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

[haxxxen]

btw, I already have once told this on this thread, you can also brick (no hardbrick=no boot) with underclocking. this stupid piece of hardware does not like to be pushing around

 

[LuanTeles]

btw, I already have once told this on this thread, you can also brick (no hardbrick=no boot) with underclocking. this stupid piece of hardware does not like to be pushing around

I see no point in underclocking the RSX. The default values already run in a pitiful way; imagine at lower frequencies. I don't think anyone would want to do it, except for testing purposes.

 

[haxxxen]

hihi, less is more...I know, this is why only I just want it...well, there is no point in doing so, but I have not noticed any drawback ingame back then, but without any scientological proof. now you @LuanTeles are telling, that it could improve almost about 50% is not right either.

well, I just want to run my consoles flawlessly in 100 years again, lol

ps
don't take too seriously

 

[RIP-Felix]

CELL and RSX clock settings are always stored in the Syscon EEPROM.
CELL reads it's own clock settings before the config ring is send.
Lv1 uses the EEPROM values to initialize RSX, but versions >0.85 removed that and instead use a fallback value. I can patch that but it's easier to patch the hardcoded value.

If you patch the CFW to remove the fallback value then we can use UART to set and revert changes without a hardware flasher. Which is much easier. I believe this would make overclocking much easier and safer.

Also related, if you can please provide insight into the Cell OC that would be helpful. @zecoxao posted a video showing how to OC CELL to 4.8GHz, which you also posted a teaser of on twitter a few years ago. Amazing BTW! We have easy access to SYSCON now, but OFC do not understand how to set clock speeds arbitrarily. Just the default 400 x8 (3.2GHz) and the unstable 600 x 8 (4.8GHz). I tried ChatGPT to see if it could find any patters, but came up empty. If we knew how the bytes worked, we could try more reasonable increases. 450 x 8 (3.6GHz) would be massive and interesting to see.

 

[Nascar1243]

If you patch the CFW to remove the fallback value then we can use UART to set and revert changes without a hardware flasher. Which is much easier. I believe this would make overclocking much easier and safer.

Also related, if you can please provide insight into the Cell OC that would be helpful. @zecoxao posted a video showing how to OC CELL to 4.8GHz, which you also posted a teaser of on twitter a few years ago. Amazing BTW! We have easy access to SYSCON now, but OFC do not understand how to set clock speeds arbitrarily. Just the default 400 x8 (3.2GHz) and the unstable 600 x 8 (4.8GHz). I tried ChatGPT to see if it could find any patters, but came up empty. If we knew how the bytes worked, we could try more reasonable increases. 450 x 8 (3.6GHz) would be massive and interesting to see.

Also noticed when looking at the testbench logs on an A01 that the Cell had a CLK multiplier of 40 and the CLK was at 79.8mhz instead of the 8x multiplier and the 400mhz CLK.

tb_ppu_dpfe_test: tbclkr=40, beclk=3192000000, tbclk=79800000

Not sure if this is significant but I was just looking around at random places and saw this unless I am reading this wrong.

 

[PoppaJerry]

I've been testing my CECH2502B with 1A date code after work today, currently at 850/850 and everything is tracking smoothly so far. I'll keep pushing it further tomorrow, hopefully I'll be able to get to 900/1000!

My CECH2502B appears to have reached its limits at 850/950. I encountered no artifacts but had frequent crashing in games on 850/1000, whereas 900/750 had both artifacts and crashes in games. 850/950 seems to be holding solid and performs pretty nicely, would love to see what a CELL overclock could do in tandem with these speeds on the RSX.

If you don't mind, then yes, I'm still interested !

The RSX model for my CECH2002A is CXD2991GGB!

Just so you guys know I did lots of testing last night on my ps3 slim 2500b date code 0c and was doing really well I got to 850/900 and tested lots of games but one game gta 4 had artifacting and I had to reduce the core speed down to 650mhz to get it to go away and even then I started to notice slight artifacting onces the rsx had warmed up. This is the only game to do it so I would appreciate if anyone else could test there insane overclocks on gta 4 I know its a bit weird as the game is heavily cpu bond on ps3 but I thought I would share it with you guys.

I know its not the memory speed because I reduced it to stock speeds with the core overclock and it remained there.

I played GTA IV (BLUS30127) for an hour on my CECH2502B at 850/950 and didn't encounter any artifacts.

 

[Mitsu™]

Hey !

I saw this comment on YouTube and it makes me wonder...
With this firmware (a REBUG one), this user needed to reactivate OC each time by poking LVL1, so in case of brick would base clocks be restored on shutdown ? If so, testing clocks would be safer (a bit less convenient, but safer.).
Just asking ! Maybe it's totally stupid.

The RSX model for my CECH2002A is CXD2991GGB!

Thank you ! So this is a different RSX model from my two CECH-2004B. They have the "CXD2991EGB" model.
We can also note that even if your 2002A has a more recent date code (9D) than my second 2004B (9C), it still has a """less powerful""" RSX. (No offense)
Maybe date codes/manufacture dates aren't that reliable to determine OC potential.

My CECH2002A with a 9D date code can only do 600/750. 650/800 causes artifacts and crashes within 30 seconds on the XMB, had to re-install a firmware with base clocks through safe mode which also failed multiple times before being successful.

And this is why your 2002A slim can't do 650/800.
RSX models vary in the same "series" and they have their own limits.

For now, we can say that :
- CXD2991EGB stable limits : 700MHz core / 900MHz VRAM
- CXD2991GGB stable limits : 600MHz core / 750MHz VRAM

Did you run any more tests with it ? Like 600/800 or 650/750 for example ?

 

[cha0shacker]

you have a link to the downloads? I have a 2501a that I'll start testing with. I already have the base OC on it.

@DelixDex already posted the link to my OneDrive by now. Sorry that im replying late, i was banned for not combining posts into one.

 

[Thimblewad]

I have the opposite experience. In Black Flag, unstable core results in glitchy geometry but unstable memory results in a hard crash and likewise unstable memory results in the PS3 freezing when opening a game.

1000 memory is not stable on all systems. so far 650_900 has been the min stable for all systems.

Same was here. When trying 750 core there was really bad artifacting in Crysis and AC4:Black Flag. Very minor artifacts were present in almost all games tested. After reverting back to 700 core, artifacts were gone.

Artifacts were present even when I tried 750/750, so I know for sure the problem was the core clock. Maybe the core was eating too much power and VRAM got unstable? Because it sure as hell looked like memory artifacting to me. Nonetheless, memory is stable at 1000 and I'd even try more

 

[Mitsu™]

Same was here. When trying 750 core there was really bad artifacting in Crysis and AC4:Black Flag. Very minor artifacts were present in almost all games tested. After reverting back to 700 core, artifacts were gone.

Artifacts were present even when I tried 750/750, so I know for sure the problem was the core clock. Maybe the core was eating too much power and VRAM got unstable? Because it sure as hell looked like memory artifacting to me. Nonetheless, memory is stable at 1000 and I'd even try more

So, your PS3 is still running perfectly at 1000MHz VRAM (since you installed it) ? Nice

On my side, I tried several games at 700/900 on my 2004B and everything is fine so far.

Games I tested :

- Crysis, aka THE test. No issues. Nice framerate boost.
- Sonic Unleashed. Same. No artifacts or anything.
- Drakengard 3. No issues either.
- Asura's Wrath. No issues.
- Dark Souls, in Blighttown. Everything is fine.
- Fallout New Vegas. Nothing abnormal for a Bethesda game.
- MGS HD Collection. No issues.
- MGS4. No issues. Better framerate.
- Metal Gear Rising Revengeance. Better framerate. Crash after ~30/45 minutes of gameplay. 3 beeps, then shutdown. Probably not OC related since players reported the same issue long ago.

PS1, PS2 and PSP games works well too. No instability.

Temperatures are almost unchanged.

No instabilities on XMB to report too.

 

[Thimblewad]

So, your PS3 is still running perfectly at 1000MHz VRAM (since you installed it) ? Nice

Yes! 700/1000 is stable in all the games I tested. I even left Crysis on overnight one time, it was on for like 10+ hours and it was working perfectly in the morning
So I'm really hoping someone makes a few .PUPs like 1025, 1050 etc. and I hope I don't brick the bad boy

 

[Cratus666]

Hey guys an update from me 2504B(early 2011).I am running 750/975 and everything is still stable. I am playing yakuza zero now and works flawless same for other games no artifacts same with 700/1000 so if anyone can make 25mhz on core I am willing to test 725/1000 which I believe would be stable, on 750/1000 I had a few artifacts. Tested nearly 10 games now including the major ones(THE TEST) Crysis lol no issues also generally speaking what's more powerful between 700/1000 and 750/975? Since they both stable for me

 

[Thimblewad]

Hey guys an update from me 2504B(early 2011).I am running 750/975 and everything is still stable. I am playing yakuza zero now and works flawless same for other games no artifacts same with 700/1000 so if anyone can make 25mhz on core I am willing to test 725/1000 which I believe would be stable, on 750/1000 I had a few artifacts. Tested nearly 10 games now including the major ones(THE TEST) Crysis lol no issues also generally speaking what's more powerful between 700/1000 and 750/975? Since they both stable for me

Unfortunately, the core can only be changed by 50 MHz increments.
Memory can be changed by 25 MHz increments.

You're stuck on 700/1000 like me

 

[ryeno]

Yes! 700/1000 is stable in all the games I tested. I even left Crysis on overnight one time, it was on for like 10+ hours and it was working perfectly in the morning
So I'm really hoping someone makes a few .PUPs like 1025, 1050 etc. and I hope I don't brick the bad boy

My PS3 was game stable at 950 but it would freeze sometimes when opening games.

 

[Thimblewad]

My PS3 was game stable at 950 but it would freeze sometimes when opening games.

Please, do not attempt 1000. If you're already having issues on 950, don't go further, we don't need another expensive door-stopper <3

 

[Cratus666]

Unfortunately, the core can only be changed by 50 MHz increments.
Memory can be changed by 25 MHz increments.

You're stuck on 700/1000 like me

Still good enough for me!!! will settle with what I have, though I see 750/975 was removed on onedrive, did anyone have issues with it?

 

[bizarrewitness]

@DelixDex already posted the link to my OneDrive by now. Sorry that im replying late, i was banned for not combining posts into one.

no worries! I appreciate the reply anyhow. Played some Bioshock infinite on 650/800 yesterday. Still need to buy some of the more GPU-intensive games to test out thoroughly and I'll share my results in the future

 

[PoppaJerry]

Thank you ! So this is a different RSX model from my two CECH-2004B. They have the "CXD2991EGB" model.
We can also note that even if your 2002A has a more recent date code (9D) than my second 2004B (9C), it still has a """less powerful""" RSX. (No offense)

None taken, with the exception of my CECH2502B I've had pretty mediocre results across my consoles lol

Maybe date codes/manufacture dates aren't that reliable to determine OC potential.

I agree, I think it just comes down to the silicon lottery. My CECH2512B can only do 700/900, so it performs on par with your CECH2004B despite having a 40nm RSX and being manufactured more than a year later.

And this is why your 2002A slim can't do 650/800.
RSX models vary in the same "series" and they have their own limits.

For now, we can say that :
- CXD2991EGB stable limits : 700MHz core / 900MHz VRAM
- CXD2991GGB stable limits : 600MHz core / 750MHz VRAM

Did you run any more tests with it ? Like 600/800 or 650/750 for example ?

At the time I only had 650/800 as the 600/XXX and 650/XXX firmware variants were taken down when I was testing. I've run some tests this morning and 600/800 has artifacts in the XMB, 650/750 didn't have artifacts in the XMB but did cause crashes without even launching games. I was able to change firmware in the XMB on 600/800 but had to go through safe mode to do it on 650/750 as it would otherwise crash, thankfully it worked first try in safe mode.

 

[Dark_Ungel]

I have a fat cecha01 Frankenstein with 65nm rsx. Half NEC/TOKINS have been replaced with tamantlum caps.
What safe OC can this bad boy run and anyone got a file?

If I had a system like this I wouldn't push it hard it's just enough it's working

 

[cha0shacker]

Still good enough for me!!! will settle with what I have, though I see 750/975 was removed on onedrive, did anyone have issues with it?

I removed it because the firmware base was kinda old. All the others are based on Beta 5 of 4.91.2 but the 925 and 975 were still based on 4.90 PEX. If anyone needs it i can add it again.