i7 920 vs i7 860 direct test comparison

[BlindFreddie]

@Emu: Did you notice that in the last pic on your post #223, CPUZ shows 860 Vcore = 1.096V @ 4095 MHz. But the max core temps are up around 70deg, which is what you'd expect with 1.4V. Must be a CPUZ error (?)
Any thoughts on why your 860 seems a bit slower than the 920, while 123bob found the opposite?

Re. power consumption: The extra Vcore needed on your 860 seems to be making it run about the same max temps as the 920. If they have the same cooling, that would mean the same amount of heat energy coming out, ie same CPU power in. Other relevant info: A few days ago I read a review where they got an estimate of the mb power consumption and figured the X58 chip was causing their 920 board to use about 20W more than their P55 one.
If you get a 2nd power meter, it would be a good idea to check them against each other.

[123bob]

Good work Emu. I have not had a chance yet to fully go over your results yet. Will do this afternoon. I think one thing we would like from your tests are the 4 gig point, matched parameter for parameter. You may have already done that, I'll have to look close at the results. Power consumption for this 4 gig point as well.

I don't think you HAVE to get another power meter, I just switch mine from rig to rig when I'm testing power. But, if this works to get more HW, that's cool too. Those meters are almost cheap enough to put one on each rig. It would be nice to see one that had multiple taps or rack mounts, for my case.

As far as the variance in our 860 vs the 920 results, I'm beginning to think it has to do with the particular WUs we are running? I noticed when I switched test WUs, my numbers changed a bit.

I'm going to be on the road this weekend up until next Friday. I'll have forum access, but that's about all.

I'll leave my rigs crunching away at the 3.7 point. I'll be able to report some daily PPD from this.

Regards,
Bob

[emuexport]

@Emu: Did you notice that in the last pic on your post #223, CPUZ shows 860 Vcore = 1.096V @ 4095 MHz. But the max core temps are up around 70deg, which is what you'd expect with 1.4V. Must be a CPUZ error (?)
Any thoughts on why your 860 seems a bit slower than the 920, while 123bob found the opposite?

Re. power consumption: The extra Vcore needed on your 860 seems to be making it run about the same max temps as the 920. If they have the same cooling, that would mean the same amount of heat energy coming out, ie same CPU power in. Other relevant info: A few days ago I read a review where they got an estimate of the mb power consumption and figured the X58 chip was causing their 920 board to use about 20W more than their P55 one.
If you get a 2nd power meter, it would be a good idea to check them against each other.

That screenshot is taken at idle. Usually I set up my PCs in the evening and let them crunch and I think they finish ther 24 WUs before lunch depending on the clocks.

Not sure why my 860 is lsower than the 920 though I think it depends on a lot of things and this is the point of the study right

It will take a while to figure out what is what. I will try and stay on the 4GHz point and get power measurements.

Good work Emu. I have not had a chance yet to fully go over your results yet. Will do this afternoon. I think one thing we would like from your tests are the 4 gig point, matched parameter for parameter. You may have already done that, I'll have to look close at the results. Power consumption for this 4 gig point as well.

I don't think you HAVE to get another power meter, I just switch mine from rig to rig when I'm testing power. But, if this works to get more HW, that's cool too. Those meters are almost cheap enough to put one on each rig. It would be nice to see one that had multiple taps or rack mounts, for my case.

As far as the variance in our 860 vs the 920 results, I'm beginning to think it has to do with the particular WUs we are running? I noticed when I switched test WUs, my numbers changed a bit.

I'm going to be on the road this weekend up until next Friday. I'll have forum access, but that's about all.

I'll leave my rigs crunching away at the 3.7 point. I'll be able to report some daily PPD from this.

Regards,
Bob

I will try and match all the 4GHz results but sometimes these PCs just dont want to play

It seems my DFI forces the 2 x DRAM frequency ie 2x10xBLCK = 2 x 20 x BLCK (Minimum) so I cant make them match that way It complicates it even more as the P55 board has the uncore and qpi limited in terms of settings.

I'm not sure about WUs being a factor or not but maybe once I have a definete set of points I can get another set of WUs for comparison. Hell maybe even try a different project

[123bob]

I will try and match all the 4GHz results but sometimes these PCs just dont want to play

I'm not sure about WUs being a factor or not but maybe once I have a definete set of points I can get another set of WUs for comparison. Hell maybe even try a different project

We can only do what the rigs allow. At least you can get 4 gig on the 860....

On the other project WUs, maybe we give HCC a whirl? I won't be able to do it until I come back next week. Right now I'm scrambling to get some work stuff done before I go.

Bob

[emuexport]

We can only do what the rigs allow. At least you can get 4 gig on the 860....

On the other project WUs, maybe we give HCC a whirl? I won't be able to do it until I come back next week. Right now I'm scrambling to get some work stuff done before I go.

Bob

4.1GHz mate!

I might stop with this set and start on another project once I am happy I have enough data points. I can always go back later right!

HCC it is then for the next set of WUs

so now you are putting work before crunching where are your priorities

Have a good trip and I'll keep testing while you are away. I know you'd do the same for me

[INFRNL]

How are you doing now INFRNL? Good to see you have the 3440 @ 4gig. Screenies please, if you get a chance? As I said before, I'm real interested in the performance of the X3440 chips. That COULD be the equivalent of the Q6600 G0, for crunchers with Kentsfield farms. Cheap, Fast, and Good.....Wow, you rarely get all three....

Looks like you are having some mount problems? If you haven't read it yet, the ICD7 thread here is very good at refining mount technique. I learned A LOT during the test there. The "pea blob" is your friend, no matter what paste you use.

Can we help out with brackets? I may have some various 1366 brackets here, possibly some stuff that could be used for 1156. Let us know what cooler you are trying to match to what. I'm unclear on what mount problems you're having. I'm sure we could get you covered.

Regards,
Bob

I appreciate the concern and willingness to help. I am not the type to ask for help unless I have no other option.
To clear things up its hard to describe the original problem; However I ordered another Megahalem rev B. I love these heatsinks.

I'l try finishing off the tests at 4GHz but the 920 is proving tricky atm doesnt seem to want to play with a higher QPI needs alot moe juice!

INFRNL if you can get some results on the X3440 that would be great as it may be the next one I "need".

I am currently running a 920 D0 that I bought from a fellow member here to help him out in his journey for a Xeon. I am running 4ghz @ 1.264 ATM I havenot yet tried less voltage and I have no idea what I am doing. This is on my bloodrage. My current load temps with IC D7 are anywhere from 63-71c.

I am running my W3520 on my biostar 4ghz@1.248v, again have not tried lower voltage asnd I still do not know what I am doing This rig is running a TRUE with AS5 I think. My load temps are mostly in the 70's and sometimes to 80c. I am going to reseat with IC D7 to compare results.

I do not think the 200mghz increas gives me much more production, but as we all know every little bit counts and amounts to a lot more in the end


Now for the X3440 update. I sent my UD2 in for RMA. I am leaving on Sunday to San Diego so you will have to wait a week for any kind of real stats. I have only been running HFCC on the X3440 and 1 other rig. I have not run real test comparisons as you gys, but it looks like the X3440 holds its own

Now for the bad news. I do not know whats going on and have not had time to play, but I cannot get this UD4 and the X3440 to run @ 4ghz now. The only thing I really changed was obviously the mobo, but same manufacturer. the only other thing was the ram. I am now on Mushkin ram, was on ocz.
I will try to figure it out when I get back from vaca.
I probably need to update the bios for 1, put the ocz back in for direct comparison from the UD2. My temps are in the mid 60's. Hopefully something didn't get messed up when the UD2 got damaged. I wouldn't think so, but who knows. Maybe this mobo has slightly different settings that I am not aware of

Well at least for now I am cruning away @3.8@1.312v. I may be able to lower the voltage as well. I do not recall my voltage before for 4ghz; I thought it was 1.36v I think I tried @ 1.36 on this board.

Lastly before I put DAK to sleep again Do you guys want me to run tests like yours? I will have to re-setup everything to match again which is kind of a pain as EMU said. I still think the X3440 is the best option and the cheapest especially for a dedicated cruncher.

[INFRNL]

I updated the bios last night, I am pretty sure I set vcore to 1.4v @ 4ghz. CPU-z/id 1.52.3 is showing 4ghz@1.376v It has been running since my above post. I will monitor it till tomorrow. Hopefully it makes it.

temps showing 70-75c at full load. So temps went up a decent amount still not bad though.

I might drop it back to 3.8 ghz while I am gone to be safe. Cannot afford to lose a weeks worth of production, but would not kill me either.

EDIT: After this message, it froze up. Had to restart, went into the bios. I guess it was on auto for the vcore. I set it to 1.4v and its been running for the past 7 hrs. Will see what it does by morning. I might just see if it will make it for a week I will have to play with it more when I get back from vaca to see if I can get it stable at a lower vcore END EDIT

[emuexport]

Well all of the PCs in my study are now water cooled and have a triple rad either fesser or thermo on them. The i7920 now has my 4870X2 even considering this the temps are at least 10 DegC cooler than with air

So have we given up on this study? I couldnt match any more combinations of the 920 vs 860 so gave up.

emu

[Otis11]

Haha, I don't think you can call it giving up... just finished enough cases - the whole diminishing returns thing...

Great job guys!

[Snow Crash]

How about software aspects ...
There all all kinds of tweaks for benchies, how about for cruching?
Starting with boinc settings (how often write to HDD, not using the screensaver, memory % usage, does reducing network connection from always on to 2 times per day matter?)
Then how about Winblows settings ... how many/ which services should be shut down? Does killing explorer.exe on headless cruncher help, turn off automatic updates, windows time ???. I know if you run in diagnotics mode all you need to do is turn rpc on and boinc will run just fine but I wonder if this produces any better results and leaves the question of how to get the data to/from WCG. I am not sure what the minimum networking services necessary are ... is it better to connect through a proxy and let that machine handle all AV duties?

[123bob]

How about software aspects ...
There all all kinds of tweaks for benchies, how about for cruching?
Starting with boinc settings (how often write to HDD, not using the screensaver, memory % usage, does reducing network connection from always on to 2 times per day matter?)
Then how about Winblows settings ... how many/ which services should be shut down? Does killing explorer.exe on headless cruncher help, turn off automatic updates, windows time ???. I know if you run in diagnotics mode all you need to do is turn rpc on and boinc will run just fine but I wonder if this produces any better results and leaves the question of how to get the data to/from WCG. I am not sure what the minimum networking services necessary are ... is it better to connect through a proxy and let that machine handle all AV duties?

That's a topic for another thread. I think this one is done as far as whether the 860 can keep up with the 920. It can. That was the purpose of this thread.

You are most welcome to test the items you suggest. I have done some of that in the past. I'm out of time for now. Others here have played with some of the things you suggest too.

My next task is to reshuffle the farm and make some room for the new rigs. I've got to get these new rigs off my bench.

I've also been experiencing way more blue screens on my GPUGrid rigs than I like. That has to get fixed. It's going to be an nVidia driver issue I'm afraid. All the blue screens relate to video issues with an nvidia driver file.

I may have to drop the Grid if I can't fix it. (They will go and fold for awhile...) This is occurring on the rigs I have on Scott's account, all with 260-216s....It may be time to bring them home and re-load OS too.

Regards,
Bob

EDIT: Final data for this test. 7 days of crunching scores. The 920 had about a half a day of being down in the last couple of days in this. All HCC.

860
10/25/09 0:007:16:37:19 23,547 38
10/24/09 0:007:00:38:43 20,930 35
10/23/09 0:008:01:18:14 24,455 40
10/22/09 0:008:11:02:22 25,449 42
10/21/09 0:006:06:00:00 18,917 31
10/20/09 0:007:06:14:45 20,492 36
10/19/09 0:008:15:56:43 25,560 43
Average - 22,764


920
10/25/09 0:005:01:41:53 17,182 28
10/24/09 0:005:09:08:49 18,097 30
10/23/09 0:008:21:54:00 29,707 50
10/22/09 0:007:03:35:25 24,326 40
10/21/09 0:008:06:12:26 26,991 46
10/20/09 0:005:21:24:32 20,646 33
10/19/09 0:004:19:39:05 16,645 27
Average - 21,942

For the varience we see in the qourum, I'd call it pretty close to even. :END EDIT

[Jaco]

Anyone familiar with Windows7 ?

I've read that win7 is very energy efficient !
System consumes less power with Windows7 than Vista or other windows versions.

http://www.anandtech.com/systems/sho...px?i=3666&p=13

Another interesting thing we should investigate

[artemm]

Anyone familiar with Windows7 ?

I've read that win7 is very energy efficient !
System consumes less power with Windows7 than Vista or other windows versions.

http://www.anandtech.com/systems/sho...px?i=3666&p=13

Another interesting thing we should investigate

I just built an AMD rig for the family and installed Win7 as the primary OS. I did notice some very interesting power settings.

Still, seems like they're more useful for PCs that idle for long periods of time, not rigs at full bore 24/7 on CPU and GPU.

The new option that jumped out at me is the Active/Passive cooling policy, but, again, it's only useful when slowing the processor.

http://blogs.msdn.com/e7/archive/200...fficiency.aspx

I think you can save some energy by making BOINC access the HDD/Network less frequently and setting an aggressive policy to shut these things down.

[Jaco]

artemm,

check this out

http://www.infoworld.com/d/windows/w...r-325?page=0,2

[Otis11]

Yeah, but that's running 4 threads instead of 8 like we do. We shouldn't have nearly as many transistors idle...

Although I can see it being more power efficient with all 8 threads active by using the halt command on the (few) idle transistors each clock cycle, that would only add up to a few watts. Good, but i don't think it'll ever add up to enough to repay your investment...

[artemm]

artemm,

check this out

http://www.infoworld.com/d/windows/w...r-325?page=0,2

Very cool. Intel SMT only? I guess they may be working with Intel because HT appears to be energy inefficient to start with:
http://www.theinquirer.net/inquirer/...hyperthreading

Still, it's awesome if Win7 "fixes" that issue.

The only way to find out for sure is to 123bob it and find out what a system draws with Vista with HT off, HT on, AMD and Win7 with HT off, HT on, AMD.

I'd do it, but I don't have my i7 build completed yet!

Edit: Nvm. Mailman proved me wrong. No reason why I can't do it now!

[emuexport]

Noice mate!

So you're gonna look into the power saving options and HT. Could be interesting!

Good Luck and remember to take some patience!

[INFRNL]

I agree that P55/X58 are damn near identical. I did not get to do the tests you guys did, but I have been watching and I am getting pretty much the same production between the 2. Been running @ 4ghz no problem

[123bob]

The only way to find out for sure is to 123bob it and find out what a system draws with Vista with HT off, HT on, AMD and Win7 with HT off, HT on, AMD.

I'd do it, but I don't have my i7 build completed yet!

Edit: Nvm. Mailman proved me wrong. No reason why I can't do it now!

Or maybe this should be "Emu-bob-it"? Are we creating an addition to our unique cruncher language? At least it is "123bob-it", not "Bobbitt" like the famous case here. Ouch!! That would be some different type of testing....

Everyone here is capable of running this kind of detail testing. Just throw out a hypothesis, and test according to that. I HIGHLY RECOMMEND THIS LINK HERE, for careful reading.

Bring your idea on here, start a thread, and like this thread, the team will help clarify what constitutes valid test methods and results. Read a few pages back. Snow Crash provided a really valuable suggestion to how we can take out some variance in the results, without having to go through weeks of test time. I was really struggling with that. Kudos to SC!

Critical thinking skills are very valuable for many things other than crunching, forum browsing, rig testing, etc. It can provide a very large income, keep you out of the worst mistakes you can make in your life, help balance your checkbook, yada, yada.....

Keep asking questions!! And be willing to pony up for the pain of providing answers to those questions. That's what makes our team so effective.....So few, and yet so much...

...and as to the mailman providing, what are your results so far? What makes a fair test? Two identical rigs? ......It might be best to start a thread. This topic of an op sys providing power efficiency to our crunchers deserves a thread. I'm watching this with interest now. I can see 5 "family packs" of win 7 if it proves out......

Or, someone send me 4-5 valid Win 7 keys and SW. I'll help out.......

Regards,
Bob

EDIT: Just for yucks, I looked at the two most scientific threads I've done on XS and looked at the read counts on the thread. The best was the ICD7 thread with just over 43 thousand reads. The next is this thread, with almost 7 thousand reads. Hmmm. Science wins.... :END EDIT

[emuexport]

I like 123bobit personally!

In other news the 860 is much happier not sitting at 1.4Vcore!

Just trying to find the sweet spot for it in terms of power consumption.

[123bob]

It looks like 3.7 gig was the sweet spot for my 860. It's been happily crunching away cool and calm with no hitches.

Bob

[emuexport]

Thats great Bob!

I have mine running at 3.7 GHz as well with 1.24VCore load BOINC will see if I can get it any lower

Pushing the 920 to 4.2GHz in the mean time just for kicks doesnt need as much VCore to get there compared to the 860 either

[INFRNL]

Did we ever determine the difference between 4ghz and 3.8 for ppd comparison? Can't remember and I am a bit lazy this week.

I am starting to think the same as Emu. I could have sworn I was running my X3440 @ 4ghz@1.36v, but now I need at least 1.4v or I have lost my head. It still runs fairly cool on air, but cooler is always better. I do not really like the fact of running 1.4v either.

3.8/4.0 shouldn't really make much of a difference. my 920 produce comparably 3.8/4.0

[onex]

canceled

[Chumbucket843]

Yeah, but that's running 4 threads instead of 8 like we do. We shouldn't have nearly as many transistors idle...

Although I can see it being more power efficient with all 8 threads active by using the halt command on the (few) idle transistors each clock cycle, that would only add up to a few watts. Good, but i don't think it'll ever add up to enough to repay your investment...

HT is a big jump in efficiency compared to when it is off. it does a much better job of keeping the cores under full load. thats why its hotter. the wider you make a cpu, the harder it is to keep under load.