Throttlestop overclocking Desktop PCs

[frankr2994]

That must itch something fierce!


That would work well.

That's a common saying at least in the Pittsburgh area....looks screwed up typed out lol. And for those dual 2687s I think I could only justify that on maybe a new Plex server. Our 4th gen i7 and gtx1650 super can run about 14 simultaneous transcodes now on gig internet. Maybe if fiber internet makes its way to 2.5gb in the near future something like that system with a 3080 could probably do 40 or more transcodes. Otherwise unless I come into some nice play money (doubtful) it's just talk.

 

[frankr2994]

Ok proud owner of a 1680v3 now. Stuck it in and doing some early testing. So far she runs hot lol. Don't know where my voltage is going to end up but right now I'm running 1.23v at 4.0ghz all core with hyperthreading. My cooler is keeping it under 80c but not by much. I'm going to creep up on the multiplier and see how high that voltage will let me go. Hopefully it's like my 1650 which until I was in the 4.7/4.8 range I didn't really have to step up on the voltage.

Calling it for tonight. I'm not sure what's going on and I think I need unclewebbs input. I can get 4.3ghz out of it. I'm using the powercut feature and it's reading out that I'm using like 10 watts so its working however I had to go down to 1.20v now it at least in a short run seems stable but if I raise the voltage I get an edp current limit and it clocks down. I thought getting the powercut feature working solves that....I tried 4.4ghz but it hit that limit and took it to 4.3 so I pulled the voltage down to 1.19v and it crashed...so I'll get some screenshots together tomorrow. Really hoping I didn't just find the vrm limit of my board or something.

 

[lexluthermiester]

So far she runs hot lol.

Um..

right now I'm running 1.23v

..that's why.

4.0ghz all core with hyperthreading.

You could likely get that at 1.1v. Maybe 1.12v. But 1.23v for that chip is crazy high..

 

[frankr2994]

From what I can find people take the E5-1660v3 up to about 1.4v. I can't find any real info on the 1680v3 because its been too expensive in the past for people to play with. It really should be the same chip just binned higher. Anyways I'll deal with cooling somehow someway its whatever. at 4.3GHZ dropping the voltage down to 1.19 crashed the system. so I'll need to push some at at. I now have a few screenshots here with what I have goin on and my settings @unclewebb if you wouldn't mind taking a look. I took these screenshots while Cinebench R23 was running multicore.

 

[lexluthermiester]

4.3GHZ dropping the voltage down to 1.19 crashed the system.

What happens at 4ghz?

 

[frankr2994]

What happens at 4ghz?

I can volt down like you suggested. Temps are fine. I can run 4.0 all day. Seems I can run 4.3 all day. 4.4 isn't happening right now. I'll attack temps as soon as I see I "need" to. My goal is to run 4.6ghz+. If I can't I'm faced with selling this chip and going back to a 1650v3 or buying a x99 motherboard that will let me use it...

 

[unclewebb]

@frankr2994 - Can you try using ThrottleStop 9.4.6? I have never had my hands on an E5-1680 v3 so I am not sure how well ThrottleStop supports these CPUs. With the new version, go into the TPL window and check the MMIO Lock box near the top right. This will do a few things. It might help unlock the maximum multiplier. With some more voltage, you might be able to use the 44 multiplier across all cores.

After you boot up, if you see that the BIOS has locked the PP0 Current Limit register to 208, you cannot change it to 400. This request will be ignored.

Looks like you are having fun. Post some more ThrottleStop screenshots so I can see what needs to be fixed to better support your CPU.

If TDP Level is not locked when you first boot up, I would suggest setting that to Level 0. You have a lot of the the Lock boxes checked so I am not sure if you or the bios has locked some of these settings.

 

[frankr2994]

@unclewebb
Downloaded new TS version deleted config file and started over. Here are screenshots prior to any adjustments, adjustments made, and throttles that occur.

And forgot to check the MMI0 lock box. Just went in and tried to do it I cannot check it.

 

[lexluthermiester]

I can volt down like you suggested. Temps are fine. I can run 4.0 all day. Seems I can run 4.3 all day. 4.4 isn't happening right now.

That might be silicon lottery going on. With what you described, you should be able 4.1ghz all-core at 1.2v, which is a very solid boost and won't cause a lot of heat.

 

[frankr2994]

That might be silicon lottery going on. With what you described, you should be able 4.1ghz all-core at 1.2v, which is a very solid boost which won't cause a lot of heat.

And your right. I can. Temps stay low 70s. But thats only slightly better than I can get out of a 1650 which is 1/3 the cost. Money means more to me. I like to play with this stuff but if it's actually not "better" then it's wasted money. And actually with the clocks I'm getting a again much cheaper 1660 would do what I'm doing just fine. Those are about half the cost of a 1680.

 

[lexluthermiester]

But thats only slightly better than I can get out of a 1650 which is 1/3 the cost.

Yes, but also fewer cores. the 1680V3 is an 8c/16t model whereas the 1650 is only 6c/12t model.

And actually with the clocks I'm getting a again much cheaper 1660 would do what I'm doing just fine. Those are about half the cost of a 1680.

That might be true, but silicon lottery is always at play. You might get a 1660V3 that can't clock as high. A bird in the hand is worth 2 in the bush. The point being you have a solid CPU. Set that 4.1ghz/1.2v and call it good and done. There is value in being happy with want you have.

 

[frankr2994]

Your very wise. However I will continue my quest on trying to melt/burn/explode something on my Dell before I call it quits lol. And besides your going against America's moto of never be happy with anything. There's always more and better. All joking aside I do appreciate the info and I do understand that what I have going on is good enough. If I needed an 8 core daily for work I'd set it there and be thrilled. But since I don't really need this PC anymore it's just a play thing that gives me a reason to hide in my office for a bit.

@unclewebb
Don't know if this is any help to you but I made a log file for a 10min Cinebench run.

 

[lexluthermiester]

However I will continue my quest on trying to melt/burn/explode something on my Dell before I call it quits lol.

That is just sad..

 

[frankr2994]

Ok well anyways I tried my hand at lapping for the second half of the day. I did not use my new 1680 for this though. Kinda hard to kill a heatsink so I did use my good one and an old e5-2640v3 cpu which I think id have to pay someone to take. So if it's shot oh well. I have it in a bowl of rice right now because I'm not confident that I didn't fill it with water. I watched a gamers Nexus video where he and king pin lapped a cpu. King pin put a dab of grease in the IHS vent hole. So that's what I did. Pretty sure it washed out. Other thing that screwed me up was keeping sandpaper attached to the glass. I used gorilla tape and all of it worked loose with the water. I may try lapping compound straight on the glass with another cpu to see how that goes.

Ok testing today. Running prime 95 small fft core difference before 8c max temp 60c. Today core difference 6c max temp 59c. Now my fans don't ramp up until over 70c and obviously I can't make this chip really push anything but it's nice to see I didn't ruin anything and that there is some......small gain. I'm anxious to see how nice the tim is spread out when I pull it back apart. On either cpu it was very obvious with high pressure spots.

 

[Retrorockit]

I just dry sand them with the 3m assortment that goes up to about #2500 I think. Start with#600-800 to get it flat, then remove the scratches step by step down to #2500.
You should end up with a mirror finish. It's a strange thing to install your CPU and see the stucco on the ceiling reflected there.

 

[frankr2994]

Both the cpu and the heatsink took a while with 320 grit to get flat. Had to switch out the 320 when I did the cpu. The nickel finish was hard and wore the sandpaper out. I finished with 2000 grit. It was obvious to me that I could have spent more time but as it was it took me about an hour and a half to do both. Which I found ridiculous. Like I said I'm going to practice some more with some old core 2s I have lying around. When I find a process I like I'll do the 1680 and touch the heatsink up to give them both a mirror finish.

 

[Retrorockit]

The sandpaper strips can be taped to a mirror ( I use a long dressing mirror) on the workbench. I tape them down in a progression, and just move along from strip to strip. If you start with too coarse it's a bunch of work to remove the deep scratches, too fine and it takes extra time. I would start with #800, and if it's fairly flat IHS go from there. If there are high spots then #600 just for those. Don't scratch up the whole CPU over some high spots. You don't have to take it down to bare copper, it can be flat with some nickel still there in places. Some people think you're trying to make it thinner. That's not the object of it at all. Flat metal to metal contact for the thinnest TIM layer possible.

 

[frankr2994]

I have an old all glass table out in the garage. Not being used anymore. I used a table leg thats approximately 40"x14" and about 5/8" thick. I taped all my paper down like you said and just moved down the line. Like I said the water ruined the connection of the gorilla tape to the glass so that made it a pain.

 

[Retrorockit]

I MESSED around with water too. Just gave it up completely.
I found 3*C with a polished finish, but I was pushing 1.5875V. through a smaller LGA775 IHS.

 

[unclewebb]

Don't know if this is any help

The log file helps a lot. It shows what is going on.

   DATE       TIME    MULTI   C0%   CKMOD  BAT_mW  TEMP    VID   POWER
2022-04-16  18:07:38  43.00  100.0  100.0       0   72   1.2241    7.2   EDP
2022-04-16  18:07:39  43.00  100.0  100.0       0   73   1.2241    7.2   EDP
2022-04-16  18:07:40  43.00  100.0  100.0       0   73   1.2241    7.2   EDP
2022-04-16  18:07:41  43.05   92.6  100.0       0   58   1.2241    7.2
2022-04-16  18:07:42  44.00   45.9  100.0       0   69   1.2241    7.7
2022-04-16  18:07:43  44.00   96.1  100.0       0   60   1.2241    7.4
2022-04-16  18:07:44  43.02   98.6  100.0       0   71   1.2241    7.2   EDP
2022-04-16  18:07:45  43.00  100.0  100.0       0   71   1.2241    7.2   EDP
2022-04-16  18:07:46  43.00  100.0  100.0       0   71   1.2241    7.2   EDP

When running Cinebench, after each scene is finished, the load on the CPU (C0%) briefly drops as Cinebench restarts and gets ready to draw the scene again. When the load drops, EDP throttling stops and the CPU multiplier goes back up to full speed, 44.00. When full load resumes, EDP throttling resumes and the CPU is once again limited to the 43.00 multiplier.

ThrottleStop shows that the PP0 Current Limit is locked to 208 by the BIOS. Without a modified BIOS that unlocks the current limit, I think you have hit the wall. Try running a less demanding benchmark test like 2, 4 or 8 threads of the TS Bench test. At some point, less load should result in less EDP throttling and the CPU should run consistently at full speed. ThrottleStop might report a steady 44.00 multiplier during a light load test.

 

[frankr2994]

@unclewebb
I was hoping that powercut got past that 208 limitation. I guess not. Oh well. Once again greatly appreciate the input. I'm getting my laptop back today that has all my bios tools on it. My parents have had it as a loaner for a while. I have an idea on how I'm going to fix this. And yes I was able to run 4.4 and 4.5 at a lighter load. 4.6 wasn't happening as most any load hit the throttle. Voltage had to be in the 1.27 range for that speed.

 

[unclewebb]

And forgot to check the MMI0 lock box. Just went in and tried to do it I cannot check it.

Just to confirm. When you try to click on the MMIO Lock box, does nothing happen? Does a check mark not show up in that box? It will probably not solve your throttling problem but you should be able to check that box.

PowerCut

PowerCut reduces the reported power consumption. I guess it does not do anything to change the reported current consumption. It definitely seems to be the locked current register that is holding you back.

Not sure if you can mod the bios. MSR 0x601 - bit[31] is the lock bit. That bit needs to be cleared so you can increase the current limit.

 

[frankr2994]

Just to confirm. When you try to click on the MMIO Lock box, does nothing happen? Does a check mark not show up in that box? It will probably not solve your throttling problem but you should be able to check that box.

View attachment 243892

A check mark does not show up so correct nothing happens.

I did test another scenario. Turbo power limits. I can set them to whatever and hwinfo shows that they have indeed been changed. However I get a throttle even at the stock 3.8ghz turbo speed. Powercut is the only way past that which you've seen where that throttles. But touching anything in turbo power limits accomplishes absolutely nothing. I can leave them alone and use power cut to achieve the same results.

Not sure if you can mod the bios. MSR 0x601 - bit[31] is the lock bit. That bit needs to be cleared so you can increase the current limit.

I have not fully tested as in installed this mod but with my bios tools I can change things marked as default using a hex editor. It took some practice but after a while I got it to show up correctly I'm amibcp. If I can set default to a different value (almost everything has multiple values in the raw bios) I should be able to go up to another limit. It will also be locked but should... should be better.

 

[unclewebb]

hwinfo shows

Some computers use a third set of turbo power limits that are controlled by an embedded controller (EC). ThrottleStop does not have access to these power limits and HWiNFO does not report what they are set to. Increasing the MSR or MMIO power limits will be ignored if the EC power limits are set lower. The lowest set power limit wins control of the CPU. PowerCut is a good trick to keep power consumption low. TS 9.4.6 finally saves the PowerCut status so it should automatically re-enable PowerCut during a sleep resume cycle. It waits for 10 or 15 seconds so the CPU has a chance to settle down before toggling the VCCIN lock bit. If you have the FIVR window open, you should be able to watch this happen after you resume. Interesting to see the PowerCut feature still being used. Someone with a low power 5th Gen Broadwell U series CPU recently contacted me to tell me about his success story while using PowerCut.

A check mark does not show up so correct nothing happens.

Thanks for the info. I had a look at the code but I cannot see anything obvious why you are being prevented from checking that box. I will look some more but even if you could check this box, it will not be able to help. You need to find a way to unlock the MSR current limit or set it higher.

This register is in units of 1/8 Amps. 208 Amps means that register should contain 1664 which converted to hex is 0x680. Find that hex number and change it to 0x980 and that should give you some more headroom.

 

[frankr2994]

@unclewebb
Much appreciated. Your an awesome person. Just fyi I know that playing with this old stuff isn't really why you have this software....or even desktops in general but I'd gladly let you remote into any system I have and play as long as you'd like. Just let me know and I'll pop another drive in and make it a barebones fresh install.

Ok got my laptop back and my files. So I pulled one of my dump bins up in amibcp and found the PP0 current. Now I can't make changes here I should just be able to see if they have been made here after I'm done. When I select the current setting it lets me know on the bottom that if set to 2048 that the value is unlimited. can you tell me what the hex value is for 2048? Since I've started this I cannot figure out what hex means what. I find a word or number somewhere else that I can see what the hex value is for it and use it to copy over to make changes.

Ok So I won't be able to test this for a couple days. I have alot to get hooked up again to start programming a modded bios but I think I made the change I needed. I extracted the module from the bios. Opened it up in IFR extractor to read the text and get the hex values, opened it up in a hex editor and searched said values and made the changes to hex, saved the file and opened the new file in IFR extractor. here is the result.

How Dell has ami make this bios is beyond me. Very difficult. The only area in the bios I can find current is there. I changed it like I shown and it didn't show up different using amibcp. So now I'm looking at the beginning of that code that says "suppress if" what it's referring to is current manual override above it. I'm hoping after I change the hex for that it will show up different. What I don't understand is if that part of the code gets suppressed where the hell do they get a value for the current limit otherwise? I'd love to find that "hard code" area of the bios. I have a feeling that the end of the bios marked with pages and pages just saying Intel reference over and over is encrypted code that overrides all the changes I'm trying.