Throttlestop overclocking Desktop PCs

[dhrag5t]

PS: After PCIE NVME Drive if you also add a PCIE USB3 adapter, Your Dell Precision T3500 becomes at par with any modern system.

One last question ... what would the difference in speed be between NVME on a PCI express x4 2.0 compared with an SSD running off SATA 2? Still on the fence on if I should go with the NVME setup or just get an SSD for my T3500.

 

[unclewebb]

is it possible to turn off speedstep/power savings

ThrottleStop has an option so you can toggle SpeedStep on or off but there is no reason to disable SpeedStep. If you want a steady multiplier, leave SpeedStep enabled and go into your Windows Power Options and make sure the Minimum and Maximum processor state are both set to 100%. If ThrottleStop shows your multiplier dropping when lightly loaded, you will need to go into the bios and disable the C states; C1E, C3, C6.

ThrottleStop does not modify any registry keys. SpeedStep can be toggled by changing MSR 0x1A0 - bit[16]. I think bit[20] in that register is the lock bit. If the lock bit is set, you cannot change bit[16]. No need to mess around with this register. Just use ThrottleStop instead.

Here is an example of a steady multiplier with the CPU lightly loaded. No complaints.

 

[ManGupta]

One last question ... what would the difference in speed be between NVME on a PCI express x4 2.0 compared with an SSD running off SATA 2? Still on the fence on if I should go with the NVME setup or just get an SSD for my T3500.

Even SATA SSD would be a great improvement over Mechanical HDD.

But NVME SSD is around 5 times faster than SATA SSD but may cost about 50-60% to 100% more.

 

[bebop350]

ThrottleStop has an option so you can toggle SpeedStep on or off but there is no reason to disable SpeedStep. If you want a steady multiplier, leave SpeedStep enabled and go into your Windows Power Options and make sure the Minimum and Maximum processor state are both set to 100%. If ThrottleStop shows your multiplier dropping when lightly loaded, you will need to go into the bios and disable the C states; C1E, C3, C6.

ThrottleStop does not modify any registry keys. SpeedStep can be toggled by changing MSR 0x1A0 - bit[16]. I think bit[20] in that register is the lock bit. If the lock bit is set, you cannot change bit[16]. No need to mess around with this register. Just use ThrottleStop instead.

Here is an example of a steady multiplier with the CPU lightly loaded. No complaints.

Thank you for the detailed explanation, I really appreciate it. Sorry I should have mentioned again that I'm on Linux and cpufreq doesn't seem to work with my board, which is why I'm curious about MSRs. But I appreciate the refresher on Windows power options as well, I haven't used it in awhile.

Writing to certain bits in a register still confuses me, to be honest. I'd think it would be "wrmsr 0x1a0 0x0", but I see here it's suggested to use "wrmsr 0x1a0 0x4000850089"; indeed "rdmsr 0x1a0" outputs "850089". I'll give it a go on a flashdrive OS (in case I break something), this weekend hopefully. Thanks again!

Wow 5GHz , very nice.

 

[unclewebb]

@bebop350 - The proper way to change an MSR is to do a read - modify - write. First you need to read the MSR to find out what is in there. Each bit in this register means something different. Doing a wrmsr 0x1A0 0x0 would clear everything in this MSR. You do not want to do that. You just want to modify (clear) bit[16] to disable SpeedStep.

If the initial value is 0x850089, that means SpeedStep is enabled. To disable SpeedStep, you need to wrmsr 0x1A0, 0x840089

That should work in Linux or Windows. Both operating systems use the same CPU.

Wow 5GHz

It was just an unstable suicide run but it makes for a nice screenshot.
My first ever trip to 5 GHz. I love this old technology!

 

[bebop350]

@bebop350 - The proper way to change an MSR is to do a read - modify - write. First you need to read the MSR to find out what is in there. Each bit in this register means something different. Doing a wrmsr 0x1A0 0x0 would clear everything in this MSR. You do not want to do that. You just want to modify (clear) bit[16] to disable SpeedStep.

If the initial value is 0x850089, that means SpeedStep is enabled. To disable SpeedStep, you need to wrmsr 0x1A0, 0x840089

That should work in Linux or Windows. Both operating systems use the same CPU.


It was just an unstable suicide run but it makes for a nice screenshot.
My first ever trip to 5 GHz. I love this old technology!

Thank you for taking the time to explain that. I gave it a shot but unfortunately the CPU still downclocks. Maybe the lock bit is set? Although I verified that it changed MSR 0x1A0 properly. BIOS is set to C1 only, not sure if I have to disable that, will have to try that next.

 

[dhrag5t]

@bebop350 - The proper way to change an MSR is to do a read - modify - write. First you need to read the MSR to find out what is in there. Each bit in this register means something different. Doing a wrmsr 0x1A0 0x0 would clear everything in this MSR. You do not want to do that. You just want to modify (clear) bit[16] to disable SpeedStep.

If the initial value is 0x850089, that means SpeedStep is enabled. To disable SpeedStep, you need to wrmsr 0x1A0, 0x840089

That should work in Linux or Windows. Both operating systems use the same CPU.


It was just an unstable suicide run but it makes for a nice screenshot.
My first ever trip to 5 GHz. I love this old technology!

5GHz is absolutely incredible for x58, even if it was an unstable run. I'm stunned the PC didn't immediately blue screen. On my W3690 I tried pushing the multiplier to 33x (basically 4.4 Ghz) and it crashed as soon as I opened up Chrome.

 

[unclewebb]

unfortunately the CPU still downclocks

Disabling SpeedStep is the cure all for everything but it rarely needs to be done.

There is a lot of inaccurate monitoring software on the Windows side. Not sure about Linux. I know there is some Linux software that reports C state residency time. This is the most likely problem.

On my W3690 I tried pushing the multiplier to 33x

The 5 GHz screenshot was a little off topic. That was obtained on a board that allows adjustable CPU voltage. Give these CPUs lots and lots of voltage and the sky is the limit. That is why good used X58 boards still sell for ridiculous prices on EBay.

 

[bebop350]

Disabling SpeedStep is the cure all for everything but it rarely needs to be done.

There is a lot of inaccurate monitoring software on the Windows side. Not sure about Linux. I know there is some Linux software that reports C state residency time. This is the most likely problem.

Software monitoring seems to be at least as bad on Linux. So, it sounds like you're indicating it's a software or C state issue if I understand you correctly. I was mistaken, no C state control in the BIOS after all, the only power option is ACPI Suspend Type S1 or S3. Oh well, it was worth a shot, I appreciate the help.

 

[Retrorockit]

I have a user at OCN who wants to run Windows 2000 Pro. He says that OS only supports 2 core CPUs. The best CPU for a system like he wants is an X6800.
I was wondering if Throttlestop can work on an OS that old? TS4.00 maybe?

 

[Arctucas]

Can anyone explain this to me?

 

[unclewebb]

@Arctucas - PL1 is the long term turbo power limit. PL2 is the short term limit. Not sure why but when PL1 or PL2 is the cause of throttling, EDP OTHER under the RING section usually lights up at the exact same time.

Check your bios or click on the TPL button in ThrottleStop and see what your turbo power limits are set to. At default settings, some PL1 throttling is not unusual. Usually when overclocking, most motherboards increase the power limits sky high so they do not interfere with maximum performance. I think Asus usually sets PL1 and PL2 to 4095.

A yellow box in ThrottleStop just shows that some throttling took place. This might have happened during boot up, before the bios set the power limits. Yellow boxes are not a problem. A red box indicates that throttling is in progress. That is a problem. Open up Limit Reasons when testing and see if anything is red. Put a load on your CPU and post some screenshots. You can click on the CORE, GPU and RING headings to clear this throttling information out of the CPU. Good to do that before you start testing.

@Retrorockit - I have not run any version of ThrottleStop on any version of Windows 2000. I doubt ThrottleStop 8.70.6 will start up when using Windows 2000 but an older version of ThrottleStop like 6.00 or 4.00 might work OK. As long as it starts, it should work correctly.

 

[Arctucas]

@unclewebb,

Apparently, my VCore and VCCSA were a bit low.

Thanks

 

[Nisargadatta]

Hey guys,

When I'm stress testing my w3680 on a HP Z400 with throttlestop, C0% will be at 100% but sometimes throttles down and seems to affect my score. Is C0% supposed to throttle in this manner?

 

[unclewebb]

@Nisargadatta - The important number to watch in ThrottleStop is the multiplier (FID column). Increase the turbo power limits TDP / TDC so when fully loaded, the CPU is not dropping the multiplier.

The C0 number represents what percentage of time the CPU is in the C0 state working on a task. If you are using the built in TS Bench test, it is normal for the C0% to drop slightly just before finishing. If you run an 8 or 12 thread test, the TS Bench will initially divide the calculation up into 8 or 12 equally sized jobs. As individual cores or threads finish their task, the C0% will drop as that thread is no longer active. It would be unusual for all threads to finish at the exact same time so the drop in C0% near the end of this test is normal. C0% should not be dropping in the middle of a test.

What sort of multiplier are you running? My W3680 was 100% stable at 30X but not quite 100% stable at 31X. The voltage is limited on these motherboards so most of these CPUs will not run reliably any faster than that.

 

[Nisargadatta]

Thanks for the quick reply, good to know a little more info on how the ts bench works.

Same 30x multiplier 100% stable, actually bought the w3680 thanks to a post you replied to over on Linus' forums.

 

[Aaron_Henderson]

I wish I was more capable with electronics...I wonder if for a lot of these boards, there could be a volt mod done to increase vcore a bit.

 

[lexluthermiester]

1. Triple channel RAM and its bandwidth has little to nothing to do with things.

Benchmarks would disagree with you. I've done testing on X58 that shows a 43% increase from dual channel to triple channel. This translates into a very important performance increase for many applications.

2. That CPU cannot remotely keep up with modern CPU's the same core/thread count on any front.

Is that what you meant? If so, it depends on what you're doing.

3. While the clock speeds are nice, those CPUs are like 50% behind in IPC.

Again that depends on what being compared. If you are comparing a $60 X5680 to a $600 i7-7800x Then yeah, the IPC difference is going to be there, but at what cost? These S1366 based systems offer excellent vlaue in performance/cost ratio. And 1366 CPU's are very overclockable so there is additional value to consider.

 

[EarthDog]

Benchmarks would disagree with you. I've done testing on X58 that shows a 43% increase from dual channel to triple channel. This translates into a very important performance increase for many applications.

Is that what you meant? If so, it depends on what you're doing.

Again that depends on what being compared. If you are comparing a $60 X5680 to a $600 i7-7800x Then yeah, the IPC difference is going to be there, but at what cost? These S1366 based systems offer excellent vlaue in performance/cost ratio. And 1366 CPU's are very overclockable so there is additional value to consider.

1. In what? You know better than to come in here and drop a figure and not support it. Surely something responds well, but not a lot at all.. Also, context, see what I responded to in order to frame my response.
2. Of course it does... but read what I responded to nearly 1 month or so ago...
3. ???? Between an i7 920 ($300+ new) vs a 10900x at $600, the ipc difference is night and day, at least 50% ipc improvement..not to mention the clockspeed differences.) Which makes it a ton faster.

The whole point is no matter how many cores and threads you throw at that old platform or how much bandwidth it has, it's still MUCH slower than modern counterparts due to the clockspeed difference and IPC. Triple/quad channel ram Cleary isnt making it up. Remember, scope...context... who I replied to.

Heres a good read - https://www.pcworld.com/article/2982965/quad-channel-ram-vs-dual-channel-ram-the-shocking-truth-about-their-performance.html

I’m sure that somewhere out there beneath the pale moonlight, there’s a task or benchmark that truly pays the dividends you’d expect by doubling the available system bandwidth, but I’m not seeing it here.

 

[Retrorockit]

I wish I was more capable with electronics...I wonder if for a lot of these boards, there could be a volt mod done to increase vcore a bit.

There is a Volt mod possible. It's the same as the one for GPUs. I haven't done it myself but you find the V. Sense pin on the VRM controller chip and you install a variable pull down resistor to GRD from that pin so the VRM "thinks" the Voltage is lower than it actually is. You also need to install a digital Voltage display to measure ( perhaps on one of the VRM chokes) the actual Voltage achieved because the VRM will adjust the Voltage so the MB will "think" it's still at the original setting. There is an art to selecting the correct value for the resistor. I think >20x,<50x the resistance to GRD on that pin keeps it from pulling the idle voltage up too high but gives enough sensitivity to the adjustment to be practical. On LGA775 the Vcore table for the CPU ends at 1.600V. and any Voltage detected over that shuts down the VRM. Some aftermarket boards allow higher Voltages, but not per Intel specs. This mod not only raises the Voltage but spoofs the sensor too. I never looked into it for LGA1366.

 

[Retrorockit]

I came across this Dell T3610 overclock with an unlocked Xeon E5 1650 v2 at 4.2 Ghz with a GTX 1080.

Dell Precision T3610 Performance Results - UserBenchmark

So no need to troll this thread for being X58 groupies.
You could probably save some money on the RDIMM 4 channel DDR3 RAM. The PSU is proprietary on these but 685W was an option which seems adequate. Even the single CPU versions of the Dell T36xx workstations use the multiple CPU chipsets and workstation RDIMM memory.
I didn't shop around, just looked at one of the usual suspects.

Dell Precision T3610 CAD PC 4-Core 3.70GHz E5-1620 v2 32GB No HDD 55100003819 | eBay

Our technicians specialize in Dell models ensuring that your unit will arrive on your doorstep with quality components and performing at optimal efficiency. Model: Precision T3610. Installed Ram 32GB (8x 4GB).

www.ebay.com

That CPU is 4C/8T. Appears to be an i7-4820K analog.

http://www.cpu-world.com/news_2013/2013091401_Intel_Xeon_E5-1600_v2_and_E5-2600_v2_CPUs_launched.html?PROCESS=Read+more

...
In researching this system I'm finding that it doesn't require RDIMM, or even ECC RAM. This saves some latency on both counts and allows low latency gaming RAM DDR3 1866 UDIMM to be run. This may limit you to 4x8GB Modules But 4 channel DDR3 1866 is about as good as it gets.
Here is some 8GB DDR3 1866 CAS7 RAM

NEMIX RAM SILVERLINE 16GB (2x8GB) DDR3 1866 (PC3-14900) PC GAMING MEMORY - Newegg.com

Buy NEMIX RAM SILVERLINE 16GB (2x8GB) DDR3 1866 (PC3-14900) PC GAMING MEMORY with fast shipping and top-rated customer service.Once you know, you Newegg!

www.newegg.com