[SOLVED] Speed Shift / EIST / power plan questions

m.bsatgini

Honorable
Sep 21, 2017
35
0
10,530
Hi,
I run my OC with fixed vcore, LLC Turbo on my 9900k
My power plan is "normal"
Right now Speed Shift / EIST are enabled
Is there any loss of stability / performance if those are enabled?
Should I enable all of them ? One of them? Neither?
I would like to know what is the "ration" beetween performance and cooling by using or not those different setttings and also If I should use max performance on windows power plan.
Thanks
 
Solution
Small FFT or Small data set, no AVX and no AVX2, for 15 minutes. That establishes baseline thermal compliance. Then, if you know or suspect that you will be running games or applications that are specifically AVX or AVX2 (Or especially if you are for some reason running AVX512 instructions, which are, whew, let's put it that way.) then you will want to do additional tweaking by enabling those, running it again for 15 minutes and making adjustments to the AVX offset in the BIOS in order to keep thermals within the acceptable envelope. That means, MAXIMUM, 85°C and a maximum of 80°C is much preferred.

Much also might depend on what the current, average or maximum ambient temperature is, or will be during the hot season where you are, in...
I like to disable Speed shift but enable speed step (EIST). Speed shift, in my experience, has tended to either not work or be hit and miss anyhow. Speed step works fine and for an overclocked fixed vcore configuration, especially if you also have the Windows power plan set to something like Performance with a minimum processor power management setting of 8% and maximum of 100%, so that cores can flex and relax as needed. This can help with core temperatures tremendously.

What is your CPU OC set at? You might not need Turbo LLC, as that's very high and generally only necessary at the absolute highest levels of overclocking. I'd usually try with a 5 or 6 on a 1-10 scale LLC or simply High or Standard on a textually based LLC. In fact, I'd probably normally rather raise the vcore a bit in order to keep the LLC from needing to be set to turbo because an LLC of turbo is probably likely to substantially lower your maximum sustainable OC thermal compliance.

I'd put the power plan to performance, then go into the advanced settings and change the minimum to 8%. Leave EIST enabled and disable Speed shift.

Then play a bit with the vcore and LLC to see if you can get away from Turbo while still remaining stable during testing.

I have full thermal and stability testing guidelines here. Note, they are two different things and thermal testing should always come first, but stability testing is just as important and you should not cut corners no matter that it's a "gaming" system and not a "scientific machine" like some people claim. All machines, regardless of type, can suffer from data corruption the same way and there does not have to be any appreciable BSOD issues for that to be happening.

 

m.bsatgini

Honorable
Sep 21, 2017
35
0
10,530
I have my 9900K set at LLC turbo for 1.24, it does not overshoot, 1.248 on load.

I tried to lower LLC without success but haven't got time to test it more, i tried to go for high at 1.29 but it wasn't stable so I left it on turbo because i found it frustrating to go higher than 1.3 but do you think I should ? What vcore should I try?

i went back to normal performance now so 5% cpu minimum which sound ok too I guess.

For testing, i've just ran an OCCT large avx for 2h, temps maxed at 77c, vcore 1.248, vr out 1.232 with turbo llc so I guess it's not bad at all no?

Which vcore should I aim to lower llc? 1.32 ?
 
Last edited:
Sounds like you've got it dialed in well enough. I'd still make the other changes to EIST and Speed shift. You can also test with Speed shift on and off with EIST enabled AND on and off with EIST disabled. It may work fine on your board. It has never worked right for the most part on the majority of boards and CPUs I've attempted to use it on. Aside from one or maybe two occasions, it has refused to allow a variable fluctuating CPU frequency.

5% is fine, nothing wrong with it, I like 8% because it's what many of Windows power plans have used for years and there is a very nominal decrease in the amount of time it takes to react to full speed which can make things feel a bit snappier if you are sensitive to that sort of thing. Many overclockers in fact that use the power saving modes do in fact set the min to 10%. 8% also seems a happy medium. 5% is fine.

Do you have an AVX offset configured in the BIOS, because if you do then testing with AVX or AVX2 instructions should be a later detail, for fine tuning. OCCT is great, has it's uses, but I'd highly recommend using Prime95 Small FFT for thermal testing with AVX and AVX2 disabled, at least initially, and then later you can enable AVX and AVX2 and retest while tweaking the offset in the BIOS if desired. If you don't KNOW you run AVX specific games or applications then it's probably not terribly important to test using AVX instructions. You are likely to seriously limit your overhead that way and while that can be a useful tweaking tool it's also a good way to see unrealistic temperatures and instability for many people. If you do use AVX or AVX2 programs or games, then it's definitely an essential part of your process but it shouldn't be part of the baseline in my opinion.

Honestly, if you don't have the time to do these types of things then overclocking is often a bad idea because attempts at high overclock goals using quick methods generally results in problems or complications, if not outright catastrophe in some cases.

I agree, you are on the right track and what you've done looks good so far, but it's hard for me to say because until I see baseline validation which I tend to believe is gained by 15 minutes of thermal testing using Small FFT and 8 hours of Realbench. That's my baseline.

For the vcore, the Intel specification lists 1.5v, which we know is grossly high and could never be maintained without sub-zero cooling regardless of multiplier and would likely lead quickly to electromigration and VT shift, as per the Intel temperature guide found here:

In truth, an absolute maximum of 1.45v should be considered and only in pretty extreme circumstances for very high or temporary overclocking conditions. In reality, for a daily driver, in my opinion and on this architecture, anything up to and including 1.35v is probably safe assuming you have cooling capable of keeping it thermally compliant. I'd like to see things stay much closer to 1.3v if possible but obviously there are various ways of doing this including setting adaptive, offset and fixed voltages and each will require a fundamentally different approach and voltage. So if you NEED to go higher than 1.29v and you can keep thermals in check with your cooling, and have a board that is capable enough, and most especially, if you are not running four DIMMs, then it does at least offer you some room for experimentation.

But it sounds like you might already be where you need to be, and perhaps just need a bit more testing to validate the stability of the overclock.
 

m.bsatgini

Honorable
Sep 21, 2017
35
0
10,530
Thanks !

For prime and occt small with avx honestly I don't see the point. With large my temps are fine, with small without avx I hit 90c so with avx I know i won't pass it but for my usage my CPU will never be as much stressed as with that kind of test.

Also, I have coil whine when c state C1E is activated but is it worth it to keep the other c state enabled?
 
Large FFT is absolutely not a useful metric for thermal testing. It is not Steady-State, and any test that is not steady state is not recommended for thermal testing. How can you possibly determine if you are thermally compliant using a data set that allows non-standard behavior? Prime95's default test, Blend, is a cyclic workload for testing memory stability, and Large FFT's combines CPU and memory tests. As such, Blend and Large FFT's both have cyclic workloads which are unsuitable for CPU thermal testing.

Thermal testing should be done with either Prime95 Small FFT with AVX and AVX2 disabled, which you can do on the main page options that pop up when you open the program. When you disable AVX2, it will un-gray the option to disable AVX as well. You can also use the OCCT small data set, which is primarily the same data set and test as Prime95 small FFT, but is about 3% "lighter" according to my colleague Computronix who authors the Intel temperature guide and has done extensive testing with practically every type of software out there.

"Gaming" workloads are "fluctuating" workloads that vary greatly depending on how a particular game title allocates CPU / GPU workloads. This makes gaming a poor metric for measuring thermal performance because there's no "standard". As such, gaming workloads don't conform in any way to Intel's datasheets. We use Prime95 Small FFT's without AVX because it's a "steady" 100% TDP workload that does conform to the datasheets, which provide the standards we follow to perform a valid thermal test.

However, if your heaviest workloads are games, and you don't run, or have any intentions of running more demanding workloads such as rendering or transcoding which can approach or equal the workload of Prime Small FFT's without AVX, then although your thermal performance doesn't conform to the datasheets, it's certainly adequate for gaming. Also, keep in mind that 91°C hottest Core during Prime95 Small FFT's without AVX at 21°C ambient leaves very little headroom for higher seasonal indoor temperatures.

Moreover, a 240mm AIO is considered marginal for the 9700K as well as the 9900K; the principal difference being Hyper-Threading, which creates higher Core temperatures. We instead recommend a minimum AIO of 280mm, and preferably 360mm for these processors. Since all AIO's do and will eventually fail, keep this in mind when replacing your 240mm AIO down the road.

Darkbreeze suggested that you read the Intel Temperature Guide, so if you really want to get yourself up to speed on this topic, then I would do so: https://forums.tomshardware.com/threads/intel-temperature-guide.1488337/

-Computronix
Author - Intel temperature guide

Relevant threads.


 
Small FFT or Small data set, no AVX and no AVX2, for 15 minutes. That establishes baseline thermal compliance. Then, if you know or suspect that you will be running games or applications that are specifically AVX or AVX2 (Or especially if you are for some reason running AVX512 instructions, which are, whew, let's put it that way.) then you will want to do additional tweaking by enabling those, running it again for 15 minutes and making adjustments to the AVX offset in the BIOS in order to keep thermals within the acceptable envelope. That means, MAXIMUM, 85°C and a maximum of 80°C is much preferred.

Much also might depend on what the current, average or maximum ambient temperature is, or will be during the hot season where you are, in the room where the computer is normally used. Ambient temperature affects thermal levels significantly.

What are your FULL hardware specifications including motherboard model and all cooling components, as well as the arrangement and orientation of the case or CPU cooling components? That can have a substantial effect on cooling as well. Two different configurations and orientations of the same cooling components could have as much as a 20°C difference in temperature. That's assuming a major problem in the way the cooling is configured, but we see that OFTEN, even from users that are supposedly well versed in building systems. So, it happens.
 
Solution