i7 920 vs i7 860 direct test comparison

[INFRNL]

Todays update,
K so my impatience got the best of me. I do have room for a full atx board, but I ended up getting the P55M-UD4 from my local store. Main reason is that I cannot really wait till mid next week and get everything set back up. I am currently going through a major remodeling of all mt rigs due to the hardware I have to work with atm. You can see from all my stats that my production is down. I got my new 920 rig up last night but still have some things to do. lost train of thought; I am going on vaca to San Diego this coming weekend, so I wanted to get the X3440 back up.

Good news is it was only the mobo; So I will get it rma'd and resell when it arrives back.
I have the X3440 on the ud4@4GHZ@1.3xV. I am currently in the process of switching to a new HDD. This rig will be back up sometime today. I think I need to reseat the megahalem as my idle tems are a little higher than the last time.

Even though this rig now costs as much as a 920 setup; I still wanted to see for myself how it compares to the 920 in all respects.
I will let you know how it goes.

One of my 920 rigs is down. I have a Bloodrage and had to use the 775 heatsing mount; for some reason my idle temps are in the 60's. I have reseated 3 times with no luck. I am not sure what is going on. may have to run stock clocks till I can get another true or megahalem.

I think this covers it for atm. I will keep yopu posted with any updates.

[DAK1640]

Geesh INFRNL, I need to take a nap after reading about how busy you are with all your rigs. Congrats on impatience winning out and getting your UD4. I hope it delivers everything you expected and more.

[123bob]

Hooray, this just got more complicated

Hehe, it got even more complicated after I tried to start the run on the 920...

I loaded the test WUs off the USB stick I've been using only to find that for some reason, the first WU shows up as aborted....Hmmmm. I tried to recover it for about the last 5 hours. I screwed around with the client_state.xml file trying to get it to believe the WU was not aborted. It's beyond me...No luck.

Sooo, I have a new set running on both rigs now. I will have to re-baseline everything since The WUs changed...

This time I copied the test WUs in about four different locations. Emu, you might want to do this with your WUs too....

I have absolutely no idea what happend to the stick, or why it would only change the status of one WU. It gave me a "can't find file" error on slot #3 once, I'm wondering if the USB stick blew out a file.

Through this process, I tried to re-format the stick and it couldn't do it.....The stick is going in the trash can, after a "tune up" with my 5 pound sledge.

Oh well, back to the drawing board...I hope the offset of the old units to the new ones is consistent. I should be able to work with that.

So, from this point in the thread forward, be aware I'm using a new set of test WUs.

I'll keep the data sorted out to avoid mixing old and new. That's about all I can do.

Bob

EDIT: As of 16:00 PDT, both rigs are running the new WUs at 3.7 gig. I decided to re-baseline first. Report tomorrow....:END EDIT

UPDATE EDIT: The probelm above was caused by trying to run WUs that were expired past the deadline time. Setting the test rig clock backward while running a test pass fixed the issue. I just have to remember to set it back to real time when I finish a run. :END EDIT

[123bob]

OK, almost had an "Oh Shiz" moment. I noticed I was running two different CPUz versions on the tests. 1.49 is what I ran on the 920. I ran 1.52.2 on the 860. There's a RAM parameter there called "Raw Refresh Cycle Time". I didn't have that on the version 1.49 prog.

I installed 1.52.2 on the 920 rig and you will all be happy to know that they match. 88 clocks for "Raw Refresh Cycle Time on both rigs...... Whew, I almost would have had to hang myself on that one. <NOT> Now I can put the rope away...

Both rigs are still cooking, with 9 WUs complete on the new 3.7 gig baseline. Right now the 920 is winning by about 2.8 min/WU, but we'll see, 15 each to go yet.

Emu had an even bigger delta in his test. I wonder if these WUs are different from my first batch somehow? This is indeed getting more interesting....

We may need some more "clock-for-clock", "setting-for-setting" data before we go off on any other tangents? Let's see how my new baseline comes in...

We're trying to make this test useful, without a lot of side tracks, particularly since they cost Emu and I about 12 to 16 hours of dead production for each run, depending on clock......Check my thinking here..... The initial goal of the thread was to verify that the 860 and by extension, it's Lynnfield brothers, were up to what the socket 1366 chips had to offer in terms of crunch power. A second goal was to show which socket had the best power efficiency for the same production.

This is to offer choices to our teammates to buy with. Some of you are already going the Lynnfield route. DAK and INFRNL have joined us. (BTW, I'm still drooling over an X3440 rig....I just might have to have one...) Based on our earlier testing, I don't think you will be at all dissapointed. My 860 cranks out the points. It's the only way RSMura is not running me over while my other rigs are gone....

I would ask you, the audience, are our goals as they were above? Anyone want to see something different? I took the short side track to get a feel for how the exact same parameters, on the same rig vary over different runs. While this showed some variance with only one trial, it is hardly statistically valid, but it costs production time.... This test method is new to all of us. Perhaps more runs like that are of value?

The other factor to remember is that Emu and I are running different sets of test WUs. The only thing each of us can do is run a 920 against an 860. Any value to us getting together and figuring out who sends test WUs to whom? Remember the problem with this is in the various account keys. Those go with the data. I'd be willing to send Emu mine, it's just a lot of data, about 130meg. I've come to trust Emu. Just don't let those WUs escape into the wild and bring down the planet's mainframes.... The issue here is that Emu's have more run time since I managed to corrupt my first ones....

One thing I mentioned before that may be of use is to refine jcool's formula for PPD estimating. I think we could do that with this test method, but we would need to decide on what we are looking for. We all know straight bclk makes a good difference. We have some indication that QPI makes a pretty good difference too, Uncore/NB makes some difference, and RAM timings may make a yet-to-be-determined difference. This is with HFCC WUs though....

What I will say is that I'm not willing to burn an infinite amount of CPU time here. I'm about producing valid, useful WUs. That's why I farm. On the other hand, I'm willing to maximize my production too. Where is the balance point is what I'm asking the team?

Bob

[emuexport]

Both rigs are still cooking, with 9 WUs complete on the new 3.7 gig baseline. Right now the 920 is winning by about 2.8 min/WU, but we'll see, 15 each to go yet.

Emu had an even bigger delta in his test. I wonder if these WUs are different from my first batch somehow? This is indeed getting more interesting....

We may need some more "clock-for-clock", "setting-for-setting" data before we go off on any other tangents? Let's see how my new baseline comes in...

We're trying to make this test useful, without a lot of side tracks, particularly since they cost Emu and I about 12 to 16 hours of dead production for each run, depending on clock......Check my thinking here..... The initial goal of the thread was to verify that the 860 and by extension, it's Lynnfield brothers, were up to what the socket 1366 chips had to offer in terms of crunch power. A second goal was to show which socket had the best power efficiency for the same production.

This is to offer choices to our teammates to buy with. Some of you are already going the Lynnfield route. DAK and INFRNL have joined us. (BTW, I'm still drooling over an X3440 rig....I just might have to have one...) Based on our earlier testing, I don't think you will be at all dissapointed. My 860 cranks out the points. It's the only way RSMura is not running me over while my other rigs are gone....

I would ask you, the audience, are our goals as they were above? Anyone want to see something different? I took the short side track to get a feel for how the exact same parameters, on the same rig vary over different runs. While this showed some variance with only one trial, it is hardly statistically valid, but it costs production time.... This test method is new to all of us. Perhaps more runs like that are of value?

The other factor to remember is that Emu and I are running different sets of test WUs. The only thing each of us can do is run a 920 against an 860. Any value to us getting together and figuring out who sends test WUs to whom? Remember the problem with this is in the various account keys. Those go with the data. I'd be willing to send Emu mine, it's just a lot of data, about 130meg. I've come to trust Emu. Just don't let those WUs escape into the wild and bring down the planet's mainframes.... The issue here is that Emu's have more run time since I managed to corrupt my first ones....

One thing I mentioned before that may be of use is to refine jcool's formula for PPD estimating. I think we could do that with this test method, but we would need to decide on what we are looking for. We all know straight bclk makes a good difference. We have some indication that QPI makes a pretty good difference too, Uncore/NB makes some difference, and RAM timings may make a yet-to-be-determined difference. This is with HFCC WUs though....

What I will say is that I'm not willing to burn an infinite amount of CPU time here. I'm about producing valid, useful WUs. That's why I farm. On the other hand, I'm willing to maximize my production too. Where is the balance point is what I'm asking the team?

Bob

hmmm you are probably right Bob. Maybe we need to collect our thoughts a little more I just went off on a tangent while I waited for guidance over the weekend

I'm happy to keep testing if you want to get back to farming Bob . I see this as being a valuable lesson for us all as well, even so much in optimising our crunchers for maximum performance.

FYI my WUs are 239MB uncompressed and 68MB RAR'd

So guys as Bob asked what do you think would be a valuable set of tests. I think Bob should grab my WUs simply as I have more data points on HFCC but perhaps he should start back on his fav project HCC

I'll have more results shortly but am quite willing to be a sheep and move with the masses!

[123bob]

hmmm you are probably right Bob. Maybe we need to collect our thoughts a little more I just went off on a tangent while I waited for guidance over the weekend

I'm happy to keep testing if you want to get back to farming Bob . I see this as being a valuable lesson for us all as well, even so much in optimising our crunchers for maximum performance.

FYI my WUs are 239MB uncompressed and 68MB RAR'd

So guys as Bob asked what do you think would be a valuable set of tests. I think Bob should grab my WUs simply as I have more data points on HFCC but perhaps he should start back on his fav project HCC

I'll have more results shortly but am quite willing to be a sheep and move with the masses!

I think finishing off the tests at 4 gig might be useful? You were going to match parameters and see how it came out. Of interest are the Mem tests, apparently for HFCC at least. (Don't risk hardware though. I'm assuming you're cooling the power chips now?)

I could run your WUs, or get a batch of HCC if someone wants to see how those run. Either way.

If I don't hear otherwise from our teammates, I'll run the 3.4 gig point next, with matched parameters. I can't hit 4 gig on the 860, so I'm a bit stuck. Maybe I'll try again but I already gave it max vcore....Perhaps an X3440 rig is in order?

BTW, my new baseline is finished. I'll report up this morning, as soon as I have time to go through the calculations.

My best,
Bob

[123bob]

I did another baseline at 3.7 gig using the new set of WUs. The settings on each machine are matched to each other as before. Same BOINC versions and CPUid were used. (I gotta quit calling it CPUz...)

The 920 first...





Calculation comes out to 81:851:666. Further math gives 3.9736805 Hrs/WU.






Calculation comes out to 80:848:596. Further math gives 3.9291203 Hrs/WU.

So, the 860 was .0445602 hours, or 2.673612 minutes FASTER per WU than the 920. Interesting.

Look at the big WU I have. The 15th WU, #461 was quite the pig on both rigs. The 860 finished this one a full 23 minutes faster than the 920. I wonder what that's about....

In any case, I still call these rigs on nearly the same level. Using the new WUs might prove interesting, particularly on that big one.

I'm running the 3.4 gig point right now. I'll have power numbers to report on that point too.

Regards,
Bob

[Otis11]

Nice work. We'll have this down pat if this keeps up much longer...

Only thing we don't know (or I don't know) is what you can expect an 860 to clock to...

[loonym]

In a nutshell the 1366 wins by a small margin?

[123bob]

The other way around. The 860 wins by a little bit on this run. It's a couple of minutes faster at doing a WU.

Again, the difference probably is not that significant. Later, I'll re-run them at 3.7 gig just to see how consistent they are.

Bob

[123bob]

Todays update,
K so my impatience got the best of me. I do have room for a full atx board, but I ended up getting the P55M-UD4 from my local store. Main reason is that I cannot really wait till mid next week and get everything set back up. I am currently going through a major remodeling of all mt rigs due to the hardware I have to work with atm. You can see from all my stats that my production is down. I got my new 920 rig up last night but still have some things to do. lost train of thought; I am going on vaca to San Diego this coming weekend, so I wanted to get the X3440 back up.

Good news is it was only the mobo; So I will get it rma'd and resell when it arrives back.
I have the X3440 on the ud4@4GHZ@1.3xV. I am currently in the process of switching to a new HDD. This rig will be back up sometime today. I think I need to reseat the megahalem as my idle tems are a little higher than the last time.

Even though this rig now costs as much as a 920 setup; I still wanted to see for myself how it compares to the 920 in all respects.
I will let you know how it goes.

One of my 920 rigs is down. I have a Bloodrage and had to use the 775 heatsing mount; for some reason my idle temps are in the 60's. I have reseated 3 times with no luck. I am not sure what is going on. may have to run stock clocks till I can get another true or megahalem.

I think this covers it for atm. I will keep yopu posted with any updates.

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

[emuexport]

I think finishing off the tests at 4 gig might be useful? You were going to match parameters and see how it came out. Of interest are the Mem tests, apparently for HFCC at least. (Don't risk hardware though. I'm assuming you're cooling the power chips now?)

I could run your WUs, or get a batch of HCC if someone wants to see how those run. Either way.

If I don't hear otherwise from our teammates, I'll run the 3.4 gig point next, with matched parameters. I can't hit 4 gig on the 860, so I'm a bit stuck. Maybe I'll try again but I already gave it max vcore....Perhaps an X3440 rig is in order?

BTW, my new baseline is finished. I'll report up this morning, as soon as I have time to go through the calculations.

My best,
Bob

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".

[123bob]

****AN IMPORTANT TEST CONSIDERATION****

Remember above when I said I had an aborted WU show up in my test WUs? I had to get another batch of test WUs, or so I thought.

Emu PMed and reminded me that the deadline date on the WU probably had expired. IT SURE DID! That's why the WU showed up aborted. I spent about 5 hours fiddle-farting around, when all I had to do was trick BOINC and set the date back on the rig.....:

THANKS EMU!!

So, I could continue with the old WUs if I wished. However, I'm finding the new set to be interesting too, with that big fat 15th WU. I think I'll stick with them.

Now, some considerations for those doing this testing.

1. You MUST set the time back on the rig BEFORE you load and begin running the test WUs. Otherwise, it will abort expired WUs if you try to change the time after the run has begun.

2. After a test run, don't forget to set the time back to actual so when you resume crunching "real" WUs, they don't get screwed up.

Again, Thanks Emu! That was REALLY bugging me on Sunday....

Regards,
Bob

[123bob]

OK, here's a test run at 3.4 gig with the new WUs used in the last test, post #207.

The 920 Screenies





This comes out to 4.5320275 Hr/WU.



Now the 860





This comes out to 4.43528935 Hr/WU

So again, with my rigs the 860 came out FASTER than the 920 by 5.80429 minutes per WU. I don't know why.....This is an interesting set of WUs.


Now, to compare just the raw bclk changes from 3.7 to 3.4;

On the 860;
3.4 gig 4.4352894 Hr/WU
3.7 gig 3.9291203 Hr/WU

3.7 gig is 30.37 minutes per WU FASTER than 3.4 gig on the 860 rig......


On the 920;
3.4 gig 4.5320275 Hr/WU
3.7 gig 3.9736805 Hr/WU

3.7 gig is 33.50 minutes per WU FASTER than 3.4 gig on the 920

Again, the 860 and the 920 are reasonably close on the 3.4 gig point. Not as close as 3.7 though.

The 3.4 vs 3.7 gig analysis points out what we already knew. Clock counts. And it counts a lot.....

Loaded power draw at 3.4 gig;
860 rig - 173 watts Boinc loaded
920 rig - 207 watts Boinc loaded

So, the 860 is producing slightly faster, for 34 watts less power draw.

Regards,
Bob

[Wezly]

Very interesting, 123bob.

Makes me want to replace everything with P55's

Unfortunately I have only been idly checking this thread, and I am wondering what you take into consideration when choosing WU's for these test. Could you elaborate or tell me where in the thread I got lazy and should re-read?

[INFRNL]

Nice Results Bob

We all appreciate your hard work and dedication

[BlindFreddie]

Great work, guys

I was on the point of posting that when you re-run your saved WUs in the future, you have to set your system time/date back to the values when you saved the files. But you guys seem to be more expert than me, so I thought you'd know, so I killed my part-composed post.
Also don't forget to disable any auto-update (from Internet) of your system time if you are using one.

I thought that usually, for the same overclock, 860s need more Vcore than 920s, but 123bob's screenies show 920: 1.304V, 860: 1.264V. Individual chip variations I guess. 1.3V for a 920@3.7GHz is a lot higher than the numbers being described in the Now THAT is a nice CPU thread.

With your big WU (#15), bob, it might have been useful to have a look at its runtime memory usage compared to the others. Bigger memory footprint would make cache less effective and mem speed more important (not sure whether latency or bandwidth). You'd expect 920 to win out though.

perfmonitor.exe (from cpuid.com) can show cache success rate/mem access rate on each core (I assume on each thread of an HT machine). Running multiple instances of perfmonitor tracks successive cores. You need to install BOINC as an ordinary unprotected process so you can assign tasks to cores (right-click on task in Task Manager), so you can see which task is where.

HTH

[123bob]

Very interesting, 123bob.

Makes me want to replace everything with P55's

Unfortunately I have only been idly checking this thread, and I am wondering what you take into consideration when choosing WU's for these test. Could you elaborate or tell me where in the thread I got lazy and should re-read?

Emu and I are both running a randomly chosen set of HFCC WUs for testing. He is running a different set than I am.

Bob

[emuexport]

Well I have to admit I have been hoarding my results a little

But only because I have been too busy to post but dont worry I'm still testing in the background. getting these rigs to match up is a real challenge.

So we have 5 results for you lot today. Almost all of them have the same memory timings now except for the first one. I will be keeping this set of memory timings till further notice. Also I'm running a single stick Cos I know how much us crunchers like to stretch the coin!

Here are the screens for proof and analysis of results.



Total run time for 24 WUs 139:51:35
Avg Runtime per WU 5:49:39



Total run time for 24 WUs 133:13:36
Avg Runtime per WU 5:33:04



Total run time for 24 WUs 98:28:27
Avg Runtime per WU 4:06:11



Total run time for 24 WUs 98:13:29
Avg Runtime per WU 4:05:34



Total run time for 24 WUs 139:45:23
Avg Runtime per WU 5:49:23

So if you are lost dont fret I'm putting together a table to make it easier to follow.

I was really struggling with the i7 860 so it missed a run last night.

On another note my RAID6 Array finished rebuilding after two days I added two more drives and it took forever to move all those bits around.

So where are we at Bob?

[emuexport]

I've been slack and not recorded the power stats for all these configurations. So I'll need to get this if this is useful too.

Let the tables and charts begin.
Total crunching time to complete 24 WUs


Avg Crunch time per WU (of the 24)


The configuration of each setup.

[INFRNL]

damn EMU, you went all out. You & Bob are the sheet!

I will have to mess around; ever since my UD2 took a dump; I have been having some problems on my m-UD4

I have been to busy at work to have enough time or energy to play around. I am going on vaca this weekend, so I will see how it goes when I get back.

[jcool]

Awesome work guys, will have to study your results later when I have the time

[emuexport]

damn EMU, you went all out. You & Bob are the sheet!

I will have to mess around; ever since my UD2 took a dump; I have been having some problems on my m-UD4

I have been to busy at work to have enough time or energy to play around. I am going on vaca this weekend, so I will see how it goes when I get back.

No worries

It's important to remember that we are a team and at the end of the day if my sacrificed crunching time helps my fellow crunchers whatever the team then its win win in my books

Whenever you get time I'd love to see what that X3440 is capable of!

Awesome work guys, will have to study your results later when I have the time

I'll update my post above with the results as I go (while it remains on this page at least). Maybe I should add Bobs results to this as well? And get off my arse and take the power readings too. Might just have to get a 2nd power meter at $30 I'm sure I can offload it to someone else at some point or being a reliability engineer I'll keep it as my standby (backup) No seriously thats a good idea!



Total run time for 24 WUs 98:13:29
Avg Runtime per WU 4:05:36



Total run time for 24 WUs 96:26:14
Avg Runtime per WU 4:01:06

[Otis11]

Emu, you are aware that those are two different clock speeds right? Just checking!

And thanks for all the work you guys are doing!

[emuexport]

Emu, you are aware that those are two different clock speeds right? Just checking!

And thanks for all the work you guys are doing!

Yes mate.

Just run outta time at nights to get comparable clocks. (If you notice the last two had the same BLCK but different multis its because I was tired and forgot I needed to put turbo and EIST on on the 920 thats all.

I am trying to capture all of the runs I complete so we can compare performance. 4GHz is my target but from the results it appears that 4.1GHz 860 is simialr to 4GHz 920