Throttlestop overclocking Desktop PCs

[frankr2994]

I will not be delidding. It's not worth it to have a better chance than not of ruining it. I do however have a decent supply of various chips that are worthless. Maybe sometime I'll play with yanking the lids

 

[unclewebb]

@frankr2994 - I search the forums day and night and I do not ever remember seeing POWER BUDGET light up red in Limit Reasons. I had a look at the Intel documentation for this and MAX TURBO.

Power Budget Management Status
When set, frequency is reduced below the operating system request due to PBM limit.

Multi-Core Turbo Status
When set, frequency is reduced below the operating system request due to Multi-Core Turbo limits.

That is useless information. No explanation of what causes either of these or what needs to be changed to try and fix the problem. I think you have hit the wall.

TCase is 66.7C

This specification is only for system builders so they can design heatsinks and system fans. It can only be accurately measured by embedding a temperature sensor into the top of the CPU. This is done by cutting a groove into the heat spreader on top of the CPU with a Dremel or similar tool. Intel does not want their retail customers hacking up their CPUs so they included core temperature sensors at the hottest spots on the core. The temperature data from these core sensors is the only information used to control thermal throttling. The TCase temperature cannot be easily measured so it is not used to control anything.

I will not be delidding.

Good idea. The lids on these CPUs are soldered on to the CPU cores. You are more likely to destroy a good CPU than accomplish anything productive.

 

[totalfreq]

@frankr2994

This specification is only for system builders so they can design heatsinks and system fans. It can only be accurately measured by embedding a temperature sensor into the top of the CPU. This is done by cutting a groove into the heat spreader on top of the CPU with a Dremel or similar tool. Intel does not want their retail customers hacking up their CPUs so they included core temperature sensors at the hottest spots on the core. The temperature data from these core sensors is the only information used to control thermal throttling. The TCase temperature cannot be easily measured so it is not used to control anything.

Agreed. Impossible to calculate without a sensor at the contact point but does it not usually fall within 10-20c of core temps? I know when I've entered the +10-20c range I'm approaching the point of diminishing returns and the end of the road is in sight.

And I understand you know a lot more than me so please correct if i am wrong.

 

[frankr2994]

Wow unclewebb you are a much better searcher than me. Atleast you got a description. I couldn't find that anywhere. With those descriptions I'll do a little bit more digging. I'm new to most of this as usually I'm just constructing oems for friends and family and don't go modifying anything for others. This is my rig so ya lol. And I don't think I have thanked you for your throttle stop software which is just outstanding. Completely destroys what I've found with xtu.

 

[Retrorockit]

Thanks for the comparo to XTU., and posting your newer CPU/Dell project here. A lot of people thought this was an X58 forum.

 

[unclewebb]

Intel included overclocking as an option for these Xeons but probably intended for this feature to only be available in the BIOS. That might be why XTU does not fully support the Xeons. Great to see ThrottleStop pumping some new life into an old CPU.

@frankr2994 - Thanks for sharing your E5-1650 v3 adventure. The Dell workstations with these CPUs are still expensive where I live. It might be a while before I find one at a decent price.

Did you try increasing or decreasing VCCIN 0.10V? I remember my laptop would get confused and would report lower power consumption when I increased the VCCIN. This trick allowed for more MHz while staying within the power budget. I took VCCIN all the way to 2.30V in the name of science. Surprisingly, that CPU still works fine.

 

[frankr2994]

I had to increase the vcci to 2 volts. I've been reading up on overclocking the I7-5820k which is sorta the same chip. I did increase beyond that and it didn't help. It seems to me with the limits presented that maybe it just flat out can't get enough power in. I may try heatsinks on the mosfets and vrm. I mean it was a calculated load of 155 watts and with 2v vcci that's what 77.5 Amps getting processed by that row of mosfets which seems crazy. Amps cause heat and heat causes resistance which increases the load rinse and repeat. I've currently been running settings for 4.5ghz all day at full load with folding at home. Came home and the computer was still up and running so I think I'm passing my stability test for 4.5ghz anyways.

The Dell workstations with these CPUs are still expensive where I live. It might be a while before I find one at a decent price.

They flooded the market here about a year ago. There are gaps in this stuff though. Over a year ago nothing was cheap and it's been about 3 or 4 years that the T3610 flooded the market. The T5810s are just about done. Lease return stock seems to be getting low so the prices will go up. I've been watching the T5820s for when they hit but I'm guessing 2 years until that happens. I never paid any attention to any of the dual cpu workstations because that to anything I could do is Overkill.

 

[totalfreq]

I've currently been running settings for 4.5ghz all day at full load with folding at home. Came home and the computer was still up and running so I think I'm passing my stability test for 4.5ghz anyways.

Folding is a solid test but have you tried linpack or prime95. I've found it's an excellent way to quickly shake out the bugs on system. Folding and Crypto Mining, add a steady load on the processors but it's more stabile and symmetrical. If the system isnt being used for gaming or 3d/video editing then you are probably fine, but if you need to render a large bim model or play GTA V you may hit a load that BSODs.

Just a suggestion if you havnt used those for testing. Have your core temps surpassed 77c?

 

[Retrorockit]

I've had good results heatsinking Dell MOSFETs. It does make more power available. Dell always seems to leave them off, probably for cost reasons. I found an extra .05V on a BTX with vertically mounted MOSFETs right next to the fan, so the cooling was already pretty good there.

 

[frankr2994]

Folding is a solid test but have you tried linpack or prime95. I've found it's an excellent way to quickly shake out the bugs on system. Folding and Crypto Mining, add a steady load on the processors but it's more stabile and symmetrical. If the system isnt being used for gaming or 3d/video editing then you are probably fine, but if you need to render a large bim model or play GTA V you may hit a load that BSODs.

Just a suggestion if you havnt used those for testing. Have your core temps surpassed 77c?

Prime 95 seems evil. I read that there were issues with my system and using it but it takes like 3 seconds for every limiter to go red and the computer shuts off. Now that was before folding which I actually did have to raise my voltage to get folding stable so I could try prime again. And max temp folding all day was 82c

 

[totalfreq]

Prime 95 seems evil. I read that there were issues with my system and using it but it takes like 3 seconds for every limiter to go red and the computer shuts off.

A xeon is built for 100% load for 1095 days straight so I highly doubt the person who used prime 95 had a solid system to begin with or was overclocking. Now when overclocking, if you cant offset AVX, use prime 95 version 26.6 to avoid the AVX voltage spike.

But if you are afraid of Prime95 have you tried linpack? US National Laboratories and large manufacturers like IBM use linpack as the defacto scoring system for supercomputers. If they were worried linpack would break their systems they definately wouldnt be using it on multimillion/billion dollar systems. Some of these systems run 60kW per rack node which is absolutely insane.

But I'm not encouraging you to break your computer. If folding is getting the stability you normally need then so be it. I've just found that when I OC my systems if I dont run a heavy CPU test like prime/linpack and a heavy GPU test, I'll find my computer BSODing in the middle of a huge raid in warcraft or playing witcher, gta v, CoD or doing the final render on a high poly count BIM model. If you dont game on ultra settings with max resolutions or are not an architect/engineer, then I'm sure your tests are sufficient. I would also recommend if you have mission critical data stored on the system you use a full burn in test or you may risk corruption down the road. I've had to send systems out for deep recovery before and 500gb of data will cost around USD$1500 to recover if it gets corrupted.

But I have to say this conversion really has me interested in seeing what my E5s can do. I have a T620 with dual E5-2667v2 sitting around and now I'm itching to see if I can even squeeze a few % out of it. As rocket said, thank you for exploring outside of the typical X56 we have been playing with in this thread.

 

[Retrorockit]

Could you run your T5810 at userbenchmark.com
There are only 4 examples of that system there but the E5-1650v3 is sampled there at about the same performance level as an OC T3500.
It's a very soft benchmark. I wouldn't be surprised if you could run it at 4.6GHz.

UserBenchmark: Dell Precision T5810 Compatible Builds

 

[frankr2994]

I will try linpak. Ultimately I'll probably game at 4.5ghz and otherwise it will probably be around 4.2 for actual work. My drives will also be a mirrored refs storage pool. Not fool proof but it's atleast it's designed to fight corruption.

And for user benchmark I did run it. A few times actually. I was comparing my xeon with others on the list. However I was not logged in and I can no longer find my score. At 4.6 I know one of my scores was 96.4%. there were slower speeds scoring better but the 100% belongs to someone that got it to 4.7

 

[totalfreq]

but the 100% belongs to someone that got it to 4.7

They may be using a liquid cooling solution, fish tank cooler or some other form...also I had friends way overclock a system to get the highest user benchmark/CPU-Z score while holding a can of dust off upside down and spraying the heatsink. They got the top scores but it was unsustainable.

For instance I can pass a cpu-z at 5.386ghz on an off the shelf AIO, but at those temps I'm risking catastrophic failure so I dont actually run there. So I wouldnt compare your 4.6 against the 4.7 and feel unworthy...ultimate benching and stabile overclocking are completely different things. It's possible they hit 4.7 stabile by just hitting the silicon lottery too.

Intel Core i7 8086K @ 5386.71 MHz - CPU-Z VALIDATOR

[icwqxj] Validated Dump by TotalFreq (2018-11-13 03:53:14) - MB: Asus ROG MAXIMUS X CODE - RAM: 16384 MB

valid.x86.fr

 

[Retrorockit]

Here's a better link for that system at Userbenchmark. Apparently the word"Tower"5810 is required to find them.

UserBenchmark: Dell Precision Tower 5810 Compatible Builds

And for user benchmark I did run it. A few times actually. I was comparing my xeon with others on the list. However I was not logged in and I can no longer find my score. At 4.6 I know one of my scores was 96.4%. there were slower speeds scoring better but the 100% belongs to someone that got it to 4.7

So this system has mid 90% CPU capability with a software overclock. certainly a nice bump up from the T3500.

 

[frankr2994]

ok not out of the woods yet. got linpak up and running. turns out prime 95 wasn't evil..... so I can run the long bench on TS 12 thread no problem. I can run F@H for 16 hours full tilt no problem. 3 seconds running prime 95 or linpak does those same 3 throttles in red. The only clock speed that it doesn't do that on is 3.8GHz which is the stock turbo speed. What I have noticed because its easier to catch at the lower speeds is the PKG Power. when only running 3.9GHZ power is like 80 watts and temp is around 65 running 100% load. all of a sudden PKG Power goes to about 158 watts my VID drops atleast .500 volts and all the throttles light up. I'm really clueless here because apparently from a stability standpoint I have only accomplish all core turbo which the computer typically doesn't do.

 

[unclewebb]

If you do not plan to run Prime95 then using 12 threads of it for testing is overkill. It is not unusual for it to put a load 30% or 40% greater on your computer compared to any other load you are likely to run. When throttling starts, the MHz and VID voltage will both drop to try to reduce power consumption. Try running a 6 thread test instead of 12 threads. Even 6 threads of Prime95 is a bigger load than most apps are going to put on your CPU.

 

[frankr2994]

4 threads on linpak still does it. at 2 threads its not an issue.

 

[totalfreq]

4 threads on linpak still does it. at 2 threads its not an issue.

Are you OCing bclk or just multiplier? If so are you more than 103mhz? I find most systems get funny if I go more than 2% up, some can get to 103mhz but past that is always unstable.

 

[frankr2994]

just the multiplier. I have no way of changing bclk.

I'm really starting to think this is a vrm issue. Didn't find any monitoring for them and don't have a laser temp gun only metallic thermocouple that I'm not going to move around my motherboard lol. The vrm is marked for a heatsink but doesn't have one I'm thinking I could go with little mosfet heatsinks or maybe an alienware one will work. But here's a few pictures. I yanked the cpu cooler and you can see that awesome contact patch....I'm going to replace that too.

And the vrm heatsink that I think may work https://www.newegg.com/p/2S7-05E8-012G6

 

[Retrorockit]

That's a very nice heatsink, but looking at the airflow in that system I think it will be sitting sideways in the air stream. Arctic Cooling offers a bag of assorted small heatsinks in their accessory section.

Accelero Xtreme III - Heatsink Set | ARCTIC

VRM and VRAM Heatsinks for Accelero Xtreme III and Twin Turbo II | ARCTIC Spare Parts | ARCTIC Service | ARCTIC Store

www.arctic.de

You could get them pointed into the wind.

 

[frankr2994]

Good point. Being so small I didn't even notice the direction of the cooling fins. As far as an oem heatsink I can't find them on any of the motherboards until I hit a 7920. That has a bunch of passive coolers and a big fan module similar to hp workstations. I'm willing to bet the vrm on all the other motherboards was provisioned to have a heatsink if liquid cooling was ever going to be an option which I believe you could order a 7810 with liquid cooling. I still may buy the alienware heatsink. If it's the right size I'll shave all the fins off of it and just use it like a heat spreader for universal heatsinks. I also want to use a cpu cooler off a t7810 which is larger and looks like it has a raised section to go over the vrm. It would be nice to actually tie those 2 together right there.

 

[Retrorockit]

Dell Typically doesn't put heatsinks on their VRM MOSFETs. Probably costs less, but if you add them it gains some head room for overclocking.
Most of the dual CPU systems, and some of the singles use a narrow ILM heatsink mounting. Most mainstream computers use the standard square Intel heatsink layout. Thermalright has some narrow ILM kits.
You might grab a Nidec M35105-57 fan. It's 92x38mm 130cfm and runs off of just 12v. No PWM. It's exhaust only due to thermistor control.
Hanging one off the back of the stock heatsink usually improves things quite a bit.They can be found for about $5.

 

[frankr2994]

Well because I was close on the thermals I wanted I'm going to try a t7810 2nd cpu heatsink. It's a decent amount larger and shouldn't require an adapters or fan connecters. They are also pretty cheap on ebay.

 

[frankr2994]

Ok I bought 2 things...or wasted money on 2 things. Got an alienware vrm heatsink..... doesn't fit. Got a dell t7810 cpu heatsink. Never thought it would be a square ilm especially being a dual cpu and my single socket is a narrow ilm. The heatsink is awesome though. It's big,has more heat pipes and even has a shroud to direct air onto the vrm. Does anyone know of brackets to put on a narrow ilm to fit a standard 1150 cooler? I'd lose 2 dimm slots which....bothers me a little bit not enough lol.