Throttlestop overclocking Desktop PCs

[lexluthermiester]

have seen that you talk mostly about DELLs here. As I also have one T5500

Throttlestop works on any system from any OEM that has a CPU with an unlocked multi. Dell's just happen to be the most popular for various reasons.

 

[ManGupta]

Has anybody got any suggestion for coolers upgrade without removing the HDD/SDD bracket?

If you are using this CPU Cooler on one of the CPU then you need to replace it first with the copper heatpipe one.



like this one (Dell 0U016F)

or this one (Dynatron G17 )

I use Nidec M35105-57 fan attached to he back of the cooler (it costs $2.4 on Aliexpress) ......... you can go through the post on Page 24 & 29 of this thread.

7.78US $ 10% OFF|TA350DC M35105 58 NIDEC 92*92*38mm 9238 9cm 12V DC 1.8A 21.6W 130CFM 5700RPM Schneider Inverter Axial Cooling Fan|HVAC Systems & Parts| - AliExpress

Smarter Shopping, Better Living! Aliexpress.com

www.aliexpress.com

In case of T3500, I use extra fans to cool the NB & SB, but T5500 already have active chipset cooling.

 

[Retrorockit]

The Nidec M35105-57 fan attached to he back of the cooler will help a lot. It's self regulating so just 12V. and Grd. will get it running.
Mangupta runs it in India without A/C and reports good results. ModDiy has them for $20 new, but it's just an old Dell Dimension Pentium 4 era CPU fan. So they can be found for scrap prices with a little digging around. Due to the thermistor control they need to be outlet only from the CPU
On page 24 there is a Dynatron G17 cooler swap shown on a T5500 MB. It fits the footprint of the stock cooler.

Throttlestop overclocking Desktop PCs

Always wanted an older Mac Pro...Where'd you put the pics?

www.techpowerup.com

I don't have the 2nd CPU riser to play with so you're on your own there. If it works for that also please post photos here, or in the Dell Workstation Owners Club.

 

[KLiKzg]

If you are using this CPU Cooler on one of the CPU then you need to replace it first with the copper heatpipe one.

View attachment 166506

like this one (Dell 0U016F) View attachment 166507 or this one (Dynatron G17 )View attachment 166557

I use Nidec M35105-57 fan attached to he back of the cooler (it costs $2.4 on Aliexpress) ......... you can go through the post on Page 24 & 29 of this thread.

7.78US $ 10% OFF|TA350DC M35105 58 NIDEC 92*92*38mm 9238 9cm 12V DC 1.8A 21.6W 130CFM 5700RPM Schneider Inverter Axial Cooling Fan|HVAC Systems & Parts| - AliExpress

Smarter Shopping, Better Living! Aliexpress.com

www.aliexpress.com

In case of T3500, I use extra fans to cool the NB & SB, but T5500 already have active chipset cooling.

Thanks, though Nidec is not available in Croatia, EU. So will need another one to get, unless I go over internet.

The Nidec M35105-57 fan attached to he back of the cooler will help a lot. It's self regulating so just 12V. and Grd. will get it running.
Mangupta runs it in India without A/C and reports good results. ModDiy has them for $20 new, but it's just an old Dell Dimension Pentium 4 era CPU fan. So they can be found for scrap prices with a little digging around. Due to the thermistor control they need to be outlet only from the CPU
On page 24 there is a Dynatron G17 cooler swap shown on a T5500 MB. It fits the footprint of the stock cooler.

Throttlestop overclocking Desktop PCs

Always wanted an older Mac Pro...Where'd you put the pics?

www.techpowerup.com

I don't have the 2nd CPU riser to play with so you're on your own there. If it works for that also please post photos here, or in the Dell Workstation Owners Club.

That Dynatron is not also available in Croatia.

So far only found Noctua NH-D9DX i4 3U, which is supposed to be under 110mm for 1st fan.
But it's about $90 & for that price I can get water cooling - which is way more better & efficient!

So now, thinking about this Cooler Master Hyper T20, but not sure if will it fit to T5500, as it's not for 1366 processors.
Which is only TDP130, but I have 95W processors...& could even try it without the fan.

Though I'm more concerned about the 2nd one, as that one has so much smaller fins, that it gets clogged frequently. ‍♂
Anybody tested something on 2nd CPU on T5500?

 

[ManGupta]

So now, thinking about this Cooler Master Hyper T20, but not sure if will it fit to T5500, as it's not for 1366 processors.


Though I'm more concerned about the 2nd one, as that one has so much smaller fins, that it gets clogged frequently. ‍♂
Anybody tested something on 2nd CPU on T5500?

Cooler Master Hyper T20 have a Clip Mount System. If the CM CPU Holder Base doesn't fit you will have to search for another one on internet. On Aliexpress you can get it for 1-2 USD.

2.28US $ 7% OFF|Desktop All Models 3in1 Cpu Cooler Fan Bracket Heatsink Holder Base For Lga775 1150 1156 1155 775 1366 Or 2011 Am2 Am3 Am4 1700 - Pc Components Cooling & Tools - AliExpress

Smarter Shopping, Better Living! Aliexpress.com

www.aliexpress.com

Regarding 2nd CPU Cooler, If this is your 2nd CPU Cooler, it should be fine.

It is a cooper Heatpipe cooler with Fan attached. Most of users of T5500 reports that it works fine. T5500 is non overclockable , so the Copper Heatpipe Cooler with 80-90 mm fan should be adequate , even with 24 X 7 Use. Issues like cleaning and maintenance apply on any system - pre-built or custom made.

Most of the T5500 users report issue only with this cooler.

. Even if you replace it with this cooler

and attach a 80-90 mm fan to it, it should be adequate for a Non-Overclockable System. Because most of T3500 users with W3680 find this cooler (with a 80-90 mm fan attached) adequate even for Overclocking.

 

[Retrorockit]

Many overclockers just attach any 80-92mm fan to the heatpipe cooler and report good results.
There are many other fans available. The 92mm fans come in 25mm,32mm, and 38mm thickness. The thicker fans can move more air at lower rpm and can be quiet due to that. The 2nd CPU fan can be a 4 wire PWM, and run off of the fan header. But Dell uses their own connector and pinout.
The added 1st CPU fan should be thermal control. These are 3 wire fans with a usually blue thermistor visible on the motor because it sticks up into the airflow. The big Nidec was the cooling fan on higher powered Dell Dimension socket 478 computers. Many of them have been discarded over the years. This being the overclocking thread I tend to show the biggest stuff that will fit. But there are a lot of other options for your situation.

 

[KLiKzg]

Cooler Master Hyper T20 have a Clip Mount System. If the CM CPU Holder Base doesn't fit you will have to search for another one on internet. On Aliexpress you can get it for 1-2 USD.

2.28US $ 7% OFF|Desktop All Models 3in1 Cpu Cooler Fan Bracket Heatsink Holder Base For Lga775 1150 1156 1155 775 1366 Or 2011 Am2 Am3 Am4 1700 - Pc Components Cooling & Tools - AliExpress

Smarter Shopping, Better Living! Aliexpress.com

www.aliexpress.com

Thanks, though the item cams with all parts in: https://edigital.hr/univerzalni-int...r-hyper-t20-ventilator-rr-t20-20fk-r1-p636037

 

[ZeusAlpha]

W3570/80 i7-965/975 are the unlocked 4 core 8T 45nm, and w3680/90 are the 32nm 6 cores 12T. i7-980X/990X also. That's it for LGA1366 The Xeons are 1333 RAM, the i7s 1066.
LGA775 is all Core2 extremes 65nm X6800,QX6700/6800/6850 45nm QX9650, and 400fsb QX9770 and LGA771 QX9775 unlocked SMP. There are some mobile Core2 extremes also.
ASFIK if it boots an unlocked CPU TS can OC it. I haven't tested anything newer. I see a lot of people with 2500K in Dell Optiplexes but nobody seems to be trying this. A lot of computers keep the unlocked CPUs out with a CPU Wattage limit in the BIOS, or Chipset, and some OC will be limited by a weak VRM that won't support Voltage increases. I always check userbenchmark.com for unlocked CPUs running before I would buy one.
PM me if you want to do the Macho 120. I ended up making a drill fixture to get the holes in the right place. I underestimated it the first time I tried and it bit me.

Hi Retrorockit, I realize this is a very old thread, but I hope you are still around and doing well. I have several T7500's and am very impressed by your work and guts. I've been searching all over looking for an "aftermarket" solution that wouldn't require drilling out the motherboard, but it seems that's what you resorted to? I'd love to learn more about your technique to do the Macho 120. I already ordered some U016F and plan to add a 90mm fan, which I've read works decently well. I'm also curious if you could help point me in the right direction for the VRM heatsink, which I have on some of my D881F's. But I can't find the part number for this anywhere. I've been considering slapping something in, but I'd really like to find the right ones if they can be found for a fair price. Thanks in advance. This is my first post on this forum, but I've been tinkering with and repairing PC's since the first generation 8086.

 

[ManGupta]

Hi Retrorockit, I realize this is a very old thread, but I hope you are still around and doing well. I have several T7500's and am very impressed by your work and guts. I've been searching all over looking for an "aftermarket" solution that wouldn't require drilling out the motherboard, but it seems that's what you resorted to? I'd love to learn more about your technique to do the Macho 120. I already ordered some U016F and plan to add a 90mm fan, which I've read works decently well. I'm also curious if you could help point me in the right direction for the VRM heatsink, which I have on some of my D881F's. But I can't find the part number for this anywhere. I've been considering slapping something in, but I'd really like to find the right ones if they can be found for a fair price. Thanks in advance. This is my first post on this forum, but I've been tinkering with and repairing PC's since the first generation 8086.

Do you use app like Real Temp / Hardware Monitor/ HWinfo ? What CPU temps are you getting Now ? Are the temps in alarming Zone ?

How do you measure temps of VRM ? Do you use Infrared Thermometer ?

You can search for Retrorockit's earlier post on Dell Workstation Owners Club thread. The one large Aluminium HS, the pic of which you posted may have issue of expansion & Contraction due to heat resulting in the contact with mosfets becoming loose over time.

Retrorockit had suggested many smaller copper HS to apply individually on each mosfet separately like Alphacool smaller Copper Heatsinks which is also used on GPU RAM.

Dell Workstation Owners Club

But The RDIMM question has merit. The more I read about systems that are socket 1366 based, some of which can use RDIMMS, the more it seems likely that this may be doable. DDR3 RDIMMS are cheap right now. Thinking I might try this with 1rank and 2rank modules and report the results.

www.techpowerup.com

Dell Workstation Owners Club

Internal Speaker for giving Diagnostics Beep Codes This.

www.techpowerup.com

Dell Workstation Owners Club

Internal Speaker for giving Diagnostics Beep Codes This.

www.techpowerup.com

 

[Retrorockit]

The Thermalright hardware works with the Dell heatsink mounting bosses. It's the same thread, no drilling required. Just a little adaptation for height.
be careful with the VRM heatsinking. Some of the components next to them are a little taller, especially on the 2 CPU machines. Also all the VRM are not the same height. A bunch of small individual heatsinks is one solution, the other is a thick enough thermal pad to clear the components and even out the level of the MOSFETs. The T7500 had an even taller cooler available due to the HDD being mounted next to the PSU and not over the CPU1 heatsink. U402F it doesn't get mentioned much because I believe it's too tall to fit in the T3500/5500 cases.

https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Fsc02.alicdn.com%2Fkf%2FHTB1f6wcRpXXXXb4XXXXq6xXFXXXq.jpg&f=1&nofb=1

 

[KMO_2000]

I recently decided to finally do something nice to my old Asus P6T Deluxe with Core i7 920, and started off down the six-core Xeon rabbit hole, so found my way here. I've spent many hours reading this thread - great stuff. My thanks to all the contributors.

I went with an X5675 initially, but the way it hit the locked 95W TDP limit regularly and wouldn't hold max turbo bugged me, so I decided to go nuts and give the thing a W3690.

I'd always wanted to play around with an unlocked CPU - never did particularly like the "brute force" BCLK approach, so was interested in trying a "softer" overclock, using the multipliers and turbo as designed, retaining the option to go slow. So I started experimenting with BCLK=133MHz.

Learnt lots of interesting things in this thread, but also picked up a number of other tips I think it would be worth sharing. Maybe a few years too late for the X58-specific bits, but hey... Most of this should be applicable to locked X58 T3500-type platforms, and I think some remains applicable to current multiplier-unlocked CPUs. Work done on Windows 7, but I don't think much has changed in Windows 10.

First point is that a BCLK overclock forces the core to go faster, all the time. Whereas if you're using Throttlestop and turbo multipliers, you have to kind of persuade your machine to enter the "turbo zone" to use your increased multipliers. All sorts of subtle settings become important.

One of the most important is the rarely-mentioned "energy-performance bias" MSR register in the CPU, which I've seen very little discussion of, outside of some Linux folks. (I've been monitoring and messing with MSRs using the RW-Everything utility).

Windows sets this energy-performance bias from the selected power plan. Balanced sets it to "5", at which point the W3690 seems very reluctant to enter turbo state given just a single-thread load. Even with Windows telling it to enter power state "27 T" (as Throttlestop would call it) via the PERF_CTL MSR, PERF_STATUS usually reports back at "26", and you see x26.0 in Throttlestop. No Turbo. It's possible that this is due in part to the way Windows smears a single thread across multiple cores - no single core is loaded 100% long enough to trip the "turbo" threshold, perhaps. OS keeps requesting 100%/turbo, CPU can't be bothered, despite there being a flat-out single-thread workload to process, hopping between the cores. An apparently horrible Windows-vs-Intel-Turbo interaction. I gather Linux works better.

When you set Windows to "High performance", it sets the "energy-performance bias" MSR to "0" - ie maximum performance. At that point the W3690 actually seems to ignore PERF_CTL, ie any request for lower processor state, and PERF_STATUS shows it always as "27 T". Multipliers only drop when idle. And Turbo is always attempted. Yay!

(It's important to know that the Power plans have a lot of hidden settings - not just the ones you can modify by default. They are different. You can expose a lot of settings with powercfg commands, but some like the CPU "energy-performance bias" seem to be totally hidden and hard coded: "High performance"=0, "Balanced"=5, "Power saver"=7. Manually poking the register with RW-Everything confirmed that this is the key make-turbo-work difference for "High performance", not any of the other setting differences).

The next point I found important is to actually use the per-core multiplier settings. You've got an unlocked core, so do what Intel do by default, and run fewer cores faster. You don't have to limit yourself to your 12-thread stable speed.

In my tests (with auto voltage and standard Vdroop), I established that I was linx/prime95 stable at x29 (3.87GHz), but with immediate linx death at x30 (4GHz). I was then able to get quite a bit better on fewer cores; apparently stable stock voltage multipliers were 29/29/30/30/31/32. I can single-thread linx or prime95 happily at 4.27GHz, stock voltage. Similarly, overvolting (Vcore 1.4125V, standard droop, CPU PLL 1.86V), I got stable with 32/32/33/33/34/35, so 6-core 4.27GHz, 1-core 4.67GHz.

I believe to some extent this arises naturally from the Vdroop. As the number of active cores decreases, current goes down, voltage goes up, so maximum speed increases. And obviously 1-core 4.67GHz is much less of an overheating problem than 6-core would be...

(Stability testing to establish fewer-core multipliers was using 1-4 threads in linx and prime95. I guess for full robustness I should shut off physical cores in the BIOS or msconfig to let me load 2 threads per core with fewer cores, but in practice the Windows scheduler should never be loading a second thread on a core until all 6 cores are active anyway. If you've activated a fewer-than-6-core multiplier, you're basically not hyperthreading. I've also not done corresponding testing with LLC compensation - my guess is that there is less to be gained.)

Next, with a 400MHz difference between 6 core and 1 core, you need to actually get into the fewer-core states as much as possible to keep the multiplier up. Again, the Windows scheduler doesn't seem to help - its tendency to spread tasks across all cores tends to push you towards lower multipliers with too many cores active. One tip I found for Windows 7 (maybe not needed for Windows 8+) was to mess with powercfg to unlock the "Processor performance core parking core override" setting, and disable it. When enabled (as by default) Windows 7 always keeps 1 thread per physical core unparked, which encourages the scheduler to spread work across all 6 cores, reducing your turbo multiplier. Once disabled, it will fully park some cores, reducing the number of cores that low-thread loads are spread across.

This seems to work in theory - at least it makes my Throttlestop average multipliers more like "34.6" rather than "34.1", for example - but whether it helps in real life scenarios is harder to say. Should we squeeze extra work into the 4th core, competing with its other users, or do we slow all 4 cores a bit and bring a 5th online for the new work? And I understand there can be latency effects of the parking/unparking, and many like to disable it.

This little chart of all my stable speeds nicely illustrates the difference between a BCLK and a turbo multiplier overclock. The 920 and 5675 were BCLK overclocked, and their +2 or +3 turbo boost is fairly incidental, if you even bother to leave turbo on. But if you're pushing an unlocked chip without raising BCLK, like the W3690 here, the difference between base and turbo frequency becomes huge, and you really have to make sure you don't end up pootling at 3.5GHz.

Given the above - the W3690 does have one minor advantage over the cheaper W3680 - the locked base multiplier is one step up. So if for whatever reason the turbo does fail to engage, you're left on a faster base clock. That means you may often get better performance in "Balanced" or "Power saver" modes, and there may be a slight responsiveness benefit there. But once you've got the turbo engaged in "High performance" with overclocked multipliers, there's no real difference.

Another tip is to set a sensible TDP limit. You don't have to set it to "high enough to never lower the multiplier". Set it to "high enough to never limit in a real interactive application". Letting maniacs like linx or prime95 get power-limited seems reasonable, if it's going to stop your CPU frying and your fans going berserk. My tests showed that 150W was sufficient for sustained 4.27GHz in Cinebench, and stopped linx getting close to 90C, so I've left it there. (Just raising it when I want to stress it).

Given all of this, a BCLK overclock is certainly more consistent for peak performance, but the multiplier overclock turns out to be great fun with just as many knobs. Maybe can't quite reach the heights of a BCLK overclock, but perhaps a better daily drive.

 

[lexluthermiester]

I've spent many hours reading this thread - great stuff. My thanks to all the contributors.

I think I can speak for everyone here when I say Welcome to TPU! Glad you found our discussions helpful! Looks like you had some fun!

 

[Retrorockit]

Thanks for that comparison. Most people take an either or approach to this, but you've done the research on BCLK, VS TS on an aftermarket MB.
Many of us here use TS on locked BIOS computers and workstations which can be very affordable. I've felt for a while that an overclock on 1 or 2 cores would be better if you don't need 12 cores.
FWIW it's possible to edit TS in a word processor to only apply to 2 cores # 0,1 .
It's nice to have a knowledgeable BCLK overclocker give TS overclocking a good comparison test.

 

[KMO_2000]

Been doing a bit more stress testing. I think my default-voltage 29/29/30/30/31/32 was a bit optimistic for the fewer-core multipliers. Had a couple of failures in normal use, despite no problems in the stress tests. Still working on it.

Looked into shutting cores down for maximum stress, so I can run all-thread linx/prime95 on fewer cores. There are 2 apparent ways to do this - set the number of processors to 8 in msconfig, or set the number of cores to 4 in the BIOS. For some reason limiting the processors in msconfig stops turbo ever activating. Don't know why, and seen no discussion of it. But if you set to 4 cores in the BIOS, you can turbo fine on the 4 cores. And that did show 8-thread 4-core x30 linx wasn't stable. Dialled that back to x29, and I'll see if that helps with my normal-workload stability.

Just wrote out a load of stuff urging unclewebb to look into exposing the energy-performance bias (EPB) register, but seems I've been somewhat fooled by Throttlestop's adaptive UI. It does already have the corresponding EPP control for newer CPUs. So, my request would be: expose the EPB control instead if EPP ("SpeedShift"?) is not enabled. Looks like if you poke the MSR, Windows doesn't touch it again until you change power plan. I imagine whatever setting flow you've found works for the EPP MSR in Windows will also work for EPB.

For comparison, it looks like Linux only lets you set it to 0,4,8 or 12. There was a recent patch to let you set more values to the EPP register, for finer-grained tuning, although the older EPB register is still limited to 4 values.

Done a bit of fiddling, and it's possible that in the W3690, it isn't particularly fine-grained. Not convinced I can see a difference from 0-3, and 4 is the point at which it starts getting resistant to single-thread turbo, similar to Balanced's 5. Given that the original Linux EPB code (dating back to Nehalem) only bothered with 4 values, and it was implemented by an Intel guy, it seems likely that only 2 bits of the 4-bit value are significant for that chip. It's possible there's finer EPB control on later chips, but maybe not until EPP is available anyway.

Regardless, exposing the EPB setting in Throttlestop would let you do things like running Balanced power plan (for the balanced SpeedStep/PERF_CTL response), but with low EPB, so the core isn't reluctant to Turbo once the Windows governor has requested 100% via PERF_CTL. I imagine this would already work for people with new chips and EPP - if they show the same turbo reluctance.

After more targetted stress testing, using reduced core count in the BIOS, I've scaled my multipliers back to 29/29/29/30/31/31 (default auto voltage, about 1.28V) and 32/32/33/33/34/34 (1.41V). Lost the +1 multiplier bonuses for single-core, and had to not speed up at 4-core at default voltage.

Another thing to think about - if you're running a limited-core boot to do synthetic soak tests for your limited-core multipliers, your "100% load" tests won't run as hot as an all-core system could get. So I took manual control of the fans to get the temperature back up to the same level as the all-core tests to run them hot, so I don't get caught out by a hot shift to fewer cores in a normal boot.

 

[KMO_2000]

One tip I found for Windows 7 (maybe not needed for Windows 8+) was to mess with powercfg to unlock the "Processor performance core parking core override" setting, and disable it. ... Once disabled, it will fully park some cores, reducing the number of cores that low-thread loads are spread across.

Okay, don't do this. Yes it does what I described, but the downside is significant latency problems. Don't know why I didn't notice sooner - probably running too many synthetic loads

People do generally recommend turning core parking off altogether to solve latency problems - if you turn that override off so it parks more than half the cores, latency problems are severe. (Audio pops, mouse pointer judder, and like 20x the "good" latency measured in LatencyMon when I checked). This is no doubt why the override was added in the first place.

Having backed off on that, I went to dig deeper into core parking - some claim it's there to help balance hyperthreading, by shutting off one thread per physical core when it can. Others go further and claim it's the only part of the Windows scheduler that understands hyperthreading, so the scheduler doesn't work properly at all without it.

I found that's not true - the scheduler understands logical cores fine and tries to distribute work on different physical cores when possible, even if everything is unparked. It's not always obvious, as it can look like all cores are busy in the task manager, but if you look carefully, you'll see that when running 6 tasks on a 12-thread CPU, the usage graphs within each pair are mirror images. Logical core 1 is busy when logical core 0 isn't.

The Windows scheduler is quite lax about keeping stuff on a core long term, but it does try short term. This old article talks a bit about the way it works, and I've got some good diagrams showing exactly how it works in practice. Bit off topic for the thread, but I think it'll interest people who are very interested in being in Turbo mode.

Here is a timeline of a fraction of a second running 5 linx tasks, captured on Windows 7 using Windows Process Recorder:

As we start, the system is lovely and stable, with the 5 linx tasks each on 5 separate physical cores. Minor background tasks are tending to favour the free physical core (6/7).

Then there's a burst of activity at 19.495s. For some reason the second task jumps from CPU 2 to CPU 4 - there are now 2 tasks on one core (CPUs 4/5).

But a short while later (19.525s), the system has some sort of periodic audit and realises it's settled in a bad situation and it tidies up, shoving the third task from CPU 5 to CPU 7 to get back to 1 task per physical core.

The rest of the trace is nice and balanced with background work mainly using the now-idle second core. Somewhat later there's another slight shuffle at 19.60s, but that's just a thread moving within one core, so uninteresting.

Now, if cores had been parked, there might be less scope for random shuffling, and we might have avoided that 30ms of double-loading a core, BUT there would also be fewer free threads to run background tasks on, so there could have been more interference with the linx tasks anyway. My guess is that core parking may have been necessary for good HT performance in older scheduler versions, but it's no longer necessary, and is not active by default in Windows 10.

Now, we can look at WHY the task jumped. It's actually somewhat rational, from a scheduling point of view, and given the way the scheduler works (which appears consistent with that 1997 article).

What happened at 19.49275s is someone (Steam!) decided to fire off 7 parallel tasks of work in red. And on top of that, a couple of other apps decide this is a good time to do some processing in green, light blue and yellow. This might be coincidence, or it might be several applications all using a low resolution clock and all deciding to do their work at the same time ("on the hour" sort of thing).

At first we're fine. If processors are idle, then new ready tasks are assigned to them. The 7 Steam tasks start the 7 idle logical CPUs and run on them. But then there's more work. When the light blue task is ready, we are out of CPUs. That light blue task last ran on CPU 2, and as there are no CPUs free and it is deemed more important than the linx task currently on CPU 2, the linx task is booted off.

When the red Steam tasks finish, other tasks take over on the CPUs. By chance, very soon after being booted off CPU 2, the linx task is chosen to take over CPU 4. (If that red task had finished a tiny bit earlier, the light blue would have gone there in the first place). So there are now 2 linx tasks sharing a core, but at this point there are 10 tasks active, so someone has to share. And it stays there, while the light blue then green task use the second core it was on.

When the system calms down again, so there are 6 or fewer active tasks, at that point the scheduler takes time to recheck distribution and rebalance the tasks still running on the same physical core, which we saw in the previous diagram.

I found those diagrams interesting - you can see at a micro level why there's so much noise in the task manager. Whenever there are more than 12 tasks running at once, things all shuffle. And that is quite common. Sometimes a physical core will be double-loaded, but it doesn't persist long.

In all my traces, it was Steam, just sitting there on the task bar, causing the most disruption, periodically spawning those 7 simultaneous threads, when it probably could have run 1 thread. (It's more multithreaded than most games!). I'm sure it could have done whatever it was doing 1 at a time. It's those sort of tasks that the core parking basically tames, stopping them from immediately running on all cores until the load is more sustained.

And that is why hyperthreading sometimes wins and sometimes loses with light loads - depends whether the benefit of having more free threads outweighs the losses from brief intervals where the system is incorrectly balanced.

The scheduler could be more aggressive at rebalancing, I'm sure, but there would be another cost to that - it's yet another move, which has a cache cost, and the move might turn out to be unnecessary. Maybe one of the tasks on the double-load core was about to finish anyway.

TLDR - the Windows scheduler isn't totally dumb. It may look like chaos in the Task Manager, but it's putting some effort in. And if you see yourself not getting close to your target 1-core or 2-core average multiplier, it's because there really are lots of silly little tasks that keep wanting to run in the background, briefly waking other 3rd, 4th or more cores.

 

[KLiKzg]

Cooler Master Hyper T20 have a Clip Mount System. If the CM CPU Holder Base doesn't fit you will have to search for another one on internet. On Aliexpress you can get it for 1-2 USD.

2.28US $ 7% OFF|Desktop All Models 3in1 Cpu Cooler Fan Bracket Heatsink Holder Base For Lga775 1150 1156 1155 775 1366 Or 2011 Am2 Am3 Am4 1700 - Pc Components Cooling & Tools - AliExpress

Smarter Shopping, Better Living! Aliexpress.com

www.aliexpress.com

Regarding 2nd CPU Cooler, If this is your 2nd CPU Cooler, it should be fine. View attachment 166612
It is a cooper Heatpipe cooler with Fan attached. Most of users of T5500 reports that it works fine. T5500 is non overclockable , so the Copper Heatpipe Cooler with 80-90 mm fan should be adequate , even with 24 X 7 Use. Issues like cleaning and maintenance apply on any system - pre-built or custom made.

Most of the T5500 users report issue only with this cooler.View attachment 166613 . Even if you replace it with this cooler View attachment 166614 and attach a 80-90 mm fan to it, it should be adequate for a Non-Overclockable System. Because most of T3500 users with W3680 find this cooler (with a 80-90 mm fan attached) adequate even for Overclocking.

Just put yesterday Cooler Master Hyper T20 in my system, though not T5500, but on Intel DP35DP board running X3360 on 100% load in BOINC...achieved only 75°C.

The DELL version of cooler from T7500 View attachment 166614 is on it's way to me...so going to put "original equipment" into my workstation. Costed me around $20.

Now I'm running 2x 80mm fans on exit of T5500, so it's quite the draft there! Will see how the temp. will come down in T5500, as they run now on 85°C (limit what Tthrottle was set).

 

[KMO_2000]

Juggling voltage + multipliers during sleep on Nehalem/Westmere

I'm having sleep problems, but it's because I'm trying to get clever and modify voltage while running.

I've been running with the BIOS set to auto core voltage, and ThrottleStop has my 29-31 multipliers for that in its default profile. That's plenty fast enough most of the time - don't have to always run hot. Only the Vtt is raised by default for the 3.2GHz Uncore and DDR1600.

I can use Asus TurboV while running to shift up to 1.4V, and then select my high voltage 32-34 multiplier profile in ThrottleStop. (Taking care not to click "Save", cos I don't want that reselected next boot).

This works fine, unless I sleep. The problem is that after sleep the system initially comes up at auto voltage. The TurboV app does restore the higher voltage, but apparently not fast enough. The system often crashes when waking, presumably because it's briefly on high multipliers and auto voltage.

If ThrottleStop could lower the multipliers when sleeping and/or delay re-raising them a bit, that would work. (I haven't checked whether the sleep restores previous multipliers or BIOS defaults, so maybe it just needs a delay).

Is there any way to achieve this? I can see some discussion of some experimental option here, but can't see it in current version 9.2.

Obviously I can switch Throttlestop profile myself before sleeping, but I have to remember, and no good if it's an automatic timed sleep.

Or are there any other utilities out there I could try instead of Asus TurboV to play with the PMBus voltage controller? (ASP0800/ Asus EPU 2). HWiNFO 64 can read its registers but can anything poke it? If I had something with a command-line I could make sure that ThrottleStop ran it before selecting the higher multipliers (preventing instant death if I accidentally click on that profile at low voltage), and maybe that would cover the sleep case.

At least I did learn one useful thing - I was annoyed that TurboV has no "auto voltage" option, so I couldn't properly revert raising to 1.4V. But if I quit the TurboV app then sleep/resume, then that l end up back to the BIOS "auto" default.

 

[lexluthermiester]

This works fine, unless I sleep.

Turn off sleep. I never use it because generally it causes more problems than it is meant to solve, like yours for example.

 

[unclewebb]

@KMO_2000 - ThrottleStop cannot control the CPU voltage on Nehalem or Westmere processors. That experimental feature you found is only for 4th Gen and newer processors that have an integrated voltage regulator.

ThrottleStop does have a few hidden features including a time delay profile switching feature. If your computer resets the turbo multipliers to their default values during sleep, it might be possible to use some features that ThrottleStop already has to solve your problem. I just added a couple of lines of code to ThrottleStop to prevent the turbo multipliers from being immediately applied after resuming from sleep. This should give TurboV some time to increase the voltage first. The amount of time before the turbo ratios are applied can be adjusted by setting an option in the ThrottleStop.INI configuration file.

I have a soft spot in my head for the W3680 / W3690. Amazing processors for their day. With extra voltage they can run fully loaded at almost 4.7 GHz just by using the 35 multiplier.

Give me a few hours and I will send you a message with a download link and instructions on how to set this new feature up.

 

[lexluthermiester]

I have a soft spot in my head for the W3680

I'm running them in several systems. That CPU is the cream of the crop for the systems they're in. Thanks to TS, OCing them is a breeze. Can't thank you enough!

 

[KMO_2000]

Give me a few hours and I will send you a message with a download link and instructions on how to set this new feature up.

You're a gent. Seems to work nicely. (Sent more technical feedback via message).

Here's my Cinebench R20 results:

First using auto core voltage at x29/29/29/30/31/31 (I estimate the chip's max core auto VID as 1.275V; Vtt is manually raised to 1.275V for the DDR1600).



And voltage pumped up to 1.4125V for x32/32/33/33/34/34:



Power limit is set to 150W; all-core runs held x29.0 (3.87GHz) and x32.0 (4.27GHz), and single-core runs averaged around x30.6 (4.1GHz) and x33.6 (4.5GHz). All-core temps maxed at 66C and 75C as shown (with a quiet fan profile, so that would have been at ~50% and ~80% CPU fan).

Don't really want to push my only W3690 any higher than that. Not that desperate for speed, and at that voltage LinX or prime95 get super hot. It's good air cooling, but not a miracle worker, and my GF gets annoyed if I blow too hard

I have raised the high-voltage Vtt a bit recently for stability - I will have to experiment to see if I can now dial back the core a little.

Quite happy with this set-up. I'm using a modified High-Performance profile with the minimum processor state at 0% so it can spend a lot of its time chilling down at 0.96V. But being a High-Performance profile it will jump straight up to 100% processor state as soon as there's load, and the chip will go straight to full turbo.

Update:

Just figured out one of my other instability problems - I thought I'd set the BIOS to x29 multiplier (so 29/29/29/29/29/29 turbo), then ThrottleStop could take over with 29/29/29/30/31/31.

Made the whole thing unstable.

Reason is simple - in the Asus P6T BIOS, if you set x29, then "Auto" voltage sets a fixed 1.225V, rather than tracking the VID output. Presumably they think they're increasing it for an overclock, but this chip has a high VID, so that actually lowers it. The chip requests 1.275V at x29 or higher.

So it's vitally important - at least for me - to set "Auto" multiplier in the BIOS.

 

[unclewebb]

Don't really want to push my only W3690 any higher than that.

No worries. The W3680 is down to $40 bucks on EBay. An unlocked multiplier and lots of guaranteed fun at that price. Too bad there is not a TurboV or similar program available to increase the CPU voltage when using these unlocked CPUs on a Dell or HP motherboard. Some more voltage is usually good for another 500 MHz, 600 MHz or 700 MHz, depending on how crazy you want to go.

 

[DanielLong]

Hello guys! I would have some tips for the developers of ThrottleStop but I don't know how to reach them.
Basically I have 3 thoughts:

1. Please remove all the functions from the main GUI which are obsolete, because I think it just makes the tool user-unfriendly and less intuitive. Half of the options cannot be used anymore, like clock modulation, speedstep, etc. Instead of them, I would add here the power limit of the CPU, because in my opinion that is a much more important setting.
2. Please somehow make it more visible that the Speed Shift EPP value can be set differently for each profile. It was not obvious for me.
3. Please make the CPU power limit configurable for each profile. In my opinion this would be the best option, because for example 15W is far enough for me on battery mode, but running on AC I would prefer 30-35W.

Or please point me to the topic where I could get into contact with the developers!

 

[Retrorockit]

Hello guys! I would have some tips for the developers of ThrottleStop but I don't know how to reach them.
Basically I have 3 thoughts:

1. Please remove all the functions from the main GUI which are obsolete, because I think it just makes the tool user-unfriendly and less intuitive. Half of the options cannot be used anymore, like clock modulation, speedstep, etc. Instead of them, I would add here the power limit of the CPU, because in my opinion that is a much more important setting.
2. Please somehow make it more visible that the Speed Shift EPP value can be set differently for each profile. It was not obvious for me.
3. Please make the CPU power limit configurable for each profile. In my opinion this would be the best option, because for example 15W is far enough for me on battery mode, but running on AC I would prefer 30-35W.

Or please point me to the topic where I could get into contact with the developers!

Uncleweb is the developer of Throttlestop. But Notebook Review has a large Throttlestop forum for notebook users. This thread is a spinoff for overclocking desktop PCs.

TechnologyGuide

Thank you for visiting the TechnologyGuide network. Unfortunately, these forums are no longer active. We extend a heartfelt thank you to the entire community for their steadfast support—it is really you, our readers, that drove

forum.notebookreview.com

I actually use the obsolete settings to overclock old Dell BTX computers but I use version 6.00 to do this so I don't have to bother with all the newer settings.
I think Uncleweb does a pretty good job of providing something for everyone ( even a retro grouch like me).

 

[unclewebb]

some tips for the developers of ThrottleStop

There is only one ThrottleStop developer. Just me. I would love to redesign the ThrottleStop user interface but user interface design is obviously not my strength. None of the features like FIVR voltage control and turbo power limits existed when I first started working on ThrottleStop. One tool that supports everything from Core 2 Duo to 11th Gen Core i has resulted in the user interface becoming a bit of a mess. If ThrottleStop was a commercial app, I could hire a team to fix it up. With zero revenue, any team I hire is not going to last very long.

remove all the functions from the main GUI which are obsolete

Some 8th and 9th Gen computers do not use Speed Shift Technology. They still use SpeedStep and the Set Multiplier function to control the CPU speed. Clock Modulation, Disable Turbo, BD PROCHOT and C1E are used on every Intel Core i CPU released since 2008. If I start removing or hiding the features that are not important to you, I will immediately have a pile of complaints from users that still use these features.

ThrottleStop is a utility program. Check a few boxes and minimize it to the system tray like everyone else does.

Being able to set different power limits for each profile is a great idea. If I have time, I will add that feature to ThrottleStop this winter.

where I could get into contact with the developers

I am not too hard to find. Have a look in your ThrottleStop folder. My name and email address should be in the ReadMe file.