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

[Razmann4k]

4.8ghz? Wtf? That's insane. That doesn't sound right. Hell 4ghz sounded high as hell already to me. And 4ghz would do so much for so many games. 3.2ghz is low as hell and that's why games suffer so much. The fact we can't keep a measly already shit 30 stable is a crime. The cpu should have been the one thing Sony overclocked on the slims. That could have been a massive selling point, and it would have cost them nothing.

Zecoxao made a video on it:

And I see you commented on it a month ago Tanzu
In the video he overclocks the Cell in a DECR PS3 devkit to 4.8GHz, but says it's possible on retail PS3 (with 90nm) but must apparently be done by writing to the Syscon via UART (unless I misheard him)

 

[Lazard]

Hello guys can someone please give me the 750/975 file? Thanks

 

[cha0shacker]

4.8ghz? Wtf? That's insane. That doesn't sound right. Hell 4ghz sounded high as hell already to me. And 4ghz would do so much for so many games. 3.2ghz is low as hell and that's why games suffer so much. The fact we can't keep a measly already shit 30 stable is a crime. The cpu should have been the one thing Sony overclocked on the slims. That could have been a massive selling point, and it would have cost them nothing.

3.2GHz was high for the time the 7th generation came out. The problem is that it is basically just a single core that was not designed to be fast. Just fast enough to feed the SPEs. But that's not how it was used.
Increasing the CPU frequency will produce compatibility problems too. Simply due to the fact that game logic is often synchronised with the TSC (Time Stamp Counter) on the console.

 

[LuanTeles]

@zecoxao

Does it work on retail consoles on DEX mode? i mean, can we load lv1, vsh and etc from network?

 

[GuilloteTesla]

@zecoxao

Does it work on retail consoles on DEX mode? i mean, can we load lv1, vsh and etc from network?

Loading LV1 from USB or network would be really good and desirable!. No more bricks from too high overclocks and a very nice debug feature all around.

 

[M4j0r]

If I'm not mistaken, some individuals have gotten the 90nm Cell to run at 4.8GHz in OtherOS? If that's true, surely one day if we find a way to do it in GameOS and for 65/45nm Cells, 4.8GHz should be relatively easy? If so, that's a 50% clockspeed improvement, and then who knows, increasing XDR clocks too? Of course that part is entirely speculative.

Zecoxao made a video on it:

And I see you commented on it a month ago Tanzu
In the video he overclocks the Cell in a DECR PS3 devkit to 4.8GHz, but says it's possible on retail PS3 (with 90nm) but must apparently be done by writing to the Syscon via UART (unless I misheard him)

Currently only 3.2Ghz and 4.8Ghz are the possible speeds available on 90nm. 65nm and 45nm have this function deactived. I've shared the settings on twitter and here: https://pastebin.com/qy3u25Av

Does it work on retail consoles on DEX mode? i mean, can we load lv1, vsh and etc from network?

It works on any console with a Communication Process (which can be added to any console). But the DECR-1000A is the only ("retail") console which comes with it. It has never been tested on a retail platform, the lack of SB RAM might be a problem.

Loading LV1 from USB or network would be really good and desirable!. No more bricks from too high overclocks and a very nice debug feature all around.

Lv0/lv1ldr have no drivers for USB, they only support the logical console which is routed via the SB -> PCI -> TMR -> PCI -> TCP -> Ethernet -> Host PC.

 

[LuanTeles]

It works on any console with a Communication Process (which can be added to any console). But the DECR-1000A is the only ("retail") console which comes with it. It has never been tested on a retail platform, the lack of SB RAM might be a problem.

It's worth a try. I'm interested in it for safer overclocks, but I don't have a LAN port to test it on my current setup.

Can any of you guys @Tanzu15, @Mitsu™, @cha0shacker , @RIP-Felix try it?

 

[M4j0r]

It's worth a try. I'm interested in it for safer overclocks, but I don't have a LAN port to test it on my current setup.

Can any of you guys @Tanzu15, @Mitsu™, @cha0shacker , @RIP-Felix try it?

For this to work, someone would need to fully reverse CN4301 (see https://www.psdevwiki.com/ps3/TMU-520#CN4301_160p_(PCI_+_Control_+_?,_to_TMR-520) , https://www.psdevwiki.com/ps3/PCI#160_pin_pad_layout ) and then design and build a custom adapter PCB to adapt the TMR (https://www.psdevwiki.com/ps3/TMR-520) bottom port to the retail PCI connector (https://www.psdevwiki.com/ps3/PCI#70/80_pin_miniPCI_pad_layout).
You can then use a TMR-520 and TCP-520 harvested from a DECR-1000A.

 

[iTzSweetTooth]

@Mitsu™

Can you please give me 900/975 and 900/1000, 950/975 links please

I'm now running 900/950 and it's fully stable and I'm trying to see if I can go above

thank you bro

 

[Tanzu15]

It's worth a try. I'm interested in it for safer overclocks, but I don't have a LAN port to test it on my current setup.

Can any of you guys @Tanzu15, @Mitsu™, @cha0shacker , @RIP-Felix try it?

I would try but his response makes it seem impossible unless someone converts all the correct things. Hopefully someone does.

For this to work, someone would need to fully reverse CN4301 (see https://www.psdevwiki.com/ps3/TMU-520#CN4301_160p_(PCI_+_Control_+_?,_to_TMR-520) , https://www.psdevwiki.com/ps3/PCI#160_pin_pad_layout ) and then design and build a custom adapter PCB to adapt the TMR (https://www.psdevwiki.com/ps3/TMR-520) bottom port to the retail PCI connector (https://www.psdevwiki.com/ps3/PCI#70/80_pin_miniPCI_pad_layout).
You can then use a TMR-520 and TCP-520 harvested from a DECR-1000A.

so seems like mission impossible.

 

[Ketxxx]

Duck test is great, my only problem about it is that the framerate is not "free". It will fall back to 30, then 20, 15, etc...

@Mitsu™ that could be a result of vsync. Its possible to get into situations where frames are literally dropped but you either have the choice of vsync and a locked 30/60FPS but you can potentially drop frames so the frame rate will be slower than it actually is, or an unlocked frame rate where performance is all over the shop with screen tearing. Probably a good balance here would have been to set an FPS cap of 45.

Without even talking about the CPU, looking at the GPU, they improved it so much that they could have indeed release a PS3 pro.
The 90nm variant was supposed to be clocked at 550/700. But heat issue, etc... Okay.
The 65nm RSX could have been set to 600/750 easily, without issues.
Starting from the 40nm series, 650/800 (or /850).

The 50MHz change for the 90nm RSX really wouldn't have made any change in terms of heat, maybe 2-3c, the decision was probably more to do with the RSX being manufactured with crappy materials so they were trying to stave off the inevitable death of those chips. The 65nm RSX I'm also dubious about, not something I looked into in as much detail as I should have and I have since binned those boards I had with suspect GPUs but they were better for sure than the 90nm parts just not as reliable as people think they were\are.

To be honest (and really, don't take it personally, I respect your analysis and your work), when we talk about a 25XX, I have a hard time considering 650/800 as an overclock. Officially yes, it is an overclock, but taking into account the capabilities of these chips ? Not so much.

Don't worry I know where you are coming from on this, when we start talking about the 40nm RSX because anything over the specified frequency puts you in "wild territory" there is no guarantees that every chip would meet specs of 650\800 (although on the GPU frequency a 40\65nm RSX definitely should, I won't go as far to say as "Can", and I'll cover that further on) they haven't been binned or validated for those speeds meaning there likely is GPUs out there that can't hit 650MHz. The memory used for the RSX\GPU (whatever you prefer) varies quite wildly I know what memory Sony generally used the majority of the time and those modules aren't rated for more than 700MHz so while the RSX tends to take kindly to OCing the realistic performance target should be 30FPS (which is what 6th gen consoles aimed for) in which case 650\800 really is enough. The extra frequency headroom while possible becomes much harder to control thermally while the chip could run at 80c and be fine it definitely wouldn't live as long as it otherwise could the recommendation of 650\800 is for the sake of ensuring all PS3s live a long happy life Aside from these things another factor is that current data on what the GPU critical points are (this is a scaling factor that shows frequency scaling to supplied GPU voltage) is far too limited at this time. I'd only recommend higher than 650MHz on the GPU with a heatsink from a 20xx model frankenmodded to a 21\25xx model PS3. From what I remember such a mod is easy to do, remove the little lip that sits on the VRMs of 21\25xx slims then just attach the 20xx heatsink. I can help you with doing this mod if you want to try it I'm just too lazy to open one of the PS3s I have up to hammer all the details out lol.

I think one of the main reasons why Sony did not release a PS3 pro and kept the clocks at 500/650 is to avoid complicating the development of the games (same for the CPU). It took time for the developers to barely "master" (so to speak) this console, so if in addition they had to develop for consoles with various performances, it would have been hellish.
It was just not the right generation to do it/the right console nor the right time.

Mostly I agree, but the extra performance of a PS3 "Pro" would have made developers lives substantially easier there's definitely no good reason why Sony couldn't have bumped the Cell to 4GHz once the PS3 slim came along at the very least as in almost all cases developers complained about the CPU being a complete dog and with good reason latency from the Cell is absolutely horrible it can be in the region of 300ms, likely more with really badly coded games where the Cell favours bandwidth over low latency, or rather gobs of bandwidth that is near useless because data travels on that bandwidth at a snails pace, aka; poor efficiency.

 

[Tanzu15]

@Mitsu™ that could be a result of vsync. Its possible to get into situations where frames are literally dropped but you either have the choice of vsync and a locked 30/60FPS but you can potentially drop frames so the frame rate will be slower than it actually is, or an unlocked frame rate where performance is all over the shop with screen tearing. Probably a good balance here would have been to set an FPS cap of 45.




The 50MHz change for the 90nm RSX really wouldn't have made any change in terms of heat, maybe 2-3c, the decision was probably more to do with the RSX being manufactured with crappy materials so they were trying to stave off the inevitable death of those chips. The 65nm RSX I'm also dubious about, not something I looked into in as much detail as I should have and I have since binned those boards I had with suspect GPUs but they were better for sure than the 90nm parts just not as reliable as people think they were\are.



Don't worry I know where you are coming from on this, when we start talking about the 40nm RSX because anything over the specified frequency puts you in "wild territory" there is no guarantees that every chip would meet specs of 650\800 (although on the GPU frequency a 40\65nm RSX definitely should, I won't go as far to say as "Can", and I'll cover that further on) they haven't been binned or validated for those speeds meaning there likely is GPUs out there that can't hit 650MHz. The memory used for the RSX\GPU (whatever you prefer) varies quite wildly I know what memory Sony generally used the majority of the time and those modules aren't rated for more than 700MHz so while the RSX tends to take kindly to OCing the realistic performance target should be 30FPS (which is what 6th gen consoles aimed for) in which case 650\800 really is enough. The extra frequency headroom while possible becomes much harder to control thermally while the chip could run at 80c and be fine it definitely wouldn't live as long as it otherwise could the recommendation of 650\800 is for the sake of ensuring all PS3s live a long happy life Aside from these things another factor is that current data on what the GPU critical points are (this is a scaling factor that shows frequency scaling to supplied GPU voltage) is far too limited at this time. I'd only recommend higher than 650MHz on the GPU with a heatsink from a 20xx model frankenmodded to a 21\25xx model PS3. From what I remember such a mod is easy to do, remove the little lip that sits on the VRMs of 21\25xx slims then just attach the 20xx heatsink. I can help you with doing this mod if you want to try it I'm just too lazy to open one of the PS3s I have up to hammer all the details out lol.




Mostly I agree, but the extra performance of a PS3 "Pro" would have made developers lives substantially easier there's definitely no good reason why Sony couldn't have bumped the Cell to 4GHz once the PS3 slim came along at the very least as in almost all cases developers complained about the CPU being a complete dog and with good reason latency from the Cell is absolutely horrible it can be in the region of 300ms, likely more with really badly coded games where the Cell favours bandwidth over low latency, or rather gobs of bandwidth that is near useless because data travels on that bandwidth at a snails pace, aka; poor efficiency.

Amen bro. Sony not doing shit for performance on the 25xx slims is a crime. By that time, in 2010-2013, these slims needed that massive boost to clocks on cpu and gpu to help developers and games out. This would have been free performance! And would have made games night and day better. Now we are stuck hoping someone can crack the cpu overclock. That's the only thing holding games back. It's the one thing that would help massively. We know how the RSX behaves now. We know which models and which months can OC to. Now we just need the higher voltage mod and or at least a safe fully stable OC on the cpu. I'll take a good cpu OC over gpu voltage mod.

 

[Razmann4k]

3.2GHz was high for the time the 7th generation came out. The problem is that it is basically just a single core that was not designed to be fast. Just fast enough to feed the SPEs. But that's not how it was used.
Increasing the CPU frequency will produce compatibility problems too. Simply due to the fact that game logic is often synchronised with the TSC (Time Stamp Counter) on the console.

It's a shame, if 65/45nm were capable of this too (which I now see that they aren't) this could've been a great thing if we could uncouple the game logic from clock speed. I know games reference the .sprx libraries, I just wonder if any games got their time from one of these libraries, and if so, could we possibly recompile these libraries to compensate for a hypothetical higher clock by for example making that clock tick 50% slower in the case of 4.8GHz so the game logic would run at 100% speed.

Currently only 3.2Ghz and 4.8Ghz are the possible speeds available on 90nm. 65nm and 45nm have this function deactived. I've shared the settings on twitter and here: https://pastebin.com/qy3u25Av


It works on any console with a Communication Process (which can be added to any console). But the DECR-1000A is the only ("retail") console which comes with it. It has never been tested on a retail platform, the lack of SB RAM might be a problem.


Lv0/lv1ldr have no drivers for USB, they only support the logical console which is routed via the SB -> PCI -> TMR -> PCI -> TCP -> Ethernet -> Host PC.

Ah I see, 65/45 are incapable of these higher clocks. For the 90nm, is this possible in GameOS? If so, I'm sure it'd be wildly unstable and dodgy, but hypothetically do any games still run at a proper speed or do they all seem to run at 150% speed (or at least try to?)

Amen bro. Sony not doing shit for performance on the 25xx slims is a crime. By that time, in 2010-2013, these slims needed that massive boost to clocks on cpu and gpu to help developers and games out. This would have been free performance! And would have made games night and day better. Now we are stuck hoping someone can crack the cpu overclock. That's the only thing holding games back. It's the one thing that would help massively. We know how the RSX behaves now. We know which models and which months can OC to. Now we just need the higher voltage mod and or at least a safe fully stable OC on the cpu. I'll take a good cpu OC over gpu voltage mod.

I think it had more to do with not alienating old PS3 owners. Imagine in 2012 a new PS3 Slim comes out with a 4GHz Cell and 1000/1300 (double 500/650) RSX, 512MB XDR and 512MB GDDR3. That would be great for people in the market for a new console and probably wouldn't cost Sony much more to produce, but what about people who already have a PS3? Suddenly a console with the same name would run games way faster, and developers would naturally want to target their games for these new PS3 Pros. Those games would run awfully on existing PS3s and that would obviously cause massive backlash. Sony and MS eseentially solved this with their PS4 Pro and One X by essentially making them 2 different machines from their regular counterparts, but that would scale back to regular PS4/One speeds for games that didn't have Pro support (games before 2016/17).

 

[Tanzu15]

It's a shame, if 65/45nm were capable of this too (which I now see that they aren't) this could've been a great thing if we could uncouple the game logic from clock speed. I know games reference the .sprx libraries, I just wonder if any games got their time from one of these libraries, and if so, could we possibly recompile these libraries to compensate for a hypothetical higher clock by for example making that clock tick 50% slower in the case of 4.8GHz so the game logic would run at 100% speed.


Ah I see, 65/45 are incapable of these higher clocks. For the 90nm, is this possible in GameOS? If so, I'm sure it'd be wildly unstable and dodgy, but hypothetically do any games still run at a proper speed or do they all seem to run at 150% speed (or at least try to?)


I think it had more to do with not alienating old PS3 owners. Imagine in 2012 a new PS3 Slim comes out with a 4GHz Cell and 1000/1300 (double 500/650) RSX, 512MB XDR and 512MB GDDR3. That would be great for people in the market for a new console and probably wouldn't cost Sony much more to produce, but what about people who already have a PS3? Suddenly a console with the same name would run games way faster, and developers would naturally want to target their games for these new PS3 Pros. Those games would run awfully on existing PS3s and that would obviously cause massive backlash. Sony and MS eseentially solved this with their PS4 Pro and One X by essentially making them 2 different machines from their regular counterparts, but that would scale back to regular PS4/One speeds for games that didn't have Pro support (games before 2016/17).

Yes I know that. But it's annoying these manufacturing proccess made these chips so fast and efficient and we had zero access to that power. The PS3s performance and optimization was already ass. These Overclocks have proven to be powerful and game changing. It was free performance. But since the fats couldn't keep up, they didn't. And yeah they wanted to keep all PS3s equal. Which sucks cause all the 25xx and later slims get better and better and faster. The super slim ended up with such a huge leap in node shrink, these PS3s could easily have been Ps3 pros. The other thing, was the dog shit amount of vram. That's what limited us the most. Resolution and textures were worst on PS3 than Xbox in many many 3rd parties. 1st party games looked great texture wise, but imagine how much better they would be if we had a full actual 512mb or 1gb amount of vram. It would have been so awesome.

 

[Razmann4k]

Yes I know that. But it's annoying these manufacturing proccess made these chips so fast and efficient and we had zero access to that power. The PS3s performance and optimization was already ass. These Overclocks have proven to be powerful and game changing. It was free performance. But since the fats couldn't keep up, they didn't. And yeah they wanted to keep all PS3s equal. Which sucks cause all the 25xx and later slims get better and better and faster. The super slim ended up with such a huge leap in node shrink, these PS3s could easily have been Ps3 pros. The other thing, was the dog shit amount of vram. That's what limited us the most. Resolution and textures were worst on PS3 than Xbox in many many 3rd parties. 1st party games looked great texture wise, but imagine how much better they would be if we had a full actual 512mb or 1gb amount of vram. It would have been so awesome.

At least we finally get to see the PS3 reach its full potential! I really do hope we one day do get to see what the 28nm RSX is capable of. With this infomation that only 90nm are able to overclock with the method zecoxao sent, it really does show that a Frankenstein PS3 really is the best possible version, I'd love to have one!

Also yeah, Microsoft made the right call to go with a single pool of 512MB shared between the CPU and GPU, because its way more versatile. If you have a game that requires 256MB of system memory and 256MB of VRAM, then it can work like the PS3 and split it evenly. But more often than not, you didn't need system memory as much as you did VRAM, so the 360 allowed for higher resolution textures without having any more memory in total.

By the way, I was under the impression that the CECHA's never made it to European territories, but here in South Africa in a random console repair shop I saw several (with YLODs). I doubt they were imported as this was a small shop. We usually get 04 region consoles which are apparently Europe/Middle East/Africa. I'm 100% sure these were launch models, they had the flap on the front with the card reader and shiny chrome trim, and they said CECHA on the back.

 

[RIP-Felix]

UPDATE: VID HACKING WORKS
Ok, entering brick territory on my 25xx. I previously was able to do 750MHz core but 800MHz was not stable. I have VID boosted this core. So if it stabilizes 800MHz, that means VID hacking for core voltage mods is a viable method.
Bro! IT WORKS!

Went from glitchy at 0.992v to mostly stable 1.015v. There was some minor glitching, so I just boosted to 1.042. And now it appears to have removed the glitching. So yes, VID hacking does help a lot. BUT! Temps are climbing quickly. Cooling may become a problem. I noticed some minor glitching as the chip warmed up.

Next I installed 850/900. This is now 100MHz beyond stable at stock voltage. I fully expect this to require some serious voltage to make stable or recover from. What I'm doing is pretty dumb.

THIS IS WILD. It was unstable in XMB. So I boosted it another 0.05v to 1.100 volts. Measuring 1.089v. And we are glitching in games but otherwise stable. I mean, it's pretty glitchy. But I think Another 0.05v will clean that up. And yes indeed it did! Up to 1.141v to stabilize 850core. Temps took a hit.

900 core? Why not! Installed and…

Well, it seemed fine in XMB. So let's launch Murder Crysis! As soon as I got in game….
A0801002

IG I found the limits of this power delivery system! Tested again and same thing. As soon as I get in game, pop! No light of death. Just cuts powers off.

This time A0801301/A0801002 (PLL Unlock and RSX Filter caps)

Unintentionally confirmed PLL unlocks can occur from Power deliver issues. I don't normally associate 1301 with filtering. I would normally think GPU Bumps/BGA, but in this extreme scenario it happened. I did get some "System Warning : busy loop detected" in the Southbridge UART logs. I noticed them when opening XMB in game while it's loading too. So a lack of filtering is causing interference with normal signaling. The power delivery in this 2501a was not designed to be pushed as high as 1.2v! I think I have pushed my luck as far as I should with this console.

Let's conclude this test. Went from 700MHz at 0.966v to 850MHz 1.141v. Mostly stable. So each 50MHz required about 60mV more. My VID tables worked well. No surprises.

BTW, since frankies are able to deliver more power for a more powerful RSX, they should easily be able to volt mod higher than I was here with this 2501a. I think this is proof of concept. Something I didn't do, but probably will is measure the voltage ripple at stock, then the max stable OC I achieved today (850/900 @ 1.14v), then add a tantalum or 2, and measure the reduction. See if it stabilizes 900/900 and prevents the 1002. Interesting results.

 

[M4j0r]

Ah I see, 65/45 are incapable of these higher clocks. For the 90nm, is this possible in GameOS? If so, I'm sure it'd be wildly unstable and dodgy, but hypothetically do any games still run at a proper speed or do they all seem to run at 150% speed (or at least try to?)

65nm/45nm don't have the clock setting feature - they'll always run at 3.2GHz. On 90nm you can only set it via the Syscon EEPROM because the value needs to be provided during the CELL initialization - 4.8GHz is very unstable.
I'll continue to try to find other values, but that's very hard because the locked clock of CELL is a security feature, you need to provide a special value to CELL to allow any other clocks. The 4.8GHz value is from prototypes.
On the RSX you don't need to do any of that, it'll just accept the clock you set.

 

[Mitsu™]

It's worth a try. I'm interested in it for safer overclocks, but I don't have a LAN port to test it on my current setup.
Can any of you guys @Tanzu15, @Mitsu™, @cha0shacker , @RIP-Felix try it?

I'm sorry, but as Tanzu said, it seems extremely complex to me :')

@Mitsu™

Can you please give me 900/975 and 900/1000, 950/975 links please

I'm now running 900/950 and it's fully stable and I'm trying to see if I can go above

thank you bro

I can't build new firmwares for the moment, check here :
https://app.mediafire.com/0uxvp5ismnjwm
You will find what you need. Sure, it's beta 8, but that doesn't matter at all.

@Mitsu™ that could be a result of vsync.

Yup, I know that. Unfortunately, it make this tool unusable for me. That, and the fact that the VRAM isn't really used in this test, so it's only useful when benchmarking the core.

The extra frequency headroom while possible becomes much harder to control thermally while the chip could run at 80c and be fine it definitely wouldn't live as long as it otherwise could the recommendation of 650\800 is for the sake of ensuring all PS3s live a long happy life

And you're right to look for a universal safe and stable OC. 650/800 is perfect for that ! Though, regarding heat generation, 650/800 or 850/1000, I can assure you it doesn't make a significant difference. 40-45% fan speed is enough to keep the GPU below 65°C all time. In fact, it barely reach 60°C.

I'd only recommend higher than 650MHz on the GPU with a heatsink from a 20xx model frankenmodded to a 21\25xx model PS3. From what I remember such a mod is easy to do, remove the little lip that sits on the VRMs of 21\25xx slims then just attach the 20xx heatsink. I can help you with doing this mod if you want to try it I'm just too lazy to open one of the PS3s I have up to hammer all the details out lol.

I tried that with the first revision, but it's more difficult than I thought and would require more modifications.

Spoiler

I'd love to succeed, but it just doesn't match. And the 21XX revision doesn't interest me, it's a huge downgrade from the model I'm trying to adapt.

Look at this post :
https://www.psx-place.com/threads/p...peeds-ps3-cfw-only.36801/page-101#post-390716

And check this page for more informations about all the type 5 revisions :
https://www.psdevwiki.com/ps3/Type_5

UPDATE: VID HACKING WORKS
Ok, entering brick territory on my 25xx. I previously was able to do 750MHz core but 800MHz was not stable. I have VID boosted this core. So if it stabilizes 800MHz, that means VID hacking for core voltage mods is a viable method.
Bro! IT WORKS!

Went from glitchy at 0.992v to mostly stable 1.015v. There was some minor glitching, so I just boosted to 1.042. And now it appears to have removed the glitching. So yes, VID hacking does help a lot. BUT! Temps are climbing quickly. Cooling may become a problem. I noticed some minor glitching as the chip warmed up.

Next I installed 850/900. This is now 100MHz beyond stable at stock voltage. I fully expect this to require some serious voltage to make stable or recover from. What I'm doing is pretty dumb.

THIS IS WILD. It was unstable in XMB. So I boosted it another 0.05v to 1.100 volts. Measuring 1.089v. And we are glitching in games but otherwise stable. I mean, it's pretty glitchy. But I think Another 0.05v will clean that up. And yes indeed it did! Up to 1.141v to stabilize 850core. Temps took a hit.

900 core? Why not! Installed and…

Well, it seemed fine in XMB. So let's launch Murder Crysis! As soon as I got in game….
A0801002

IG I found the limits of this power delivery system! Tested again and same thing. As soon as I get in game, pop! No light of death. Just cuts powers off.

This time A0801301/A0801002 (PLL Unlock and RSX Filter caps)

Unintentionally confirmed PLL unlocks can occur from Power deliver issues. I don't normally associate 1301 with filtering. I would normally think GPU Bumps/BGA, but in this extreme scenario it happened. I did get some "System Warning : busy loop detected" in the Southbridge UART logs. I noticed them when opening XMB in game while it's loading too. So a lack of filtering is causing interference with normal signaling. The power delivery in this 2501a was not designed to be pushed as high as 1.2v! I think I have pushed my luck as far as I should with this console.

Let's conclude this test. Went from 700MHz at 0.966v to 850MHz 1.141v. Mostly stable. So each 50MHz required about 60mV more. My VID tables worked well. No surprises.

BTW, since frankies are able to deliver more power for a more powerful RSX, they should easily be able to volt mod higher than I was here with this 2501a. I think this is proof of concept. Something I didn't do, but probably will is measure the voltage ripple at stock, then the max stable OC I achieved today (850/900 @ 1.14v), then add a tantalum or 2, and measure the reduction. See if it stabilizes 900/900 and prevents the 1002. Interesting results.

Awesome ! It's more and more concrete !
When you say that temps are climbing quickly, how much hotter it gets compared to stock ?

 

[Razmann4k]

65nm/45nm don't have the clock setting feature - they'll always run at 3.2GHz. On 90nm you can only set it via the Syscon EEPROM because the value needs to be provided during the CELL initialization - 4.8GHz is very unstable.
I'll continue to try to find other values, but that's very hard because the locked clock of CELL is a security feature, you need to provide a special value to CELL to allow any other clocks. The 4.8GHz value is from prototypes.
On the RSX you don't need to do any of that, it'll just accept the clock you set.

Oh, I assumed it was 4.8GHz because that would be a 1.5x multiplier, not a special value that could possibly be changed to another frequency (if you can find other working values). Now you obviously just said 65/45 is incapable of varying from 3.2GHz, my question is: do you think the 4.8GHz clock is unstable because the 90nm struggles at that speed or do you think it's because of the software itself not coping with the speed difference? Hypothetically if those smaller nodes could be clocked higher for instance (which it can't be), would it be more stable or is it simply a limitation of GameOS?

 

[Tanzu15]

UPDATE: VID HACKING WORKS
Ok, entering brick territory on my 25xx. I previously was able to do 750MHz core but 800MHz was not stable. I have VID boosted this core. So if it stabilizes 800MHz, that means VID hacking for core voltage mods is a viable method.
Bro! IT WORKS!

Went from glitchy at 0.992v to mostly stable 1.015v. There was some minor glitching, so I just boosted to 1.042. And now it appears to have removed the glitching. So yes, VID hacking does help a lot. BUT! Temps are climbing quickly. Cooling may become a problem. I noticed some minor glitching as the chip warmed up.

Next I installed 850/900. This is now 100MHz beyond stable at stock voltage. I fully expect this to require some serious voltage to make stable or recover from. What I'm doing is pretty dumb.

THIS IS WILD. It was unstable in XMB. So I boosted it another 0.05v to 1.100 volts. Measuring 1.089v. And we are glitching in games but otherwise stable. I mean, it's pretty glitchy. But I think Another 0.05v will clean that up. And yes indeed it did! Up to 1.141v to stabilize 850core. Temps took a hit.

900 core? Why not! Installed and…

Well, it seemed fine in XMB. So let's launch Murder Crysis! As soon as I got in game….
A0801002

IG I found the limits of this power delivery system! Tested again and same thing. As soon as I get in game, pop! No light of death. Just cuts powers off.

This time A0801301/A0801002 (PLL Unlock and RSX Filter caps)

Unintentionally confirmed PLL unlocks can occur from Power deliver issues. I don't normally associate 1301 with filtering. I would normally think GPU Bumps/BGA, but in this extreme scenario it happened. I did get some "System Warning : busy loop detected" in the Southbridge UART logs. I noticed them when opening XMB in game while it's loading too. So a lack of filtering is causing interference with normal signaling. The power delivery in this 2501a was not designed to be pushed as high as 1.2v! I think I have pushed my luck as far as I should with this console.

Let's conclude this test. Went from 700MHz at 0.966v to 850MHz 1.141v. Mostly stable. So each 50MHz required about 60mV more. My VID tables worked well. No surprises.

BTW, since frankies are able to deliver more power for a more powerful RSX, they should easily be able to volt mod higher than I was here with this 2501a. I think this is proof of concept. Something I didn't do, but probably will is measure the voltage ripple at stock, then the max stable OC I achieved today (850/900 @ 1.14v), then add a tantalum or 2, and measure the reduction. See if it stabilizes 900/900 and prevents the 1002. Interesting results.

This is fantastic news. This means my 850 fully stable PS3s can easily do 900 now since all that happens is it crashes no artifacting due to lack of voltage. And my 950 can easily hit 1000 on core. So it would be 1000/1000. Hopefully you release an easy dummy proof tutorial soon. I've been bored out of my mind.