It will require soldering to gain access to the SYSCON UART and fixing the checksum at address 7fe on sherwood. This has the added benefit of giving you access to the syscon fantables to force the fan to100% in case of a brick. That can help you unbrick. So in preparation for the VID Hack, I'd suggest opening and installing the necessary wires for SYSCON UART. Here's a tutorial for that part.
As you can see the max stable overclock I could achieve without voltage increases was 700/900. And with that OC I would get voltage ripple of nearly 80mVpp. This is not enough to trigger voltage protection, but is definitely well above what a well attenuated filter should strive for. The general rule is 1-5% of the output voltage. So at 1v the ripple should be 10-50mVpp or less. As you can see the CPU Vcore (yellow line) is flat. 10mVpp, which is well controlled by it's filter at stock frequencies. But the RSX at 80mV is now outside the well controlled range. That's without increasing voltage!
850/900 required 1.141v to stabilize. Acceptable voltage ripple is 11-57mVpp or less, but I'm getting upto 120mVpp! And yet this is still not enough to trigger voltage protection.
Yesterday I tried 900/900 and while it was stable in XMB, it was unstable in game at 1.141v.
I then boosted the voltage +0.05v to 1.191v. And that made the game more stable, but there were some lingering black speckeld/glitchy artifacts. So I boosted it another +0.05v to 1.215v. Withing a few seconds of gaining control of my character it freeze for about a second, then powered off.
Unlike most unstable Overclocks, this was getting more and more stable with VID hikes. It's not due to the OC being unstable. It was caused by one of these voltage spikes! Worsening, inadequate power delivery/filtering! The SYSCON errorlog contained a new A0801002 error that had never occired on this console before. That's the error code you get when voltage ripple exceeds the Over/under voltage protection. This model (CECH-25xx) will simply shut off. There was no tripple-beep or flashing LEDs. This is a general HW fault, no different than a YLOD, but SONY changed the way it's indicated on this and later models. @Nascar1243 and I have coined the term "No Light of Death" (NLOD or noLOD) to describe this situation.
Anyway, you can see that increasing the voltage increases the voltage ripple and eventually overwhelms the filtering. I can tell that my RSX filter was not perfectly healthy to begin with. This console has 1124 days of use on it already. So the Aluminum Polymer capacitors that SONY has on here are probably wearing down to some extent and overclocking to this extreme has exposed their wear. It would be fine at stock clocks. In fact I want to get a scope image at stock also. But I suspect thes caps need replaced and perhaps I can add a few extra to help even more. That might allow me to push even further.
This is unfortunate. As I don't want to or have experience in soldering at all. And I fear taking appart these PS3s again ever since my December died just by taking them appart. Surely it can't be utterly impossible to modify voltages through just software. Is that really the only way to modify voltages?
Man I hope you or someone can find an easy way to. Then it would easily make it accessible to people. But since you said it will take a long while, guess I'm cooked.
It will require soldering to gain access to the SYSCON UART and fixing the checksum at address 7fe on sherwood. This has the added benefit of giving you access to the syscon fantables to force the fan to100% in case of a brick. That can help you unbrick. So in preparation for the VID Hack, I'd suggest opening and installing the necessary wires for SYSCON UART. Here's a tutorial for that part.
As you can see the max stable overclock I could achieve without voltage increases was 700/900. And with that OC I would get voltage ripple of nearly 80mVpp. This is not enough to trigger voltage protection, but is definitely well above what a well attenuated filter should strive for. The general rule is 1-5% of the output voltage. So at 1v the ripple should be 10-50mVpp or less. As you can see the CPU Vcore (yellow line) is flat. 10mVpp, which is well controlled by it's filter at stock frequencies. But the RSX at 80mV is now outside the well controlled range. That's without increasing voltage!
850/900 required 1.141v to stabilize. Acceptable voltage ripple is 11-57mVpp or less, but I'm getting upto 120mVpp! And yet this is still not enough to trigger voltage protection.
Yesterday I tried 900/900 and while it was stable in XMB, it was unstable in game at 1.141v.
I then boosted the voltage +0.05v to 1.191v. And that made the game more stable, but there were some lingering black speckeld/glitchy artifacts. So I boosted it another +0.05v to 1.215v. Withing a few seconds of gaining control of my character it freeze for about a second, then powered off.
Unlike most unstable Overclocks, this was getting more and more stable with VID hikes. It's not due to the OC being unstable. It was caused by one of these voltage spikes! Worsening, inadequate power delivery/filtering! The SYSCON errorlog contained a new A0801002 error that had never occired on this console before. That's the error code you get when voltage ripple exceeds the Over/under voltage protection. This model (CECH-25xx) will simply shut off. There was no tripple-beep or flashing LEDs. This is a general HW fault, no different than a YLOD, but SONY changed the way it's indicated on this and later models. @Nascar1243 and I have coined the term "No Light of Death" (NLOD or noLOD) to describe this situation.
Anyway, you can see that increasing the voltage increases the voltage ripple and eventually overwhelms the filtering. I can tell that my RSX filter was not perfectly healthy to begin with. This console has 1124 days of use on it already. So the Aluminum Polymer capacitors that SONY has on here are probably wearing down to some extent and overclocking to this extreme has exposed their wear. It would be fine at stock clocks. In fact I want to get a scope image at stock also. But I suspect thes caps need replaced and perhaps I can add a few extra to help even more. That might allow me to push even further.
It's not a bad idea simply because the OC increases the power the console consumes. I'm using an APS226 (most efficient PSU) as a bench PSU and made adapters for all the PS3's 5v connectors. It can deliver more than enough power to any console.
PSUs work most efficiently between 50-60% of their power rating. This 2501a model draws about 75W and this PSU is rated at about 400W. So it's not operating at it's most efficient range, but I do not see this being an issue.
The 12v rail is inherantly noisy. Before every DCDC converter there is a filter (caps) to clean up the noise entering the converter. For Vcore it's a buck converter. That's just so the buck converter can operate efficiently. Bucks them selves switch on and off very fast, generating noise (Swictch mode noise). That's what the capacitors next to the processor are filtering. So it's not the PSU that's generating the ripple/noise, or at least it's contributing very little in comparison to the buck converter itself. Upping the PSU may help keep the PSU alive longer since we're increasing the demand on it, but the ripple/noise I'm talking about is coming from the RSX Voltage Regulation Module itself.
Not may, is. Skewing the VRM signal in software will do it the only hurdle is finding where the VRM signals are in software.
Shunt mods is a very untidy brute force way of things, a cleaner method might be to identify where the voltage is being fed from then adding an adjustable potentiometer. Might need to replace a few caps, possibly upgrade the VRM so it doesn't burn out, but thats about it. Someone more familiar with the various PS3 PCB revisions would be able to say better if the PCB design is up to snuff for handling a mod like that.
When I'm not feeling lazy I'll pull one of my PS3s apart and sort out a refined method for taking a 20xx heatsink and putting it on a 21\25xx model. Should be straightforward and only take a few minutes to work out, just the effort involved.
So it's not the PSU that's generating the ripple/noise, or at least it's contributing very little in comparison to the buck converter itself. Upping the PSU may help keep the PSU alive longer since we're increasing the demand on it, but the ripple/noise I'm talking about is coming from the RSX Voltage Regulation Module itself.
When I'm not feeling lazy I'll pull one of my PS3s apart and sort out a refined method for taking a 20xx heatsink and putting it on a 21\25xx model. Should be straightforward and only take a few minutes to work out, just the effort involved.
It will require soldering to gain access to the SYSCON UART and fixing the checksum at address 7fe on sherwood. This has the added benefit of giving you access to the syscon fantables to force the fan to100% in case of a brick. That can help you unbrick. So in preparation for the VID Hack, I'd suggest opening and installing the necessary wires for SYSCON UART. Here's a tutorial for that part.
As you can see the max stable overclock I could achieve without voltage increases was 700/900. And with that OC I would get voltage ripple of nearly 80mVpp. This is not enough to trigger voltage protection, but is definitely well above what a well attenuated filter should strive for. The general rule is 1-5% of the output voltage. So at 1v the ripple should be 10-50mVpp or less. As you can see the CPU Vcore (yellow line) is flat. 10mVpp, which is well controlled by it's filter at stock frequencies. But the RSX at 80mV is now outside the well controlled range. That's without increasing voltage!
850/900 required 1.141v to stabilize. Acceptable voltage ripple is 11-57mVpp or less, but I'm getting upto 120mVpp! And yet this is still not enough to trigger voltage protection.
Yesterday I tried 900/900 and while it was stable in XMB, it was unstable in game at 1.141v.
I then boosted the voltage +0.05v to 1.191v. And that made the game more stable, but there were some lingering black speckeld/glitchy artifacts. So I boosted it another +0.05v to 1.215v. Withing a few seconds of gaining control of my character it freeze for about a second, then powered off.
Unlike most unstable Overclocks, this was getting more and more stable with VID hikes. It's not due to the OC being unstable. It was caused by one of these voltage spikes! Worsening, inadequate power delivery/filtering! The SYSCON errorlog contained a new A0801002 error that had never occired on this console before. That's the error code you get when voltage ripple exceeds the Over/under voltage protection. This model (CECH-25xx) will simply shut off. There was no tripple-beep or flashing LEDs. This is a general HW fault, no different than a YLOD, but SONY changed the way it's indicated on this and later models. @Nascar1243 and I have coined the term "No Light of Death" (NLOD or noLOD) to describe this situation.
Anyway, you can see that increasing the voltage increases the voltage ripple and eventually overwhelms the filtering. I can tell that my RSX filter was not perfectly healthy to begin with. This console has 1124 days of use on it already. So the Aluminum Polymer capacitors that SONY has on here are probably wearing down to some extent and overclocking to this extreme has exposed their wear. It would be fine at stock clocks. In fact I want to get a scope image at stock also. But I suspect thes caps need replaced and perhaps I can add a few extra to help even more. That might allow me to push even further.
Experiment 1 (yesterday): Overclocking a Slim 2501a using voltage increases to find it's limit.
Result 1: RSX Vcore (VDDC) voltage ripple exceeding 120mVpp at 1.215v was the limit, causing an A0801002 within a few seconds of starting the game.
Experiment 2 (today): I added in parallel 3x 470uF AlPol low ESR (3 or 6 mOhms, IDR which) processor decoupling caps.
Result 2: It prevented the 1002 error at 1.141v and allowed me to get FPS and temps data for 900/900, but there was still slight artifacting. Voltage ripple was reduced to 30mV or less. This is within the 1-5% general rule for a well attenuated filter. It wasn't enought to make 900 core stable. I boosted voltage to 1.240v which caused an A0801002 after about 5 minutes. Voltage ripple was as high as 40mVpp.
Hypothesis 1: I may have reached a current limit with this models VRM. Even though the filter was adequately attenuated, the Buck converters cannot deliver more current than their maximum rating.
Discussion: The console was drawing nearly 100w at the end. I need to look into the datasheets of the ISL6326 and the buck converters used, if I can find it, to see what the maximum current rating is. Then conduct Experiment 3: Using the current sense pads to measure the voltage drop, calculate the current passing into the RSX using ohms law. If that value is close to the Imax of the buck converter phases, then that supports hypothesis 1. If not, then Hypothesis 1 may be refuted and there is another variable not yet understood that's influencing the instability observed.
Not may, is. Skewing the VRM signal in software will do it the only hurdle is finding where the VRM signals are in software.
Shunt mods is a very untidy brute force way of things, a cleaner method might be to identify where the voltage is being fed from then adding an adjustable potentiometer. Might need to replace a few caps, possibly upgrade the VRM so it doesn't burn out, but thats about it. Someone more familiar with the various PS3 PCB revisions would be able to say better if the PCB design is up to snuff for handling a mod like that.
When I'm not feeling lazy I'll pull one of my PS3s apart and sort out a refined method for taking a 20xx heatsink and putting it on a 21\25xx model. Should be straightforward and only take a few minutes to work out, just the effort involved.
Why diminishing returns are inevitable when overclocking a GPU
This phenomenon is due to fundamental principles of physics and the design constraints of semiconductor technology.
Heat Generation :
Increasing the GPU's clock speed = more power consumption, which generates more heat. P ∝ CV²f
More heat makes the GPU less efficient and can also trigger thermal throttling, reducing performance gains.
Note : In the case of a PS3, thermal throttling is unlikely if cooled efficiently.
Note 2 : Claims that overclocking from 500/650 to 900/1000 (for example) doesn't produce more heat are incorrect and contradict basic principles of electronics and thermodynamics.
Power Limits :
GPUs have built-in power limits to prevent damage.
Without increasing voltage, you hit these limits quickly, and further overclocking gives smaller performance boosts.
Efficiency :
At lower overclocks, small increases in clock speed give good performance boosts.
As you push higher, each additional MHz gives less improvement because the GPU becomes less efficient.
Real-World Example
If you look at GPU overclocking benchmarks, you'll see a pattern like this:
Initially, you get a good FPS boost for each MHz increase, but it drops off as you push higher. Meanwhile, power consumption keeps rising significantly.
Conclusion :
All GPUs, regardless of their make and model, follow these fundamental principles of power consumption and efficiency, including the RSX. As you overclock a GPU (especially without increasing the voltage), you encounter diminishing returns due to these unavoidable factors.
Despite arguments to the contrary, the effects of heat generation, power efficiency, and architectural limits ensure that diminishing returns are an inescapable reality in GPU overclocking. Claims that OCing to high frequencies doesn't produce more heat are simply incorrect and overlook fundamental aspects of GPU operation.
• Part 1 : Core OC
I made some experiments with the duck sample test to see at which OC the diminishing returns began to accentuate.
As I said, VRAM is not very exploited by this homebrew, this part only concerns the core.
This clearly shows at which core frequency it becomes less interesting in terms of performance gains.
Up to 700MHz, it's still worth it. After that...
So, if your PS3 can't go over 700MHz core, it's not a big deal. Really.
As @Ketxxx says, if a game has a catastrophic frame rate despite that, it's not only a GPU issue. It's either the CPU that needs an overclock too, or it's just a very poorly optimized game (yes, I'm looking at you Bayonetta).
It is important to keep in mind that the GPU cannot save everything on its own, no matter how high you may OC it.
At the end of the day, it's up to you to decide if stressing your console further is worth the (smaller and decreasing) gains you'll have.
With VRAM at 1000MHz :
• Duck count for a stable 60 FPS :
800/1000 : 288
850/1000 : 294
As you can see, the VRAM doesn't matter in this Duck sample test.
An excellent illustration of diminishing returns :
Note how between 500/650 and 600/750 you win 4.96 FPS, while between 700/900 and 800/1000 you win... 0.11 FPS at the low 0,1%. Full video here
• Part 2 : VRAM OC
Now concerning VRAM, I've tried several games to find a good test and BioShock Infinite with unlocked framerate seems perfect for this.
This exact scene, at this exact place without moving, with water, reflections, etc...
Easily repeatable, since there is a checkpoint here.
While OCing beyond 800MHz does yield some performance improvements, the benefits decrease.
Personally, an OC I'll probably stick to will be 700/800-850 (~ +37% in performance).
A good balance between performance gains and """"safety"""". The minimal gains at higher frequencies aren't worth stressing the hardware further IMO.
If your PS3 can't do 700MHz core, then 650/800-850 will already improve many games.
/!\ 650/800 is safer for a 20XX. Summary in percentage and ending notes :
• VRAM Overclocking :
500/650 to 500/700: +1.69%
500/700 to 500/750: +1.37%
500/750 to 500/800: +0.98%
500/800 to 500/850: +0.78%
500/850 to 500/900: +0.37%
500/900 to 500/950: +0.52%
500/950 to 500/1000: +0.52%
• Core Overclocking :
500/650 to 550/650: +8.37%
550/650 to 600/650: +8.64%
600/650 to 650/650: +6.28%
650/650 to 700/650: +5.51%
700/650 to 750/650: +4.48%
750/650 to 800/650: +2.86%
800/650 to 850/650: +1.74%
Of course, not all games react exactly the same, some don't even benefit from this OC at all because they are more CPU-bottlenecked (ex : Bayonetta). Other games, more GPU-intensives, can get a nice boost, but bad optimization is still a thing to consider and gains might not be as spectacular as expected.
You have to temper your expectations !
• Diminishing returns tendency confirmed with 41 various games. There are small variations, but it's still consistent.
This post is not meant to dictate your actions; rather, it serves as a demonstration to help you better understand the value of GPU overclocking. This way, you can more easily weigh the benefits and risks.
Remember, OCing a GPU always has its downsides, I suggest you to read this post from RIP-Felix : https://www.psx-place.com/threads/p...peeds-ps3-cfw-only.36801/page-126#post-392938
It will require soldering to gain access to the SYSCON UART and fixing the checksum at address 7fe on sherwood. This has the added benefit of giving you access to the syscon fantables to force the fan to100% in case of a brick. That can help you unbrick. So in preparation for the VID Hack, I'd suggest opening and installing the necessary wires for SYSCON UART. Here's a tutorial for that part.
As you can see the max stable overclock I could achieve without voltage increases was 700/900. And with that OC I would get voltage ripple of nearly 80mVpp. This is not enough to trigger voltage protection, but is definitely well above what a well attenuated filter should strive for. The general rule is 1-5% of the output voltage. So at 1v the ripple should be 10-50mVpp or less. As you can see the CPU Vcore (yellow line) is flat. 10mVpp, which is well controlled by it's filter at stock frequencies. But the RSX at 80mV is now outside the well controlled range. That's without increasing voltage!
850/900 required 1.141v to stabilize. Acceptable voltage ripple is 11-57mVpp or less, but I'm getting upto 120mVpp! And yet this is still not enough to trigger voltage protection.
Yesterday I tried 900/900 and while it was stable in XMB, it was unstable in game at 1.141v.
I then boosted the voltage +0.05v to 1.191v. And that made the game more stable, but there were some lingering black speckeld/glitchy artifacts. So I boosted it another +0.05v to 1.215v. Withing a few seconds of gaining control of my character it freeze for about a second, then powered off.
Unlike most unstable Overclocks, this was getting more and more stable with VID hikes. It's not due to the OC being unstable. It was caused by one of these voltage spikes! Worsening, inadequate power delivery/filtering! The SYSCON errorlog contained a new A0801002 error that had never occired on this console before. That's the error code you get when voltage ripple exceeds the Over/under voltage protection. This model (CECH-25xx) will simply shut off. There was no tripple-beep or flashing LEDs. This is a general HW fault, no different than a YLOD, but SONY changed the way it's indicated on this and later models. @Nascar1243 and I have coined the term "No Light of Death" (NLOD or noLOD) to describe this situation.
Anyway, you can see that increasing the voltage increases the voltage ripple and eventually overwhelms the filtering. I can tell that my RSX filter was not perfectly healthy to begin with. This console has 1124 days of use on it already. So the Aluminum Polymer capacitors that SONY has on here are probably wearing down to some extent and overclocking to this extreme has exposed their wear. It would be fine at stock clocks. In fact I want to get a scope image at stock also. But I suspect thes caps need replaced and perhaps I can add a few extra to help even more. That might allow me to push even further.
Are you talking about using a syscon patches in the firmware? Please correct me if I am wrong but I don't think anyone has created custom syscon patches yet. Looking at the ps3 wiki, it seems the issue is that the "Content Data Patch" is encrypted.
Experiment 1 (yesterday): Overclocking a Slim 2501a using voltage increases to find it's limit.
Result 1: RSX Vcore (VDDC) voltage ripple exceeding 120mVpp at 1.215v was the limit, causing an A0801002 within a few seconds of starting the game.
Experiment 2 (today): I added in parallel 3x 470uF AlPol low ESR (3 or 6 mOhms, IDR which) processor decoupling caps.
Result 2: It prevented the 1002 error at 1.141v and allowed me to get FPS and temps data for 900/900, but there was still slight artifacting. Voltage ripple was reduced to 30mV or less. This is within the 1-5% general rule for a well attenuated filter. It wasn't enought to make 900 core stable. I boosted voltage to 1.240v which caused an A0801002 after about 5 minutes. Voltage ripple was as high as 40mVpp.
Hypothesis 1: I may have reached a current limit with this models VRM. Even though the filter was adequately attenuated, the Buck converters cannot deliver more current than their maximum rating.
Discussion: The console was drawing nearly 100w at the end. I need to look into the datasheets of the ISL6326 and the buck converters used, if I can find it, to see what the maximum current rating is. Then conduct Experiment 3: Using the current sense pads to measure the voltage drop, calculate the current passing into the RSX using ohms law. If that value is close to the Imax of the buck converter phases, then that supports hypothesis 1. If not, then Hypothesis 1 may be refuted and there is another variable not yet understood that's influencing the instability observed.
Are you talking about using a syscon patches in the firmware? Please correct me if I am wrong but I don't think anyone has created custom syscon patches yet. Looking at the ps3 wiki, it seems the issue is that the "Content Data Patch" is encrypted.
Based on what @M4j0r has told me the VID is controlled solely by SYSCON. So any SYSCON patch that allow you to write anywhere on the EEPROM would need to be accompanied with a LV1 patch to allow that. But he would be the person to ask about this stuff, since I understand very little about how the programming/software side of PS3 works. He's the wizard.
This was a CECH2501a. It doesn't have tokins. It had a Polymer/MLCC array from the start. But this console has 1125 days of use on it and they have lost their effectiveness and overclocking/overvolting exposed this fact. Adding a few more in parallel restored the filter to operational status and removed failing Aluminum Polymers Caps from the equation, but it wasn't the issue when I hit the limit. I think it's a limit in how much the buck converters can output. Which we can test once I release a tutorial. People here can see if their consoles can output that much voltage and still work. OFC they would be doing so at their own risk. None of this is advisable.
Based on what @M4j0r has told me the VID is controlled solely by SYSCON. So any SYSCON patch that allow you to write anywhere on the EEPROM would need to be accompanied with a LV1 patch to allow that. But he would be the person to ask about this stuff, since I understand very little about how the programming/software side of PS3 works. He's the wizard.
This was a CECH2501a. It doesn't have tokins. It had a Polymer/MLCC array from the start. But this console has 1125 days of use on it and they have lost their effectiveness and overclocking/overvolting exposed this fact. Adding a few more in parallel restored the filter to operational status and removed failing Aluminum Polymers Caps from the equation, but it wasn't the issue when I hit the limit. I think it's a limit in how much the buck converters can output. Which we can test once I release a tutorial. People here can see if their consoles can output that much voltage and still work. OFC they would be doing so at their own risk. None of this is advisable.
Enthusiast level technomancy up in here!
We tried to find a likely candidate for such an thing, but couldn't. The VRAM works that way tho, not that it was helpful at all!
Experiment 1 (yesterday): Overclocking a Slim 2501a using voltage increases to find it's limit.
Result 1: RSX Vcore (VDDC) voltage ripple exceeding 120mVpp at 1.215v was the limit, causing an A0801002 within a few seconds of starting the game.
Experiment 2 (today): I added in parallel 3x 470uF AlPol low ESR (3 or 6 mOhms, IDR which) processor decoupling caps.
Result 2: It prevented the 1002 error at 1.141v and allowed me to get FPS and temps data for 900/900, but there was still slight artifacting. Voltage ripple was reduced to 30mV or less. This is within the 1-5% general rule for a well attenuated filter. It wasn't enought to make 900 core stable. I boosted voltage to 1.240v which caused an A0801002 after about 5 minutes. Voltage ripple was as high as 40mVpp.
Hypothesis 1: I may have reached a current limit with this models VRM. Even though the filter was adequately attenuated, the Buck converters cannot deliver more current than their maximum rating.
Discussion: The console was drawing nearly 100w at the end. I need to look into the datasheets of the ISL6326 and the buck converters used, if I can find it, to see what the maximum current rating is. Then conduct Experiment 3: Using the current sense pads to measure the voltage drop, calculate the current passing into the RSX using ohms law. If that value is close to the Imax of the buck converter phases, then that supports hypothesis 1. If not, then Hypothesis 1 may be refuted and there is another variable not yet understood that's influencing the instability observed.
Hypothetically, the GPU itself should have no issues at 1.240v assuming it's being powered with appropriate VRMs and filtering, right? Another forum member noted that the Curie architecture had plenty of cards run at 1.4V, so I'm assuming the Core is designed with that kind of voltage in mind?
