Upgrade to D0?

[Shmee]

So, I got my C0 up to 3.8 Ghz with HT off. Vcore is around 1.35, QPI around 1.4. 200x19.
It seems that it gets unstable, quite so, with the 20x multi at 3.8 Ghz or above. 200x20 is a no go, even with higher vcore and slightly higher QPI. 190x20 is very unstable at same settings where 200x19 is fine.

I have a lapped TRUE black, push pull. Linpack stays below 60 C as of now on a nice cool night. I think I don't have a very good chip, seems to be rather fussy with voltages. So, I am wondering, would it be worth it to upgrade to a D0 if I can expect MB 4.2-4.6Ghz? Right now its not my cooing holding me back.

I have the UD5 from gigabyte. I read some of the x58 extreme guides from gigabyte that recommended some extreme voltages for C0's over 4 Ghz, like 1.6 QPI and over 1.4-1.5 Vcore. How safe is this? Seems rather out of spec.

I have heard that D0's can be better at much lower volts.

So, If I can get a D0 for $200, and *possibly* sell the C0 later, or not, would it be worth it for 400-800 Mhz? Also, our my guesses for the D0 way off, or realistic?

I am also thinking of flashing to one of the F8 beta bios's, here they help alot from F7, especially with D0's.

So...Stay as is? Oc the C0 more? Or get a D0?
Suggestions/advice appreciated.

 

[JAG87]

Personally I don't think it's worth it. First of all nothing is guaranteed you can be out $50 and end up with the same OC. Second, chances are you won't notice any difference, unless you stare at your encodes or renderings and giggle when they finish 2 seconds faster. Gaming wise, nada.

I'm leaning towards the idea that your board isn't up to snuff. We've got fairly similar setups, I can do 200x21 HT off with 1.36250 Vcore and 1.2 VTT. 1.375 Vcore is as high as you should go if you want to be within warranted specification (means your cpu will be trouble free for at least 3 years). 1.5 Vcore is very high, borderline with absolute max, and 1.6 VTT is way way beyond the absolute max of 1.35V. Very dangerous. If I were you I would back down the clocks as it seems that your VRM is having a hard time supplying the necessary amps. Try 3.6 with HT on if you need 8 threads, or settle for 3.8 HT off.

EDIT

http://www.hardwarecanucks.com...erboard-review-14.html

After reading this, it seems your board has pretty good voltage regulation. There is very little droop, which indicates very stable VRMs. You shoud try to enable LLC, as the voltage fluctuations from load to idle are minimal. With 1.35 Vcore +LLC you give around 1.33V under load, you might be able to get 4.0 HT off. Even 1.40V, returning 1.37V under load would be acceptable, if needed. My suggestion is to back down the VTT, 1.4V actually puts out more and it's getting a tad high IMO. Bring it down to 1.35V or even 1.30V if you can.

 

[alyarb]

i wouldn't fault the UD5, but i also would not resell a CPU that only gave me 85% of my desired overclock. I have a C1 Q95 and feel the same was as you. Ideally, I want the 200MHz that I feel I was robbed of, but the fact is that we still have great overclocks with undesirable specimens. when you can allow yourself this, have a glass of something you can nurse like a cognac and consider that there is a very nonlinear relationship between frequency and energy and its in everyone's best interest that you got the crappy OC you got.

http://files.getdropbox.com/u/594924/i7_pwr.PNG

100 extra watts just to get from 3.8 to 4.0? feel better?

 

[Shmee]

wow. ty for replies! and I find that graph hard to believe! Is it really that much extra draw just from extra voltage needed and increased frequency? i would think a D0 wouldn't be like that, as they generally need less vcore.

Anyways, I guess I will just keep it as is, 3.8 with HT off. Don't need 8 threads, I have heard it is actually worse in most games.

 

[Grnkjr0]

To Shmee

Are you sure it is your CPU chip that is the cause to your "low" overclock?

I have a C0 and it can do 205x20 with turbo on = 4.3 GHz. I think my Mugen 2 CPU cooler is somewhat similar in performance like yours.

You sure it is not your memory that prevents you from booting? What ram speeds are you running and on what timings?

You could underclock your memory and test what your max BCLK is, just make sure the Uncore frequenzy is x2 the memory speed and then +1.

 

[Grnkjr0]

To alyarb

Could you please post that link to toms hardware where you got that picture from.

Edit. I found it.

 

[alyarb]

if you play games, keep HT off. If you encode or model, turn it on. and keep in mind that graph data is under prime95 full load. a busy 4GHz i7 uses 100 more watts than a busy 3.8 GHz i7. and generally you are correct about the stepping code. If a D0 can do 4 GHz at 1.3v or so, the increase in power would be closer to 50 watts than 100. But on that same token, you could probably do 3.8 GHz with 1.15-1.25 volts. The relationship is still nonlinear and 4 GHz is still more expensive to run than 3.80. Tomshardware did the overclock with a C0 at 1.5 volts, which in my opinion is a failure or a short term overclock for lab use. 1.5 volts does not constitute 24/7 usage. The 3.8 GHz system booted at 1.336 volts, which is so-so. If you can get a D0 to do 3.8 GHz at 1.2 volts there will be definitely be a measurable reduction in power, but you just want to be in the 4 GHz club, and the problem is that it just isn't worth it unless you are guaranteed a golden D0 for all your trouble. That data certainly is prohibitive and I would like to see power consumption scaling on a D0 i7 as well, but something tells me I wouldn't run 4+ GHz unless it could be done with less than 1.3v.

here's the tom's article, starting from page 4.
http://www.tomshardware.com/re...ck-core-i7,2268-4.html

keep in mind, C0 step on a P6T.

 

[JAG87]

No generally the increased amp draw comes from the increased frequency, not the voltage. The voltage is just needed to signal clarity, which brings stability at higher switching frequencies. But with higher voltages comes higher leakage, more watts are dispersed as heat, which means the chip has to draw more amps. Sounds like a small science crash course.

The problem with the chart is Tom's sucks at overclocking, they juice 1.5V into the cpu to get 4.0, so they get massive leakage and massive power draw. Essentially they are idiots, that's why thankfully you are here on Anandtech.

And you are correct, HT actually hurts performance in most games, and is only useful if you do a lot of encoding or multithreaded rendering.

 

[Grnkjr0]

Toms´ upped the voltage to the extreme. 1.5V in Vcore and QPI at 1.5 would never be needed at all. I can run 200x20 with turbo on with my CO at 1.376 and even 1.360

Why they used those voltages was only to show a dramatically effect in the graphs.

 

[JAG87]

Originally posted by: Grnkjr0
To Shmee

Are you sure it is your CPU chip that is the cause to your "low" overclock?

I have a C0 and it can do 205x20 with turbo on = 4.3 GHz. I think my Mugen 2 CPU cooler is somewhat similar in performance like yours.

You sure it is not your memory that prevents you from booting? What ram speeds are you running and on what timings?

You could underclock your memory and test what your max BCLK is, just make sure the Uncore frequenzy is x2 the memory speed and then +1.

I'd take this into consideration as well since you are playing with BCLK. I'd leave it always at 200 to avoid problems.

 

[alyarb]

the problem is that no one writes articles about power scaling with an overclock. hardocp did one and it only went up to 3.8 GHz and they used more voltage than tom's. like i said, a definitive article with C-step and D-step processors would be ideal, but it will never happen. we'd be forced to poll the community for their data. but then your test samples are corrupt/dissimilar.

do you guys really think tom's article intentionally used more vcore than necessary simply because you have gotten better overclocks than they have? I understand many of their publications leaves a lot of data to be desired, but that doesn't mean that they add in fake or wrong data. in a simple power consumption article, would you say that they knowingly post false information, or do they make up their own data? do you guys remember anand's initial i7 review? He had a C0 i7 920 and could not go higher than 3.8 GHz, and anand was also using 1.5 volts. I will grant you that there are some good C0 chips, but they are not in the majority as with D0. It's possible that 1.5v was necessary for tom's 4 GHz test as well. but its also likely that there were better C0s for them to test with. why would they want to post an article with outrageous consumption at 4 GHz?

 

[JAG87]

alyarb, essentially anything above 1.375V produces forced frequency stability and high leakage in my experience. Whether you can get 3.8, 4.0 or 4.2 with that voltage is all up to your cpu, and whether you choose HT on or OFF is up to you, keeping in mind HT on draws a lot more power and produces a lot more heat. I'd say 4.2 HT on with less than 1.375V is still within reason, provided you have adequate cooling, but the chips that can do that are very rare.


Regarding your comments on Toms, personally I think they are idiots, because they should know when an OC comes to an end, and while figuring the voltage required is alright, adding that last configuration into the power consumption graph is plain stupid, because nobody in their right mind would run that setup. They would settle for 3.8.

It's done to show something that should never occur, but the fancy bar charts draw attention. Isn't that the reason why you posted it in this thread

 

[alyarb]

Originally posted by: JAG87
but the chips that can do that are very rare.

that's exactly what i'm saying. i'm not asking how overclocking works. what i'm asking is why you guys are implicating tomshardware for using 1.5v. It is indeed a high voltage, but they wanted data at 4GHz and that is what it took to get there. i'm not disputing the quality of the work on tom's in any way.

the reason i posted it in the thread is because shmee is in a similar situation. He has a C-step that won't reasonably do 4 GHz, and I want him to at least feel that even if he could do 4 GHz, it would be unduly costly with a C0 and unduly troublesome to acquire a D0 that is worth his while.

they were able to get up to 3.66 without increasing the voltage, and you can still see nonlinear increases in power exhibited. this behavior would carry over with a D0 even if it could do 4.2 GHz at stock voltage (but would likely not draw 100 extra watts).

 

[JAG87]

Originally posted by: alyarb
that's exactly what i'm saying. i'm not asking how overclocking works. what i'm asking is why you guys are implicating tomshardware for using 1.5v. It is indeed a high voltage, but they wanted data at 4GHz and that is what it took to get there.

Then they should leave that data out, because it is not indicative of an average 4 Ghz overclock. It's indicative of a very poor 4 Ghz overclock.

Imagine if you were an ignorant Toms-only reader, what conclusion would you draw from that article? What would be your general thoughts about Core i7 at 4 Ghz? And believe me, there are Tons of Toms-only readers, especially in europe.

Originally posted by: alyarb
this behavior would carry over with a D0 even if it could do 4.2 GHz at stock voltage.

No it wouldn't. That power consumption in the that graph is cause purely by forced leakage. The increase draw from 2.66 to 4.2 Ghz at stock voltage would be perfectly linear, just like it is up to 3.66. This can't be proved thought, because the voltage always has to be increased to reach those frequencies, and the forced leakage skews the power results.

 

[alyarb]

well that's why people in europe ride bikes and sit around in cafes. my point is that if anand did an article on his 3.8 GHz C0 at 1.496 volts, the data would be the same. it's just the best that could've been done at the time. unless you have D0 power scaling article, what else is there to go on?

No it wouldn't. That power consumption in the that graph is cause purely by forced leakage. The increase draw from 2.66 to 4.2 Ghz at stock voltage would be perfectly linear, just like it is up to 3.66. This can't be proved thought, because the voltage always has to be increased to reach those frequencies, and the forced leakage skews the power results.

i was referring to the non-overvolted behavior. not the 1.5v behavior. both are nonlinear. but the stock volted overclocks were closer to linear, and would be more representative of a 4GHz D0 @ 1.3v. I think if you gave a D0 1.5 volts at any frequency, that behavior would reappear as well.

 

[JAG87]

The difference is Anand would at least write that their results are poor, and that their power consumption charts should be taken with a grain of salt.

Instead Toms writes this as a statement:

"Reaching 4.0 GHz core clock speed required a massive core voltage increase to 1.50 V."

As if that's the rule of thumb, when it's not. And what's funny is that their article is focused exactly on taking the 920 from 2.66 to 4, and they used the worst possible chip, generating the worst possible article.

 

[alyarb]

heh. well that does compel agreement.

 

[Grnkjr0]

I have read a lot of Toms` articles and I have no issues with the general standard of their work, but many people reading articles will skip to the graphs part and not bother to read the details.

But remember the graph shows peak power and not average use. Peak in 417W is double the 217W average when doing Vantage and that is still more than when playing most games.

But their results focus on the performance gain vs power use and when doing this using such an extreme number as 1.5V will surely mess up the results. What I think most C0 revisions will need is more likely 1.4 in BIOS if they use stock cooler and even less if they have decent cooling.

what people seem to forget is that the stock speed is 2.66 GHz and that running 4.0 GHz on stock cooler would be very hard. I know my chip is an OK chip because I was able to set BCLK 200x20 with turbo on my stock cooler, but it ran very hot and I know I had to go buy a better cooler if I wanted this chip to last just a bit longer.

Back to topic:

I think that a BCLK 180 x20 + turbo = 4.0 GHz should be very easy to do, and I think the issue here is the memory and not the chip.

Edit:. That should have said 190x20 + turbo - sry.

 

[imported_Shaq]

Why did the OP stop at 200? If it is okay at 19x try to take it up to 215 BCLK or so. Most motherboards will hit the limit at 215-220 BCLK. It sounds like the CPU doesn't like the 20 multiplier. Have you also tried 21x (20x with turbo)? And 1.35v is max for qpi and 1.55v for the cpu.

 

[Grnkjr0]

Originally posted by: Shaq
Why did the OP stop at 200? If it is okay at 19x try to take it up to 215 BCLK or so. Most motherboards will hit the limit at 215-220 BCLK. It sounds like the CPU doesn't like the 20 multiplier. Have you also tried 21x (20x with turbo)? And 1.35v is max for qpi and 1.55v for the cpu.

I thought it was the other way around with QPI and Vcore?

Still I think it is an Uncore/memory multiplier/QPI/ memory voltage issue. Still it would be hard to tell unless he posts his settings for all this and what RAM he uses.

 

[JAG87]

Originally posted by: Shaq
Why did the OP stop at 200? If it is okay at 19x try to take it up to 215 BCLK or so. Most motherboards will hit the limit at 215-220 BCLK. It sounds like the CPU doesn't like the 20 multiplier. Have you also tried 21x (20x with turbo)? And 1.35v is max for qpi and 1.55v for the cpu.

Going above 200 BCLK for 24/7 takes some serious VTT. It's not recommended.

1.55V is the absolute max tolerable before instantaneous damage occurs (doesn't mean its safe for 24/7). 1.375V is warranted by intel as a tolerable range for 24/7 reliable operation.

 

[Shmee]

Ram isnt the issue, I dont think. I have it on a 6x mulit; its rated for 1333Mhz. Auto timings. I tried 19x210 earlier, but the system didnt want to post, and then it reset the BIOS settings with a failed OC warning.

 

[Grnkjr0]

What is your memory multiplier, Cas timings, Uncore frequenzy, QPI, Memory voltage?

Have you ever manually set the cas memory timings in BIOS?

You need to list all this in order for anyone to help you because I have a felling you are not familar with what settings is needed for you to get this overclock to work - sry.

Uncore multiplier always needs to be twice the memory speed and +1. QPI and memory voltage depends on memory brand and speed, but it is not same settings in general. It again depends on motherboard and CPU chip, because none are the same.

If you list all that I will try help you get your desired overclock.

 

[arkcom]

Hey guys, I've got a C0 that does 3.8 at 1.25vcore. I just bought a D0 to swap out, but maybe I shouldn't waste my time. Is that about what most D0's are doing at that speed?

 

[Tullphan]

I've just got a D0 & i'm going to wait until cooler weather before I crank mine up. My office is in the hottest room...33c, making it run in the upper 60's when running P95.
Currently it's at 3.5GHz...175x20 & 1.184v.