Throttlestop overclocking Desktop PCs

[Retrorockit]

That's the only way progress is made. Try something that doesn't work , and keep trying until it does.
It only takes 2 things to be a genius. Disagree with everyone else in the world. And make it stick.

 

[frankr2994]

That's the only way progress is made. Try something that doesn't work , and keep trying until it does.
It only takes 2 things to be a genius. Disagree with everyone else in the world. And make it stick.

I sent this guy so many bios files that bricked the board lol. he's been a champ reviving it each time.

 

[unclewebb]

What does ThrottleStop report for CPU speed when running Cinebench or any similar test? 32 multi for all or only for some of the cores?

Edit - You can try pushing the Mod heading on the monitoring table to report each CPU ID number.

 

[Retrorockit]

On my Dell Dimension I kept trying all kinds of OC stuff everybogy said I needed to do. Didn't have a clue what i was doing. Fianlly realized it crahsed every time the fan would speed up. Hot wired the fan and went straight to 2nd place at CPUZ, Then bumped that up another whole multiplier. All the stuff that did nothing for me before, paid off all of a sudden.

 

[frankr2994]

What does ThrottleStop report for CPU speed when running Cinebench or any similar test? 32 multi for all or only for some of the cores?

Edit - You can try pushing the Mod heading on the monitoring table to report each CPU ID number.

oh we haven't gotten that far yet lol. Light loads right now we can get the speed up there. I'm going through some OC settings with him . voltage and power limits.

 

[frankr2994]

What does ThrottleStop report for CPU speed when running Cinebench or any similar test? 32 multi for all or only for some of the cores?

Edit - You can try pushing the Mod heading on the monitoring table to report each CPU ID number.

Ok ended up applying the micro code hack. Which I found out is just changing some items in the bios anyways still working on going further but as of right now all core turbo 3.2ghz under load. And would you mind expanding on the pushing the mod heading comment? I also attached an image of what said hack changes. I applied them manually to my bios last night but the settings didn't take effect. I think I know why and I'll revise it tonight. Only setting I don't know what it is to change is the max package count. I cannot locate that.

 

[Retrorockit]

So 2 instances without renaming one file for the 2nd CPU?
Did you text edit to assign CPU cores to each one?
On the previous page I noticed that CPU#1 was at Turbo all cores, and CPU#0 was not. That seems interesting.

I like it when somebody moves the goal posts. It's even better when they move them both at once.

 

[frankr2994]

So 2 instances without renaming one file for the 2nd CPU?
Did you text edit to assign CPU cores to each one?
On the previous page I noticed that CPU#1 was at Turbo all cores, and CPU#0 was not. That seems interesting.

I like it when somebody moves the goal posts. It's even better when they move them both at once.

Things were getting glitchy leaving the Intel micro code in place when the multiplier got raised. It's like there were fail-safes that triggered every limit to try and slow it down whenever there was a load. And 1 CPU was getting stuff applied the other wasn't. Right now I believe hyper threading is on. And I'm not directly doing any of this. I'm assisting someone that lives who knows where lol. So I don't have all the answers just trying to report the progress being made .

 

[Retrorockit]

I saw 32 threads on the Cinebench in the background so HT must be on.
So you're playing with the CPU microcode and not the BIOS? A lot of times I though the Dell BIOS was getting blamed for stuff that was other places. Like locked PLL chips.

 

[frankr2994]

I saw 32 threads on the Cinebench in the background so HT must be on.
So you're playing with the CPU microcode and not the BIOS? A lot of times I though the Dell BIOS was getting blamed for stuff that was other places. Like locked PLL chips.

Playing with everything. With bios mods we got overclocking and voltage control. But limits set and wouldn't let clocks run high. Even with powercut. Microcode deleted chip runs at full stock turbo all core. Overclock and voltage control are gone now even with the bios mods. Trying to diag now.

 

[Retrorockit]

maybe a look at the microcode of an unlocked CPU might reweal somethingto splice into the 2 CPU code.

 

[frankr2994]

maybe a look at the microcode of an unlocked CPU might reweal somethingto splice into the 2 CPU code.

That gets in over my depth of understanding. The micro code I know of is part of the bios. Not directly tied to the chip. More of a chip series sorta thing. So any 2011-3 board should have the same microcodes regardless of unlocked chip or not. Other than some boards having updated versions of said microcodes.

 

[frankr2994]

Has anyone here went into the depths of bios modding yet? Anyone want to have a crack at why I can't unlock writing to the bios region only using fptw and cannot lift the overclock lock or enable higher power levels? This is back on my 5810. I was able to rather easily lift the overclock lock on that guys 7810. I still have to do a sleep resume to lift the lock and after that I'm still stuck with the same power levels as always.

Here is my modded file if someone wants to look at it.

unsuppress everything.rom

drive.google.com

 

[Retrorockit]

IDK how much of this is on the CPU now, But chips on the MB can be locked to not respond to the BIOS. So sometimes the PLL chip itself is locked and the BIOS has nothing to do with it. The lock command may be early in the BIOS code, or on the PLL itself. The locking bit (input) is often on an output pin so it can't be spoofed later after Power Good is achieved. A look at the Datasheet for the relevant chip may help. RWEverything can unlock some things. There can be a lock / unlock bit, and a separate frequency bit, and then an output to cover it up. I've seen these things in datasheets but I'm not good enough at programming to work with them. So you may find the PLL speed selector, but it may be locked at another place in the code. TME was one name for this feature. An unlocked PLL may have a different version number.

 

[frankr2994]

IDK how much of this is on the CPU now, But chips on the MB can be locked to not respond to the BIOS. So sometimes the PLL chip itself is locked and the BIOS has nothing to do with it. The lock command may be early in the BIOS code, or on the PLL itself. The locking bit (input) is often on an output pin so it can't be spoofed later after Power Good is achieved. A look at the Datasheet for the relevant chip may help. RWEverything can unlock some things. There can be a lock / unlock bit, and a separate frequency bit, and then an output to cover it up. I've seen these things in datasheets but I'm not good enough at programming to work with them. So you may find the PLL speed selector, but it may be locked at another place in the code. TME was one name for this feature. An unlocked PLL may have a different version number.

The hard locks are set at the very beginning of the code in the flash descriptor region. I Unlocked all of those. This gives me read write to everything except the bios region. I can only read from the bios region. The bios is locked with soft locks. Question IDs for bios lock, host flash lock and bios guard. I've found these but like all the other settings I apply straight to the chip they don't help me.

 

[frankr2994]

I got around the bios locks hahahaha. Going to take me a little bit to get it all sorted. Don't want to change too much too fast so I know what settings want to brick the bios and what doesn't. But as of right now I don't have to do a sleep resume cycle to overclock. Huge win. Now to start cranking these power limits.

Ok I have unlimited PPO current now. so there are NO limits to hit. however if I push past this...which this is the most I've ever gotten period my 5810 power cycles. no bsod. just goes into a reboot cycle and starts windows normally.

 

[frankr2994]

Ok today's project is going to be saving those multipliers and voltages straight to the bios so it just works on boot. After that going to see on overclocking the ecc memory and found options to change the pwm offset on the system fans.

Then....I'm going to apply most of this to a 7810 I just bought on the cheap. That is in prep to buy and run dual e5-2696v3s. So the Dell vrm limit seems to be around 200 watts. Because spikes happen so fast I need to stay at about 190. Prolonging over 190 is when I get a system reboot. I haven't found a way around that. 190 watts at 1.28 volts is 243 amps. I already knew the amp limit on the 6 phase vrm is 240. So I guess I'm happy I can max it out. Was hoping I was going to be able to push past and if it didn't bsod or anything I'd find the point where I fried a MOSFET on the board lol. IR power stages must be too smart for that.

 

[Retrorockit]

Ok today's project is going to be saving those multipliers and voltages straight to the bios so it just works on boot. After that going to see on overclocking the ecc memory and found options to change the pwm offset on the system fans.

Then....I'm going to apply most of this to a 7810 I just bought on the cheap. That is in prep to buy and run dual e5-2696v3s. So the Dell vrm limit seems to be around 200 watts. Because spikes happen so fast I need to stay at about 190. Prolonging over 190 is when I get a system reboot. I haven't found a way around that. 190 watts at 1.28 volts is 243 amps. I already knew the amp limit on the 6 phase vrm is 240. So I guess I'm happy I can max it out. Was hoping I was going to be able to push past and if it didn't bsod or anything I'd find the point where I fried a MOSFET on the board lol. IR power stages must be too smart for that.

On the old LGA775 there was a hard limit to the Voltage in the Intel specs. If the VRM detected >1.6V. it shut down. I could run at 1.5875V., but if I set 1.6V. any spike it would go to 0V. The VRM table settings ended at 1.6V. There were about 3 different VRM tables for LGA775 CPUs. You might look at the VRM tables and see if you're at that point.

 

[frankr2994]

On the old LGA775 there was a hard limit to the Voltage in the Intel specs. If the VRM detected >1.6V. it shut down. I could run at 1.5875V., but if I set 1.6V. any spike it would go to 0V. The VRM table settings ended at 1.6V. There were about 3 different VRM tables for LGA775 CPUs. You might look at the VRM tables and see it you're at that point.

I know I'm at a current rating limit for the mosfets themselves. I looked the datasheet up on them a long time ago. They have built in over current protection since they are a fully integrated power stage instead of just a MOSFET. 6 phase using 40 amp fets. I did look into replacing them which is obviously easier said than done. That family of fets has 3 different ones. My board was actually laid out for the next size up. I have 4mmx6mm fets. There are un used pads on the board for the 5mmx6mm which are 50 amp each. The next one 6mmx6mm are 60 amps but there isn't room or solder pads for those

I did however turn off fivr faults in the bios which is supposed to ignore over voltage and over current. It either didn't work or the fets themselves shut down.

 

[Retrorockit]

I know I'm at a current rating limit for the mosfets themselves. I looked the datasheet up on them a long time ago. They have built in over current protection since they are a fully integrated power stage instead of just a MOSFET. 6 phase using 40 amp fets. I did look into replacing them which is obviously easier said than done. That family of fets has 3 different ones. My board was actually laid out for the next size up. I have 4mmx6mm fets. There are un used pads on the board for the 5mmx6mm which are 50 amp each. The next one 6mmx6mm are 60 amps but there isn't room or solder pads for those

I did however turn off fivr faults in the bios which is supposed to ignore over voltage and over current. It either didn't work or the fets themselves shut down.

When you get that thing running as fast as it can go post it at userbenchmark.com. That will compare it only to other Dell T5810s. It's always fun to see the improvements there. It get's an overall score too compared to all computers, and you can break it down by CPU %ranking if you want.

 

[frankr2994]

When you get that thing running as fast as it can go post it at userbenchmark.com. That will compare it only to other Dell T5810s. It's always fun to see the improvements there. It get's an overall score too compared to all computers, and you can break it down by CPU %ranking if you want.

I've ran it a few times already. Have to re clock up the 1070 and get my memory speed up. Then it should be a pretty good result..... comparatively.

 

[Retrorockit]

Did you add any cooling to the MOSFETS. Dell typically leaves them bare, and you can find some headroom by adding heatsinks.

 

[frankr2994]

Did you add any cooling to the MOSFETS. Dell typically leaves them bare, and you can find some headroom by adding heatsinks.

I have a heatsink from an old Ibm server on them. I watched temps with a fluke thermocouple last night over a 30 minute stress. Stays under 50c.

I do have a heatpipe cooler that I machined an aluminum block for to mount on the vrm. It's not done yet.

So I milled that block of aluminum to fit down on the fets while also having contact with the chokes and the caps. Register on the top to fit the heatpipe cooler. Just never finished it. Needs about 4 small holes drilled and I need to buy a 2.5mm tap. Also got some .8mm copper to make small spacers to bring the top of the fets even with the small caps beside them to make good contact with the cooler. I guess I'll have to use a dab of thermal adhesive to hold them in place. If they slide to the side alot of stuff can short out.

 

[Retrorockit]

I would be careful about gluing it down. That big hunk of Al will expand and contract breaking the glue bond over time. I see a screw hole at each end of the VRM. Soft TIM to allow expansion, and a wire spring clip to hold it down. Thermal pads are the normal solution for uneven parts, but I see you've got that fixed already.
One of the big features of BTX was the VRM right at the fan before the CPU. It was part of the BTX spec. Maybe reverse the airflow rear to front and see what happens? The t3500 still had intake fan air direct to the VRM. Off to the side but still direct.

 

[frankr2994]

I would be careful about gluing it down. That big hunk of Al will expand and contract breaking the glue bond over time. I see a screw hole at each end of the VRM. Soft TIM to allow expansion, and a wire spring clip to hold it down. Thermal pads are the normal solution for uneven parts, but I see you've got that fixed already.
One of the big features of BTX was the VRM right at the fan before the CPU. It was part of the BTX spec. Maybe reverse the airflow rear to front and see what happens? The t3500 still had intake fan air direct to the VRM. Off to the side but still direct.

You misunderstand. I need to glue down the copper shim stock. The aluminum block was going to use a thermal pad and 2.5mm sprung heatsink screws through the bottom of the board. I have a 1.5mm thermal pad under the little heatsink I have right now. But it's .75mm difference in height between the fets and those little caps. Pretty big damn step as far as trying to get a heatsink to apply good pressure.

Reversing the airflow does give me an idea. On the 7810 I'm going to run a 2nd CPU heatsink on both chips. The first one is stepped to allow for the optical disc drive which is going away. Anyways the 2nd CPU always runs hot because the first sink pushes right into it. If I reversed the 2nd CPU sink so they push into themselves I could duct that straight out the panel cover so they would both be pulling ambient and would exhaust out of the case.