all slims should be able to handle but it can't be forgotten that the RSX is not a G70\71, it is merely based upon it so you must treat the RSX as an RSX, not a GTX
There will be plenty of PS3s that can of course handle above 650MHz core speed but what we are aiming for here is a maximum universally safe frequency for all slim model PS3s which is why I arrived at the figures of 650\700 because you have to consider not all PS3s are going to have core\memory that clocks well and not all PS3s (in fact, definitely are not) using the same Samsung memory some consoles are equipped with memory rated for more than 700MHz so the best thing to do is to set clocks at what should be universally attainable and anything higher than that is at the users risk. I'm thinking of saving as many peoples PS3s from becoming paperweights as possible.@Tanzu15 Nope, not that I know of.
@Mitsu™ I have a few different fans including the Nidec G10C12MS1AH-56J14, from what I've used of it the fan is super quiet, and in theory it should be having more fan blades, but I wonder about what CFM it truly moves as it's too quiet IMO, not like my other fans which are both Delta if I recall correctly. I find it a little strange too that Sony used fans that varied so wildly in terms of how many amps they use and number of fan blades. The Deltas and Nidecs use 15 and 17 respectively, while the NMB uses 24. On paper the NMB is probably the best fan, most blades at an amperage of 1.5, that's somewhere between what most of the Deltas and Nidecs use.
Last edited:
Similar threads
-
PS3 [Guide] Sourcing the right PS3 for the best Overclocking performance
- Started by M4N14C
- Replies: 1
-
PS3 {CFW Release} 4.92.2 Evilnat Custom Firmware (w/ Cobra v8.5) - Includes: CEX/PEX/D-PEX) + noBD/BT- Started by Evilnat
- Replies: 1
-
-
PS3 PS3HEN w/ Linux & Overclocking Support? BadHTAB A hardware-software based hypervisor (lv1) exploit- Started by STLcardsWS
- Replies: 52