PS3 Project RSX Boost: Overclock your Retail PS3 RSX Speeds (ps3 cfw only)

[RIP-Felix]

There are some general "rules", even if finding out your console's limits yourself is always better and somewhat fun :
• Don't OC a FAT more than 600/750. Personally, I wouldn't OC a 90nm RSX at all.
• For a 20XX slim, 650/800-850 is the maximum you should attempt. 700MHz core is likely to cause issues ( due to the NEC/Tokins capacitors.
Note that if your NEC/Tokins are particularly degraded, even 600MHz core might be too much.
• For a 21XX, 700/850-900 is possible, 650/850 is guaranteed.
• Starting from the 25XX series, we're talking about RSX revisions because there are 4 known models with different limits.
- "CXD5300A1GB" : Basically the same chip found in a 21XX, so 700/900
- "CXD5300CGB" : 800/975 is possible, even 850/1000 if you manage to keep your temps below 70°C (or it might freeze in some demanding games).
- "CXD5300DGB" : Okay, this one is a bit mysterious. We don't have many reports about it, it seems to perform like a "A1GB".
- "CXD5300GGB" : If you like big numbers, this one if for you ! The latest RSX revision you can find in a CFW-compatible PS3. 850MHz core minimum guaranteed, 1000MHz VRAM should be easy for it too.
900MHz core is possible, 950MHz too but it's super rare.

That's all !

Now about the SuperSlims... I'm curious too :p
Even if one day we'd be able to fully jailbreak them, maybe it won't be as incredible as one could think. It's a possibility, sure, but it all depends on Sony !
These models are cool, but they're also super cheap in many ways since Sony went into "money saving mode : MAX", so... I don't know about their potential.

Also, I recommend you to check these posts about diminishing returns
• https://www.psx-place.com/threads/p...peeds-ps3-cfw-only.36801/page-133#post-394237
• https://www.psx-place.com/threads/p...peeds-ps3-cfw-only.36801/page-119#post-392382

DISCLAIMER!

WE DO NOT HAVE ENOUGH DATA to make these generalizations and there are NO GUARINTEES!

When making statments like those it's important to state the limitations of your data. You mention none. It could be based off 1 console you had work, or 10. Either case isnt enough to establish a reliable average. And even if you had an accurate mean, that doesn't tell you about the distrobution or outliers. A Guarintee means 100% of all consoles with, say a -GGB, will do [insert OC]. That right there is doomed to be a wrong statment some small percentage of the time, even if it's extreemly conservative.

Do not cater to lazyness. 600/750 is still the closest to a safe OC there is and should be WHERE YOU START. If you want to start higher because you're impatient, DO NOT OVERCLOCK.

At 500MHz the SPE has basically time for 6 cycles to do its work for the RSX (Because the SPEs run at 3200MHz). When you start to OC to to 550MHz or higher suddenly only 5 cycles fit and the GPU stalls hard because it doesn't get the information in time anymore that it is coded to wait for.

Help me underatand something. Early poor implimentations barely used SPEs if at all, so in that scenario this is a mute point. But doesn't the RSX have to process a frame, however long that takes? Ideally this should take less than one frame. 16.7ms for 60FPS target. If it can process and deliver the frame in less than that time, it sits idle until more data is fed to it for processing. If it needs the Cell to do certain processing functions before it can complete yhe current frame, I dont see how it being able to process and deliver it's part of the workload faster would have any effect other than reducing the time it takes. The cell just gets the instruction sooner and it takes as long as the cell always takes. But when it's done and feeds the data back to the RSX, the RSX's faster Clock speeds can finish the frame faster at that point.

This is probably an overly simplistic explanation. Maybe you can fill the gaps in my understanding.

 

[cha0shacker]

DISCLAIMER!

WE DO NOT HAVE ENOUGH DATA to make these generalizations and there are NO GUARINTEES!

When making statments like those it's important to state the limitations of your data. You mention none. It could be based off 1 console you had work, or 10. Either case isnt enough to establish a reliable average. And even if you had an accurate mean, that doesn't tell you about the distrobution or outliers. A Guarintee means 100% of all consoles with, say a -GGB, will do [insert OC]. That right there is doomed to be a wrong statment some small percentage of the time, even if it's extreemly conservative.

Do not cater to lazyness. 600/750 is still the closest to a safe OC there is and should be WHERE YOU START. If you want to start higher because you're impatient, DO NOT OVERCLOCK.


Help me underatand something. Early poor implimentations barely used SPEs if at all, so in that scenario this is a mute point. But doesn't the RSX have to process a frame, however long that takes? Ideally this should take less than one frame. 16.7ms for 60FPS target. If it can process and deliver the frame in less than that time, it sits idle until more data is fed to it for processing. If it needs the Cell to do certain processing functions before it can complete yhe current frame, I dont see how it being able to process and deliver it's part of the workload faster would have any effect other than reducing the time it takes. The cell just gets the instruction sooner and it takes as long as the cell always takes. But when it's done and feeds the data back to the RSX, the RSX's faster Clock speeds can finish the frame faster at that point.

This is probably an overly simplistic explanation. Maybe you can fill the gaps in my understanding.

It has to do with how you synchronize the data processing between the GPU and the SPUs, you can do that by polling, by cycle counting or by interrupt.
On the PC side whats happening in GTHD is called a race condition. Basically code runs too fast and that results in a crash.
As the RSX is ready to transfer the data before the SPE expects it and doesn't try to transfer the data again. It does so after a certain amount of cycles.

The PS3 is a pretty manual machine in terms of programming, it basically doesn't do anything automatically unless specificly told to. It makes it a dream for people trying to optimize everything to the last cycle but pretty cumbersome for "normal" devs. My favorite example is a Hello World program running on an SPE. It takes around 170 lines of code to do that due to having to manually transfer the code to the SPE, run it and then manually retrieve the result later to finally display it on screen.

And SPE usage is not that low in GTHD, probably due to it being an exclusive game. Japanse exclusives were pretty fast on SPE utilization even in the early days.
Just the synchronization is pretty crude in this case.

 

[Mitsu™]

Sir Mitsu, I am currently using 700 850 and it is very stable. I tested it for a long time, thank you for everything. My question now is, I want to know the limits. However, if I increase it by 50, 50, 50, for example, now it is 700, 750, and then 800, 900, will the device be bricked? Or can I recover it with minor corruption and return to the old file?

It depends on how thorough your testing is. If you ensure that the overclock is stable, run extended tests, and pay close attention to any signs of instability, you should be fine. Keep in mind, though, that the VRAM is more likely to cause a brick when jumping from one frequency to another without warning compared to the core. For example, I tested 900MHz on the core with my 2504 (CXD5300CGB), and it crashed within seconds once in-game. I just reinstalled the 850MHz core, and everything went back to normal. Also, don't bite off more than you can chew, as they say.

Note that it's recommended to have an e3flasher on hand in case an unfortunate hard brick occurs

Unrelated, but my first attempt at delidding an RSX is... successful !
That's my dead 2004B. A LOT of patience is required, but I didn't use any force (or almost none).
Just sawing the glue slowly with a very thin paper cutter, some isopropyl alcohol and heating the IHS moderately. Not a single scratch.
Very satisfied with the result !
Now I need to craft a tool for the CELL, just a matter of time. I hope that it'll work just as well ^^'

I'll add the RSX string later, when I'll have the time.
I probably still have it somewhere.
It could support 700MHz core. Don't remember the VRAM frequency.

 

[RIP-Felix]

Unrelated, but my first attempt in delidding an RSX is... successful !
That's my dead 2004B. A LOT of patience is required, but I didn't use any force (or almost none).
Just sawing the glue slowly with a very thin paper cutter, some isopropyl alcohol and heating the IHS moderately. Not a single scratch.
Very satisfied with the result !
Now I need to craft a tool for the CELL, just a matter of time. I hope that it'll work just as well ^^'
View attachment 44133

Beware that 90nm and 65nm RSX have softer thermal plaster that pops off easily with the heat/lever method. The 40nm on 21xx and later are the issue.

Where'd you get that tool? What's it called? You called it a paper cutter, but it looks like a scalpel with a bent blade. Looks like it might work for the CPU.

 

[Mitsu™]

Beware that 90nm and 65nm RSX have softer thermal plaster that pops off easily with the heat/lever method. The 40nm on 21xx and later are the issue.

Where'd you get that tool? What's it called? You called it a paper cutter, but it looks like a scalpel with a bent blade. Looks like it might work for the CPU.

I'll be very VERY careful if I ever need to delid a 40nm RSX. I'm not using the lever method though.

It's a paper knife (that's probably what I should have called it instead of 'paper cutter' ^^') that I've bent a little (about 2 cm). Unfortunately, it's still too thick for the CELL. I need to work on it.

 

[RIP-Felix]

Letter opener? "Paper knife" doesn't yield any products similar to that. Just exacto knifes and box cutters.

 

[Mitsu™]

Letter opener? "Paper knife" doesn't yield any products similar to that. Just exacto knifes and box cutters.

Mmh, maybe. In French, we call it a "coupe-papier," and yes, it's just a kind of knife used for opening letters.

You were right, by the way ! It's perfect for the CELL too ! After struggling to sand down a small spatula for hours without managing to make it thin enough to fit correctly beneath the IHS (it even left a small superficial scratch ^^'), I tried gently sawing the corners with this knife and succeeded !

I also used some heat and "Essence F" (you can see it in the picture, I don't know how to translate it) to soften the silicone. This "Essence F" seems to be quite effective without attacking the PCB like something such as acetone would.

It was relatively easy and took even less time than the RSX. Now I need to try this on a working console, though not my main Slim. Since it's a 2504A, I probably won't ever touch the CELL on it (except as a last resort, eventually). There are just too many examples of destroyed ones.

 

[RIP-Felix]

So what you see in the corners is an issue. You want to start in the gap where there are not traces to damage and ride on a coushion of silicone as you cut around the outside. Sawing should be avoided. It should be able to cut on it's own. The blade should be tapered so it cuts upward against the IHS insted of down toward the mask.

 

[Mitsu™]

So what you see in the corners is an issue. You want to start in the gap where there are not traces to damage and ride on a coushion of silicone as you cut around the outside. Sawing should be avoided. It should be able to cut on it's own. The blade should be tapered so it cuts upward against the IHS insted of down toward the mask.

I know you're right, this is the safest and most proper way to do it, but in fact, it wasn't cutting against the mask. The silicone was removed from the corners because the 'Essence F' (naphta ?) + heat really softened it to the point where I could easily remove it with my nails. I was literally cutting through it like butter. The knife didn't scratch anything; the thin spatula did, but not the knife. I really couldn't get the thin spatula into the gap, which is why I tried another method.

EDIT: A second CELL easily delidded this way, the RSX too. No scratches at all. I think I'll stick to this method.
Unfortunately, I can't test it finally. It died for another reason xD
It was already dying when I got it, that's why I bought a 2504A.
I need to find some dead 25XX MoBo to practice.


EDIT2 : Even a 360 S easily and perfectly delidded thanks to this knife !
And this time, I'm 100% sure it's a success.

 

[quake]

I can try, it's not that easy to explain.
The way the work distribution in all GT games on PS3 is done is in lock step mode.
Basically a certain amount of clock cycles on the GPU/CPU is counted to measure when the SPEs have to do a certain task for the game that then has to be processed by the GPU and then output to the screen.

At 500MHz the SPE has basically time for 6 cycles to do its work for the RSX (Because the SPEs run at 3200MHz). When you start to OC to to 550MHz or higher suddenly only 5 cycles fit and the GPU stalls hard because it doesn't get the information in time anymore that it is coded to wait for.
That's also why it isn't a complete crash, only that the system is very very unbearably slow.

Later GT games thankfully used additional measures to ensure synchronization.

GTHD is just an early title which are usually only very crudely adapted to the target hardware. And it's also an example why later console versions don't just change their parts for faster ones or run the original ones at a higher clock. It will lead to compatibility problems which a console manufacturer will want to avoid at all costs. Nobody wants a second PS2 situation.

That's really interesting, but why does the problem only occur when pressing the ps button to open xmb in-game, and not ever during gameplay?

Any other known titles that have done this?

 

[Dhmhtrhs]

Hi guys should i be worried with this error codes or not?

 

[cha0shacker]

That's really interesting, but why does the problem only occur when pressing the ps button to open xmb in-game, and not ever during gameplay?

Any other known titles that have done this?

I can only guess but I think it's because the XMB overlay generates an interrupt and the whole thing falls appart.

 

[RIP-Felix]

Hi guys should i be worried with this error codes or not?

Possably, the 2111 isn't a great code. It's possable that errorlog is from normal unexpected/improper shutdows. People using the rocker instead of a gracefull shutdown or power outtages. The 1001 and 1004s can ovvur that way. But if you observe the YLOD, it shuts off during a stressful game for example, then the 1001's mean bad CPU tokins.

The 2111 is what complicates this log. Perhaps the console is full of dust amd in need of cleaning/repaste. That error has been seen (rare) in the case of overheating, but isn't diagnostic on it's own and it could indicate a more serious issue behind the scenes.

What model console, MB revision, and other details about the console can you offer. Has it been opened? What work was done? Was it sealed still? Any flux residues around the CPU/GPU...That sort of thing.

 

[Dhmhtrhs]

Possably, the 2111 isn't a great code. It's possable that errorlog is from normal unexpected/improper shutdows. People using the rocker instead of a gracefull shutdown or power outtages. The 1001 and 1004s can ovvur that way. But if you observe the YLOD, it shuts off during a stressful game for example, then the 1001's mean bad CPU tokins.

The 2111 is what complicates this log. Perhaps the console is full of dust amd in need of cleaning/repaste. That error has been seen (rare) in the case of overheating, but isn't diagnostic on it's own and it could indicate a more serious issue behind the scenes.

What model console, MB revision, and other details about the console can you offer. Has it been opened? What work was done? Was it sealed still? Any flux residues around the CPU/GPU...That sort of thing.

Iam going to be straight forward... I wanted a backwards compatible ps3 ceach04 60 gb after some mouths it was super dirty from the inside you could se the dust trough the small air vents etc after i tried opening it for the first time no cfw install or anything everything seems to be going normal after a few minutes the fan stared going higher and higher i don't know how much the fan speed reach without evilnut install but i knew it was somewhere between 65-70 I've never heard this kind of loudness in my life after some day i actually thought why won't i take it out of this cabin and place it somewhere where this thing can breath the thing dropped allot iam with 50% fan speed and the max it can reach on intensive games is 64-65C from 75C etc tonight i thought why not clean it for the first time it toke me 4 hours to open it dissemble it I've reach where the fan its and i give up lol I wanted to see the gpu and gpu if the thermal paste is good i could but after i put my fingers on the mid of it push it a little bit i shaw that it had some still there but dried i said meh and just closed it back because i didn't have any thermal paste anyways but before i unscrewed it where you can se the nek tokins at the middle where that aluminium thing hold it i shaw some kind of black tape and i thought of it that is some kind of thermal pad or something to lower the temperature? So someone probably replace the tokins or just did a cheap thing so is going to last i actually don't have any problem with it I've tried bioshock infinite Prince of Persia 2008 sonic generations prototype 2 all of then 2 hours no problem no crash no nothing you think this errors aren't begin right maybe? 2300 days and 73 inappropriate shut downs have been diagnosed

 

[RIP-Felix]

... kind of black tape and i thought of it that is some kind of thermal pad or something to lower the temperature? So someone probably replace the tokins or just did a cheap thing so is going to last I actually don't have any problem with it...

Sometimes there's this black tape on the tokins. That's OEM. I dont think it does much. Not sure what their point was with it, but it appeaes they don't always install it. Pay it no attention.

Sounds to me like you may have been having thermal monitor related overheating issues from the console being caked in dust and possably it's in need of a delid. However, It was hard to understand your response. Sounds like you didn't replace the paste on the heatsink/IHS interface? Or you did and now it seems fine?

If the cleaning has made the console stable, then you may have lucked out. Keep an eye on it.

 

[Dhmhtrhs]

Sometimes there's this black tape on the tokins. That's OEM. I dont think it does much. Not sure what their point was with it, but it appeaes they don't always install it. Pay it no attention.

Sounds to me like you may have been having thermal monitor related overheating issues from the console being caked in dust and possably it's in need of a delid. However, It was hard to understand your response. Sounds like you didn't replace the paste on the heatsink/IHS interface? Or you did and now it seems fine?

If the cleaning has made the console stable, then you may have lucked out. Keep an eye on it.

Yes i didn't replace the paste but it seems stable edit just got my first problem 1 time i tried opening the console manually with my hand it turns on and immediately the red light keep flashing after 2 attemp it turns on no problem weird?

 

[Dhmhtrhs]

And i wonderd why my ps3 was so loud -.-

 

[cha0shacker]

Iam going to be straight forward... I wanted a backwards compatible ps3 ceach04 60 gb after some mouths it was super dirty from the inside you could se the dust trough the small air vents etc after i tried opening it for the first time no cfw install or anything everything seems to be going normal after a few minutes the fan stared going higher and higher i don't know how much the fan speed reach without evilnut install but i knew it was somewhere between 65-70 I've never heard this kind of loudness in my life after some day i actually thought why won't i take it out of this cabin and place it somewhere where this thing can breath the thing dropped allot iam with 50% fan speed and the max it can reach on intensive games is 64-65C from 75C etc tonight i thought why not clean it for the first time it toke me 4 hours to open it dissemble it I've reach where the fan its and i give up lol I wanted to see the gpu and gpu if the thermal paste is good i could but after i put my fingers on the mid of it push it a little bit i shaw that it had some still there but dried i said meh and just closed it back because i didn't have any thermal paste anyways but before i unscrewed it where you can se the nek tokins at the middle where that aluminium thing hold it i shaw some kind of black tape and i thought of it that is some kind of thermal pad or something to lower the temperature? So someone probably replace the tokins or just did a cheap thing so is going to last i actually don't have any problem with it I've tried bioshock infinite Prince of Persia 2008 sonic generations prototype 2 all of then 2 hours no problem no crash no nothing you think this errors aren't begin right maybe? 2300 days and 73 inappropriate shut downs have been diagnosed

2300 days? thats 6 years and 4 months constant runtime. Was it used in a compute cluster before?.
My CECHA has 67 days.

But 73 inappropriate shutdowns is a really low number for 2300 days total runtime

 

[Mitsu™]

great thanks a lot Mitsu lets hope that maybe it can make a difference on the OC tests that you will make to check on that one

Okay I'm super late, but I tried to compare the temperature between CEX and PEX and it doesn't seems to change anything on my 2504A.

On XMB after an hour : 53°C CELL / 55C RSX at 32% fan speed with both CEX and PEX at the same frequencies (700/850) and ~22°C in my room.

In Crysis HD, at the beginning of the 4th mission after an hour : 61°C CELL / 69°C RSX at 32% fan speed with both CEX and PEX at the same frequencies (700/850). ~22°C in my room.

Still, my console's results aren't representative of all PS3s around. I'm really curious. I wonder what could be causing or explaining the difference you pointed out.

 

[LuanTeles]

@Mitsu™ can you make a 850/1025 EvilNat PeX Beta 10 ?