Does C1E, EIST lag?

[Deadtrees]

My Q6600 is overclocked @ 3.2 Ghz(400mhz x8) and I do have C1E turned on. With it, the cpu runs at the speed of 2.4(400mhz x 6) when much hosrepower isn't needed. The voltage remains the same regardless of the clock difference. EIST is turned off because it just doesn't make any difference.

Now, I have two questions that I failed to find answers.

1. Do I save energy by having C1E enabled? I think so because the cpu runs at slower speed although same vCore is applied, but I'm just not sure about it. If I'm not saving any energy by running it slow, I think I should just run it @ 3.2 full time.

2. Does C1E create latency? I used the word 'latency' because I'm not talking about overall speed difference but a 'lag' it might cause.
If I'm running a program that requires the CPU to run @ full speed, I wouldn't care about slight speed difference that might be caused by CPU going from 2.4ghz to 3.2ghz. However, I use programs that require speed in very short amount of time, e.g. When I browse through 22mp RAW images in full screen. In that case, the speed is only needed less than a sec. Because it's a very short amount of time (1 sec,) I think the job is done before the CPU goes into full speed. If so, I think I'm wasting CPU power by not using it its full potential. Am I right on this one?

 

[aigomorla]

When your overclocked and over volted, i dont think C1E / EIST works very effectively in saving you power.

On your Kentsfield, those things are pretty dayam hot... if anything it might make your chip run a bit cooler when not needed, but saving electricity, i think your better off turning your bathroom light off when no one is in it.

As for latency... i cant answer that for ya, as i have never had those things on to begin with.

If my system is clocked at X.XX mhz, there is no way i would want it running slower when heat isnt an issue.

 

[NA1NSXR]

I'm of the opposite opinion. If my CPU is overclocked and overvolted there is no way I would want it to remain at that state if I am idling. Also, I don't think switching takes any perceivable time at all for these features. Even though I'm no expert on C1E or EIST, we are talking about circuitry here, even if the switching time was slow it would still be perceivably fast from the user's standpoint. I would be surprised if these features caused any perceivable system response latency at all.

 

[Rubycon]

Yes it does. Run benchmarks particularly Everest and you will see dramatic differences. Will it affect real world results? I have no idea. However if you are fine tuning you must have faithful output by your benchmarking programs so you want to leave it off until you are satisfied. Then at that point turn it on THEN check for instability one last time to see if stability has been compromised.

 

[Idontcare]

Originally posted by: Deadtrees
When I browse through 22mp RAW images in full screen. In that case, the speed is only needed less than a sec. Because it's a very short amount of time (1 sec,) I think the job is done before the CPU goes into full speed. If so, I think I'm wasting CPU power by not using it its full potential. Am I right on this one?

This speed is rate-limited by your hard-drives and their latency + bandwidth. This is one of those examples where you'd be blown away by an SSD like the X-25M (where its >200MB/s read performance is what you want combined with 0.085ms access time).

As for C1E/EIST yes I use them on my overclocked systems and yes they do save you power/money. I have a kill-a-watt and have characterized my system's power consumption with and without these power savers (Q6600 as well as QX6700) and the difference can be as much as 50-75W.

Yes that is no more than an extra light bulb turned on in your house but it is free to you to save that money (you don't leave the bathroom light on when no one is in the bathroom do you? even though its "just one light bulb" right?) but you also have that extra heat 24/7 heating up your house. If you live in Alaska or Michigan then that might not be a problem but if you live in a warm location then you are probably paying for air conditioning as well, so that 50-75W of extra heat now means you need to also pay for another bit of electricity to move that heat from inside your house to the outside of your house. And all for what?

Recommendation: Keep C1E/EIST enabled and get an Intel SSD or an OCZ Vertex SSD.

 

[NA1NSXR]

Which Everest benchmarks in particular?

I'm asking because I'm in the middle of a big encode on my computer and I cannot restart it now to find out for myself. I'll run them as soon as this is done later today though.

 

[Rubycon]

Cache and memory benchmark.

 

[lopri]

You can also use this tool for a quick latency test. (direct download link from cpuid.org)

C1E/EIST does affect latency, thus performance, but IMO its benefit outweighs the (ever so) slight performance loss. You're likely notice more lags from anti-virus program than C1E/EIST. Intel CPU's power saving features are very robust and the performance loss is minimal. OS level power saving features (particularly S3) works better with C1E on as well. Unless a system is folding or benchmarking, I'd leave C1E/EIST on.

 

[NA1NSXR]

I ran the all of the Everest memory benches and DPC Latency checker and there is no difference on my system, doing 5 runs of each benchmark and averaging the numbers (the results did bounce around slightly even with the computer totally idling). In DPC latency checker there was also no improvement after turning off both C1E and EIST.

I did not try the CPUID latency test I will do that now.

edit - no difference using the cpuid latency test either.

After going through this for myself I will now say there is zero memory and CPU cache performance penalty from C1E and EIST. The Everest read/write/copy benchmarks did not deviate over 100MB/s between runs, the Everest latency benchmark never deviated over 1ms and the cpuid latency test never deviated over 1 cycle. That is about as close to the same as its going to get for these particular tests.

Anyway, maybe others will run through the Everest memory benchmarks and CPUID latency test as well and confirm my results? Do you guys have links to discussions or material that lead you to believe C1E/EIST introduces system latency in the first place? My search came up with a couple but its mostly newbies making claims and no substantial (or consequential) discussion.

edit 2 - I think the reason my scores were the same with the power saving features turned on and off is because the latency checker and memory benchmarks did not trigger a clockspeed transition on my system with the features turned on. EIST is reported to consume up to 10ms for each "step" change on a merom. I don't know if Speedstep latency has improved in newer chips. If you're loading up a computer though its only a one time penalty for getting up to speed and assuming EIST is still as bad as 10ms that is the maximum penalty you are likely to see per computing load.

 

[MagickMan]

I've not tried it, but the scoop is that C1E/EIST could cause instability if your CPU is overclocked, especially one that requires a healthy bump in voltage.

I'd love to be able to use some energy-saving features, but not if my system is going to randomly reboot and f#$# up my OS.

 

[Rubycon]

It's strange your scores are not different. Every system I've had since this was supported has exhibited this behavior from P4 to i7. Are you sure speed step is actually working and changing the multiplier from idle to full load?

 

[aigomorla]

Originally posted by: Idontcare
(you don't leave the bathroom light on when no one is in the bathroom do you? even though its "just one light bulb" right?)

You'd be suprised at how many people do this...

 

[taltamir]

didn't anandtech test those in regards to the lower than expected performance of TLB fixed phenoms? they found that lag DOES exist. I don't remember if it was UP TO or ABOUT 2% on intel C2 implementation and 10% on AMD phenom (the first) implementation (aka, that was the performance loss from turning on power saving). I think it was about.
I use it despite the 2% loss because it is worthwhile from a power cost / performance standpoint.

 

[Idontcare]

Originally posted by: taltamir
didn't anandtech test those in regards to the lower than expected performance of TLB fixed phenoms? they found that lag DOES exist. I don't remember if it was UP TO or ABOUT 2% on intel C2 implementation and 10% on AMD phenom (the first) implementation (aka, that was the performance loss from turning on power saving). I think it was about.
I use it despite the 2% loss because it is worthwhile from a power cost / performance standpoint.

The AMD CnQ bugginess stuff is something else entirely, not to be conflated with the C1E/EIST cpu state switching on Intel chips in an otherwise properly implemented/functioning power-state enabled system.

The issue with some AMD platforms is that the cpu's are capable of properly switching between power-states but not all mobo makers have properly implemented the code to do so in their BIOSes...the result is that some AMD mobos induce these power-state switching bugs and it seems like your cpu is the source of the problem.

I can't remember which AT article delved into this, but it was probably the one that debuted PhII back in Jan if you want to track it down.

 

[taltamir]

what I am talking about is the Phenom ONE versions that end with 50 and have the solved TLB bug... when anandtech tested those they noticed that performance was not all that improved. So they delved deeper and found a lot of performance loss from CnQ in those phenoms.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3344&p=4

This is NOT it, I remember seeing a more detailed article after this one, but I cannot find it at the moment.

 

[Idontcare]

Originally posted by: taltamir
what I am talking about is the Phenom ONE versions that end with 50 and have the solved TLB bug... when anandtech tested those they noticed that performance was not all that improved. So they delved deeper and found a lot of performance loss from CnQ in those phenoms.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3344&p=4

This is NOT it, I remember seeing a more detailed article after this one, but I cannot find it at the moment.

That's exactly what I am talking...but you have to read the follow-up article to get the full picture and the follow-up info is contained in the PhII article where they uncovered the source of the CnQ symptoms which they observed and documented in the initial PhI B3 review.

 

[Idontcare]

Originally posted by: Idontcare

Originally posted by: taltamir
what I am talking about is the Phenom ONE versions that end with 50 and have the solved TLB bug... when anandtech tested those they noticed that performance was not all that improved. So they delved deeper and found a lot of performance loss from CnQ in those phenoms.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3344&p=4

This is NOT it, I remember seeing a more detailed article after this one, but I cannot find it at the moment.

That's exactly what I am talking...but you have to read the follow-up article to get the full picture and the follow-up info is contained in the PhII article where they uncovered the source of the CnQ symptoms which they observed and documented in the initial PhI B3 review.

I was looking up X2 links for another poster and came across the review info I was recollecting when writing this post:

Regardless of how AMD enabled it, motherboard makers were constantly botching it up in their BIOS which would result in different motherboards having very different power consumption levels especially at idle. It appears that some vendors were properly enabling this software-hack C1E state while others weren?t.

AMD always expressed frustration to me that the motherboard vendors kept screwing things up and I?m guessing they got tired of dealing with it. The new Athlon II X2 has microcode level support for the C1E state; when the OS executes a halt instruction, the CPU now knows to both shut off its clock and drop its voltage. No BIOS trickery necessary.

The problem with this, as you can guess, is that not all current motherboards have proper BIOS support for it. Yep.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3572&p=2

 

[Deadtrees]

Thank you all for great reponses!
I decided to just leave it on based on Idontcare's advice.