i7 920 vs i7 860 direct test comparison

[123bob]

OK, I now have one of each. I threw the 920 together today and got it clocked up to 3.7 gig to match the 860 rig reported in the 1156 thread. Then, I took the 860 and matched the multiplier of the 920 (x20), put the QPI multi back to x18, so now the rigs are configured exactly alike. I'm running HFCC on both rigs to.

My plan is to analyze the c**p out of these with respect to power, PPD and WU production, and cost. Some of that, with detailed hardware list will come in later posts. What I want to show here is the setup screenies.

Emu is doing the same thing but at 4 gig with HFCC. This should get interesting...

First, the build pic, if it can be called that....



The 920 is on the left, the 860 on the right. And I'll say again, those Zalman 10x extremes are working REALLY well. Of course, they both have ICD7 paste on them....

OK, setup screenie of the 860 rig first. This is different and updated from the one I posted in the 1156 thread, but at the same CPU clock speed. It has lower CPU multi, higher bclk, and higher QPI speed.




Here's the screenie of the 920 rig. I had to do this in two pics since I'm running an old PCI virge card in it...This is on an MSI Platinum SLI board.





There was a trick to getting it to bios overclock since it has this set of dip switches for "automatic" overclocking. It was set default to on and kept overwriting my bios settings... Of course, I could find nothing in the manual on how to disable it. I just tried putting all the dip switches in the off position and that worked..... After that, the clock came easy. I have a good amount of margin left in this board so I should end up at 4 gig when all is said and done. This will be part of the power analysis. I want to establish the "sweet spot" on both rigs.

OK, so both are crunching away, I should have initial PPD/production results in a few days. I got curious when jcool posted data that indicated one of his 920 rigs was producing significantly more WUs per day than my 860.

Now I have a 920 "control" to compare the 860 to. (At least that's my excuse for buying it... Add that to the fact that I picked up an extra PSU a few days ago and it was getting lonely....)

The 860 was pulling 183 watts, Boinc loaded. I'll have to check that again since I changed the QPI around. The 920 is pulling 224 watts Boinc loaded. These are just initial figures. I'll put more detail around this later.

The BOINC bench on the 860 is 3,376/11,851. The 920 rig is 3,382/10,853.

I would also say to disregard the estimated completion WU times on both rigs. The 860 just got a batch of new HFCC WUs, it was running HCC. The 920 rig has not completed any WUs yet so it doesn't really know what the real times should be. I expect that to come down into the 4 hour range.

Wow, busy day today, and more hamsters in the stable. That will teach me to buy a power supply....

Bob

[INFRNL]

this will be interesting. thanks for your time Bob!!!

[123bob]

You're most welcome! Any excuse to add some hamsters on...This is in the name of science after all...

OK, the first of many updates, this one somewhat unexpected. I was browsing the forum on the 860 rig (the one I'm typing on now) and it blue screened on me...

I had noted with the changes I made to bring up the QPI, it somehow dropped the reported vcore in CPUz. I don't know what that is about... I suppose the QPI boost requires more work from the CPU. So, I boosted the vcore by .007 volts. My Boinc loaded CPUz is now 1.232 on the 860 rig. We'll see how she runs. I can't wait to see how it produces with this subtle observation of vcore change.

Bob

[DAK1640]

ok 123bob///I am on a need to know basis...report in already

[123bob]

ok 123bob///I am on a need to know basis...report in already

Yes Sir! Reporting in. We wouldn't want to keep Da Goosey waiting now....

I had another blue screen overnight on the 860 rig. I'm thinking I didn't boost vcore enough. So, I did the math and boosted it back up to the level it ran on the 1156 thread. Namely, CPUz loaded with Boinc now reports 1.264 volts. I think it should run smooth now.

The 920 rig ran fine and the estimated completion ranges for HFCC WUs is down to about 5 hours each. I expect that to drop a bit more.

One thing to note is that my numbers for the first few days on the 860 will be skewed a little higher since I still have some HCC WUs pending validation. Those need to clear out before I'll get true numbers. I may try to back them out of my score by doing the math on individual WUs, but that might be more of a PITA than it's worth.

Stay tuned,
Bob

[123bob]

OK, nine hours later and the vcore bump on the 860 seems to have done the job. Crunching away on HFCC.

Bob

[NaeKuh]

need to see the times on the same work group like bob's. :P

Anyhow great thread bob, waiting for your results.

[123bob]

Thx Emu! This should get real interesting over the next few days.

One thing I noticed though is that HFCC now appears to be at a quorum of 2 also! We'll see how we are treated in the point department. What a time for them to change it up....Oh well, I'll keep on running it for a bit.

My rigs seem to have both settled into estimated completion times of just a few minutes over 4 hours. We'll see how that works out too. If it's correct, I should do about 48 WUs per day on each. That would put it right in line with jcool's data.

Bob

[[XC] gomeler]

Emu, does your Core i7 860 require that much vcore to be stable? Insane

Good luck with the tests guys, I suspect the Core i7 860 will become the chip of choice due to the lower platform power consumption.

[emuexport]

Thx Emu! This should get real interesting over the next few days.

One thing I noticed though is that HFCC now appears to be at a quorum of 2 also! We'll see how we are treated in the point department. What a time for them to change it up....Oh well, I'll keep on running it for a bit.
Bob

I reckon! At least we can learn from each other right! My feeling is that my i7920 will win out purely due to the high Vcore my 860 chip requires to reach 4 GHz. As you did I noticed a significant jump in VCore requirements when pushing above 3.7GHz. As we will be running these 24/7 the power requirements will definetely add up.

I've taken some measurements but as one is WC and the other is not I need to work out what the draw of the pump is on the system as otherwise they are identical in terms of cooling fans!

Emu, does your Core i7 860 require that much vcore to be stable? Insane

Good luck with the tests guys, I suspect the Core i7 860 will become the chip of choice due to the lower platform power consumption.

Not quite that much I accidently left the Vdroop control on which goes a bit bonkers IMHO. So now its back to 1.4 in bios which is 1.435 under load overcompensating for droop perhaps! I thought normally it went lower than what you set it at!
I'll leave it crunching at this setting just so we can see what the difference is in terms of PPD when comparing clock for clock 1156 to 1366

Oh and gomeler if you have a bit of a hunt around you will find 1.4Vcore is pretty normal from my reading. I do however believe my chip isn't the best but it is cooled by a triple radiator and is dumping some serious heat in the loop!

[INFRNL]

Does the 860 still draw less pwr than the 920 when overclocked and crunchin its butt off?

[jcool]

Emu, can you bump the Uncore mult on the i7 to 18x? So it will match the 860

[DAK1640]

Da Goose wants a prelim update...Report in soldiers...

[emuexport]

Does the 860 still draw less pwr than the 920 when overclocked and crunchin its butt off?

Not on mine it doesnt. I think it takes too much juice past 3.7GHz and mine is probably not a star performer.

Emu, can you bump the Uncore mult on the i7 to 18x? So it will match the 860

uhhh wut I am such a noob at this still. So used to clocking Q6600s the terminology is still growing on me.

Anyways good pickup jcool and now Uncore is at 18x to match the i7 860!

Da Goose wants a prelim update...Report in soldiers...

Still crunching here. My study is nice and toasty now and very messy. Gotta stop playing with the crunchers and tidy up its a little embarrasing


And yes the radiators are a little dirty but when I get a WC CPU block for the i7 920 they'll get a wash and dry!

[Shaotai]

And yes the radiators are a little dirty but when I get a WC CPU block for the i7 920 they'll get a wash and dry!

Dang! THOSE ARE DIRTY!! How much dust you have in that house?!! LOL

Great test guys. Very interested in hearing how it turns out!

[Otis11]

And yes the radiators are a little dirty but when I get a WC CPU block for the i7 920 they'll get a wash and dry!

Why not just take a shop vac to it quick? That'll suck the dust right off...

[123bob]

Da Goose wants a prelim update...Report in soldiers...

I will release no data before it's time....

I want to see tonight's update. Then I'll post up some individual WU results. I won't post overall machine totals until they settle in. The HCC WUs have to clear out of the pending val queue, on the 860 rig, before I can accurately report machine totals.

I hate rushing data out just because. It does no good and there's no science to be had. There is no "sound byte" results on this test. It will take a week or longer to draw final conclusions.

As to the power question, I'll move the kill-a-watt back over to the 860 rig. That has had some changes to it, enough to warrant re-visiting it.

The 920 rig is drawing 223 watts, Boinc loaded. It's drawing 126 watts at idle.

The 860 rig is drawing 186 watts, Boinc loaded. It's drawing 94 watts idle.

So, the slight bump in Vcore, along with a significant boost to QPI speed only cost me about 3 watts! That's cool, I expected it to be more.

I suspect Emu and my chip are similar in that in order to hit 4 gig, significant vcore will be needed. I'll try that out in a couple of days. Right now it's 3.7 gig, head to head.

Oh, and now it's been more than 8 minutes since Hicks post below.....I'm a cra*py typer.....

Bob

[[XC] Hicks121]

The 860 rig is drawing....<placeholder, be back to update in a few....>

Hey! IT's been 8 minutes......

[123bob]

OK, some results.....

First a rig total. I'm surprised they came this close today....Remember though, take this with a grain of salt since I have HCC units still in the pending val queue....We'll see how it settles out. The queue should largely empty out by tomorrow night's update, with any luck. I'll keep an eye on it and report when it's gone.



This is the interesting one. Some WUs.



This would imply that the QPI setting has a profound effect on the machine's production. I bumped the 860 up from a low level and the average completion time is VERY close to the 920 rig.

Note that I'm still taking somewhat of a beating in the quorum score, so I put more value on the completion time category. The total machine scores will be interesting in a few days. Right now, no conclusion can be drawn about that....

Bob

[123bob]

Quick question for the Euro folks. Do your power companies also charge by the KWH like the USA? What are some sample costs for you guys, in Euros? With the dollar value going up and down, I'm trying to make it more meaningful for you guys. Your power costs don't fluctuate like our exchange rates do....

If you have suggestions, let me know.

The reason I ask is that when I do power cost analysis, I'd like to have Euro columns in it too. I plan on having a power cost range of about $0.05 to $0.35 per KWH for the US folks.

Thx,
Bob

[Martijn]

Quick question for the Euro folks. Do your power companies also charge by the KWH like the USA? What are some sample costs for you guys, in Euros? With the dollar value going up and down, I'm trying to make it more meaningful for you guys. Your power costs don't fluctuate like our exchange rates do....

If you have suggestions, let me know.

The reason I ask is that when I do power cost analysis, I'd like to have Euro columns in it too. I plan on having a power cost range of about $0.05 to $0.35 per KWH for the US folks.

Thx,
Bob

You honestly don't want to know. the Netherlands is quite expensive, we pay ~€.40 net for 1 kWh. I believe Belgium and Germany were about as expensive, but I could be wrong. Do keep in mind that all our power lines are underground and we rarely have black- or brownouts. I never needed my UPS, the last blackout was 3 years ago.

[123bob]

You honestly don't want to know. the Netherlands is quite expensive, we pay ~€.40 net for 1 kWh. I believe Belgium and Germany were about as expensive, but I could be wrong. Do keep in mind that all our power lines are underground and we rarely have black- or brownouts. I never needed my UPS, the last blackout was 3 years ago.

Yes, I want to honestly know! Jeebers, I hope you retired that northy I sent you..... It's a power pig......

JEEPERS KHRISTOS!! Hope you don't mind if I mix US slang with Greek.....

Power here in the Pac NW is $0.05 per KWH. It's been that cost for years....

So for you Euro folks, the high end of the power cost analysis is what you would be interested in. I can see why. I will ask for your help in the initial cost of these rig items too.

This thread just got WAY more important to me. For any of you across the pond, LMK any specific tests you would like me to run. I intended this post to find the right production vs "green power" point. I think the 860 rig is there now. The point for the 920 remains to be seen...

Honestly, I live in the biggest consumer nation in the world. We want it cheap, fast, and good. In my Engineering career, I've always been told you can have any two, but not all three. So far, that has held true.

My best,
Bob

[Martijn]

I did indeed retire it, but I gave it to a friend of mine. He's now using it for his server. 2.26 northy's aren't too bad though, especially when undervolted.

[v0dka]

You honestly don't want to know. the Netherlands is quite expensive, we pay ~€.40 net for 1 kWh. I believe Belgium and Germany were about as expensive, but I could be wrong. Do keep in mind that all our power lines are underground and we rarely have black- or brownouts. I never needed my UPS, the last blackout was 3 years ago.

Sorry, but as a fellow Dutchman Im going to call this BS. Two minutes in google gave me the following result, and this is from a company that is certainly no price fighter.

http://www.essent.nl/content/thuis/p...rder/index.jsp

€0.09 for the power and €0.12 for taxes. So that's €0.21 per KWH

BTW, the fact taht our net is underground does not warrant a high electricity price. You pay addiotnal costs for the power grid (a fixed amount).

[Martijn]

Sorry, but as a fellow Dutchman Im going to call this BS. Two minutes in google gave me the following result, and this is from a company that is certainly no price fighter.

http://www.essent.nl/content/thuis/p...rder/index.jsp

€0.09 for the power and €0.12 for taxes. So that's €0.21 per KWH

BTW, the fact taht our net is underground does not warrant a high electricity price. You pay addiotnal costs for the power grid (a fixed amount).

Hmm, that's certainly interesting. I never really bothered to look into eneco's exact prices. Apparently they're at 25c/kWh, which is nowhere near as high as the 40c/kWh I was told about. Adding the additional costs to that might up that 25c to 30c at most. I guess I am going to have a talk with the dad, it looks like he was exaggerating like I was

Bob, you had B3 quads left?