Core i7 Overclocking by Batch #

[roller11]

Edited my other post with updated 4.0G info. Prime blend 11 hours successful at 1.3125v. QPIv at 1.225. DRAM at 1.6v. Everything else on Auto. Going to try 1.30v later today.

http://www.xtremesystems.org/forums/...postcount=2464

-td

You have a truly outstanding chip considering your results were at 89-93C. Try disabling HT and see what it'll do, could be a record ! And be sure to increase Qtt up to about whatever Vcore you are using.

[buddatech]

my results so they may be added to the list eVGA X58 i7 920 @3.9 186*20+Turbo Batch# 3937B367 with 1.3625v HT ON
Zalman 9700 HSF temps are with Lin X and P95 running simultaneously.

[Loser777]

It's kind of counter-intuitive, but having both P95 and LinX running at the same time won't produce the same amount of stress/heat that LinX would alone, as the cores are constantly switching between tasks... P95 isn't as stressful. It's like having background processes running during stress testing... the less stressful process will cause the temperatures to go down as the CPU is spending time running a less stressful process.

[INFRNL]

does anyone else besides Nemesis have a B317? Thats what I have and looks like it might be a little weaker than the B194. I havent gotten my system up yet as I am waiting on ram...be here tomorrow! I am hoping to get at least 4ghz on decent voltage like some of the other chips

[vikingboy]

It's kind of counter-intuitive, but having both P95 and LinX running at the same time won't produce the same amount of stress/heat that LinX would alone, as the cores are constantly switching between tasks... P95 isn't as stressful. It's like having background processes running during stress testing... the less stressful process will cause the temperatures to go down as the CPU is spending time running a less stressful process.

This isnt my experience although the logic isnt flawed. However, I can run 20 loops of linx 10000 problem size with out issue but as soon as I add a blend run of P95 at the same time instability is shown.
I think the weakest part of the system is the IMC and a small LINX run doesnt show ups the IMC weaknesses that well.
Blend does stress the IMC and with the small Linx run stressing the CPU - shows those weaknesses very quickly.
The other way to test for stability, is to run a larger Linx problem, 17000 minimum. Most of the failures on Linx are ebtween cycles of calculations (watch in task manager), you can see it isnt temp that killing the system, its the reading and writing matrix data from memory for the next calculation run.

Those large linx calcs really hurt and my overclocks using that as a bench for stability are way less than any other solution - it depends what you want at the end of the day.....high numbers or stability.

Ian

Ps I can complete a 500 cycle run of Linx 10000 problem size which wont complete 2 runs at 17000...its that much of a delta for me.

Would love to see temps of a 20 run of Linx @ 17000 - if our system can handle it.

[JWMc]

Edited my other post with updated 4.0G info. Prime blend 11 hours successful at 1.3125v. QPIv at 1.225. DRAM at 1.6v. Everything else on Auto. Going to try 1.30v later today.

http://www.xtremesystems.org/forums/...postcount=2464

-td

Very nice, I did try 1.3125v load @ 3.990ghz (190x21), 6GB, HT enabled, guess what!:

http://img218.imageshack.us/my.php?i...12vht6grf0.jpg

Yep, same result pretty much. I didn't do long term Prime blend testing yet but the vCore is verified for sure.

I love this chip I got!!!!

[JWMc]

This isnt my experience although the logic isnt flawed. However, I can run 20 loops of linx 10000 problem size with out issue but as soon as I add a blend run of P95 at the same time instability is shown.
I think the weakest part of the system is the IMC and a small LINX run doesnt show ups the IMC weaknesses that well.
Blend does stress the IMC and with the small Linx run stressing the CPU - shows those weaknesses very quickly.
The other way to test for stability, is to run a larger Linx problem, 17000 minimum. Most of the failures on Linx are ebtween cycles of calculations (watch in task manager), you can see it isnt temp that killing the system, its the reading and writing matrix data from memory for the next calculation run.

Those large linx calcs really hurt and my overclocks using that as a bench for stability are way less than any other solution - it depends what you want at the end of the day.....high numbers or stability.

Ian

Ps I can complete a 500 cycle run of Linx 10000 problem size which wont complete 2 runs at 17000...its that much of a delta for me.

Would love to see temps of a 20 run of Linx @ 17000 - if our system can handle it.

Vikingboy, you are messing up your vCore testing by simultaneously running Prime with LinX or something else to stress the CPU as Loser777 said. I know it takes forever but you really should use only LinX or OCCT to find vCore and then use Prime blend to make sure you have enough VTT/vDIMM/relaxed enough timings.

Yea I noticed that about LinX. You will fool yourself if you use a low problem size. I don't know why LinX needs to use as much RAM as possible relative to the total to load the system properly but it does. It's confusing because we know that LinX itself does not stress the memory system as much as Prime blend.

I agree though, you need Prime blend to test IMC quality which is also a goal of this thread (I hope).

[icecold316]

Very nice, I did try 1.3125v load @ 3.990ghz (190x21), 6GB, HT enabled, guess what!:

http://img218.imageshack.us/my.php?i...12vht6grf0.jpg

Yep, same result pretty much. I didn't do long term Prime blend testing yet but the vCore is verified for sure.

I love this chip I got!!!!

Fantastic JWMc, I was able to match same setting of yours at 3.8 Ghz @ 1.2V after I moved from BIOS F4r to F5e. That BIOS helped me drop voltages by 0.05V from 1.248V @ Load to 1.2V @ Load. I am pretty sure, my processor could match you voltages on 3.99 Ghz too. Seems 3841A is a very good one.

Outside the case, my i7 went above 70C @ 3.8 Ghz, but inside the case its running at 74C in Linx with room temperature at 26C. I can't imagine the heat at 3.99 Ghz when it will be inside the case at 26C Room temperature( probably in 90's).

Any temperatures above 80C, seems a bit high - as my PC turns into a heater thanks to CPU, Memory, Gtx260 and NB all contributing to the heat.

I was getting "IRQL not less or equal" error with Green Lan Feature enabled. I disabled that feature and the problem is Gone.

[JWMc]

Fantastic JWMc, I was able to match same setting of yours at 3.8 Ghz @ 1.2V after I moved from BIOS F4r to F5e. That BIOS helped me drop voltages by 0.05V from 1.248V @ Load to 1.2V @ Load. I am pretty sure, my processor could match you voltages on 3.99 Ghz too. Seems 3841A is a very good one.

I'm sure it could, our chips seem like they might be almost identical. Are there many 3839s that are known to OC poorly? I've seen people say their 3841s didn't do well also.

About lots of chips hitting 4ghz on the Xtremesystems forums. I bet that's because we know how to test things properly and present/compare data honestly. I bet just about any i7 can hit 4ghz on air but many of us use water and many noobs on air aren't altering the test environment properly to see what it can do (super cold ambients).

Outside the case, my i7 went above 70C @ 3.8 Ghz, but inside the case its running at 74C in Linx with room temperature at 26C. I can't imagine the heat at 3.99 Ghz when it will be inside the case at 26C Room temperature( probably in 90's).

Any temperatures above 80C, seems a bit high - as my PC turns into a heater thanks to CPU, Memory, Gtx260 and NB all contributing to the heat.

My in-case temps at normal ambient (25C) went WAY up compared to my early testing in the cold. But I have a bad situation where I only have one exit fan for the heat. I'm using a Sanyo Denki 101 for the single exhaust fan though so I can crank it up if I want to. Entirely new chassis and WC plans are in the works.

I was willing to push the CPU to 8-9c from TJMAX on air, no more. This was also only just to see what the chip could do so I'd have more information to decide on building my first WC loop or not. I'm going for low 60s in Real Temp @ 25C ambient for my max 24/7 OC on water, whatever that turns out to be.

I was getting "IRQL not less or equal" error with Green Lan Feature enabled. I disabled that feature and the problem is Gone.

I think I may have glossed over this in the Gigabyte X58 board thread. If this helps stability it could be important that everyone learns this!

[286SX]

Vikingboy, you are messing up your vCore testing by simultaneously running Prime with LinX or something else to stress the CPU as Loser777 said. I know it takes forever but you really should use only LinX or OCCT to find vCore and then use Prime blend to make sure you have enough VTT/vDIMM/relaxed enough timings.

[...]
I agree though, you need Prime blend to test IMC quality which is also a goal of this thread (I hope).

JW,

Can you clarify something for me as I dont' have much experience with the newest chipsets and CPUs WRT overclocking.

How should I be using LINX and Prime to confirm the best system settings. Honestly, I completely skip LINX and just try to run blend for 6+ hours.

As I'm working on getting a low(ish) voltage 4.1/4.2 working, is the "best" method to set CPUv first, run LINX to see if it sticks, then run prime blend to see if it's stable.

Questions: if LINX dies, I should up the CPUv first? Will changing PLL or QPIv help the LINX stress?

If LINX passes, but prime dies, does this mean PLL or QPI should be changed first? And what about those crazy skew settings? Will they really help...

Thanks!

-td [BTW, nice ride. You should try the track instead of autocross!]

[JWMc]

JW,

Can you clarify something for me as I dont' have much experience with the newest chipsets and CPUs WRT overclocking.

How should I be using LINX and Prime to confirm the best system settings. Honestly, I completely skip LINX and just try to run blend for 6+ hours.

As I'm working on getting a low(ish) voltage 4.1/4.2 working, is the "best" method to set CPUv first, run LINX to see if it sticks, then run prime blend to see if it's stable.

Questions: if LINX dies, I should up the CPUv first? Will changing PLL or QPIv help the LINX stress?

If LINX passes, but prime dies, does this mean PLL or QPI should be changed first? And what about those crazy skew settings? Will they really help...

Thanks!

I have no clue when to alter PLL voltages or by how much. LinX stability needs (sometimes a fair bit) less QPI/VTT than Prime blend for sure!

AFAIK, LinX is the best program out there to find your true multithreaded stable vCore (or manually running Linpack of course). If it crashes in LinX but not Prime blend (I'd be surprised), you definitely need to up the load vCore until it goes up to the next closest step above what you were using, at least. LinX is just great because if you're not even close on vCore it will crash quickly and consistently so. I've only once been able to find settings that could get LinX to report an error and not BSOD me when the system fails.

Once you get it to pass LinX, vCore should be verified. However even LinX can pass with extremely borderline low vCore and then Blend will fail because you needed that one more step up on the load voltage as reported by CPU-Z. BTW I'm up to almost 4 hours on Prime blend with all the same settings, hell yes!

I have a guess that the strange behavior I was seeing when I needed one voltage bump more for Prime blend vs LinX was at 20x200. The system is MUCH happier at 190x21!

-td [BTW, nice ride. You should try the track instead of autocross!]

I love her. I have gone to SCCA Solo2 Nationals once and won some (pretty meaningless but still) divisionals. I've been taking a break from it for a while though. I did do a track day once but I wasn't willing to really push it in the corners that hard. It was at Roebling Road Raceway in Georgia. At the moment I'm riding on a hardcore autocross suspension setup: KW-V3 double adjustables with piggyback reservoirs in the front, 12k/10k F-R rates. PM me if you wanna geek out about it some more. I promise I won't clutter the thread up with car talk.

[The Nemesis]

does anyone else besides Nemesis have a B317? Thats what I have and looks like it might be a little weaker than the B194. I havent gotten my system up yet as I am waiting on ram...be here tomorrow! I am hoping to get at least 4ghz on decent voltage like some of the other chips

Yes, the B317 seems a bit weaker than the B194 but it did hit 4GHz prime stable @1.4v and 1.31vtt so its still a solid chip and better than some batches. Highest I've had is was 4827MHz on SS phase 1.6vore & 1.6vtt/qpi. There are others here with the same batch who've pm'd me so I'm sure they will post results. It's a good batch, just not golden.

[buddatech]

It's kind of counter-intuitive, but having both P95 and LinX running at the same time won't produce the same amount of stress/heat that LinX would alone, as the cores are constantly switching between tasks... P95 isn't as stressful. It's like having background processes running during stress testing... the less stressful process will cause the temperatures to go down as the CPU is spending time running a less stressful process.

The other way to test for stability, is to run a larger Linx problem, 17000 minimum. Most of the failures on Linx are ebtween cycles of calculations (watch in task manager), you can see it isnt temp that killing the system, its the reading and writing matrix data from memory for the next calculation run.

Those large linx calcs really hurt and my overclocks using that as a bench for stability are way less than any other solution - it depends what you want at the end of the day.....high numbers or stability.

Ian

Ps I can complete a 500 cycle run of Linx 10000 problem size which wont complete 2 runs at 17000...its that much of a delta for me.

Would love to see temps of a 20 run of Linx @ 17000 - if our system can handle it.

I retested with Lin X alone I've done a 20 pass test before with problem size at 17000 just never took a screen shot, also I am running p95 blend will post a pic once I get 2-3 hours minimum under test.

[headala]

Could you (cky2k6?) post the raw data spreadsheet? It's kindof hard to make any conclusions without being able to sort and filter the data. I'll be buying my 920 soon....

[roller11]

I have no clue when to alter PLL voltages or by how much. LinX stability needs (sometimes a fair bit) less QPI/VTT than Prime blend for sure!

:

JWMc,
Your results speak for themselves...Why don't you disable HT and see what your chip is capable of?

[JWMc]

JWMc,
Your results speak for themselves...Why don't you disable HT and see what your chip is capable of?

I may as time goes by waiting for WC parts to arrive and also learning how the hell it all works when it does. Once it's under water I'm definitely going to play with it until I've tested every limit!

[icecold316]

How about the Google Spreadsheet Data sheet for Sharing everyone I7 Overclock : http://spreadsheets.google.com/ccc?k...fnE-v78I7tODQg

[headala]

How about the Google Spreadsheet Data sheet for Sharing everyone I7 Overclock : http://spreadsheets.google.com/ccc?k...dzXcJAOw&hl=en

THAT is a brilliant idea! Hmmm, I shoulda thought of it...it is kindof my job.

[PytonOrm]

I know that a 10 minute run is not up to your standarts, but ambients here are near 35°C and I wont stress it with Prime for longer today.
Also, this machine have been up and crunching for 36h with this settings.

i7 920 - 3837B017

same batch as me does this at 1,64vcore http://valid.canardpc.com/show_oc.php?id=469425

[JWMc]

same batch as me does this at 1,64vcore http://valid.canardpc.com/show_oc.php?id=469425

Holy cow! I fear I'm not man enough to step up and try voltages that high on air. I will however be doing it on water.

[Astratuner]

Yes, very good news indeed! I too have a 3841A chip, and it runs really nicely. Stock voltage and cooling takes me up to 3.6Ghz without a sweat. But it does seem this batch is hit or miss, some like me report decent results, while others get not so decent results. In fact, my chip was only 123 chips in front of yours on the assembly line from this same batch, a 3841A505

Wow that's good news indeed.

Hope mine does well to

What do you mean with the good and bad results? Different programs or different users wit the same CPU batch?

[Viss]

Only half hour for now, more to come. 3835A804 under TRUE:

[icecold316]

Very nice, I did try 1.3125v load @ 3.990ghz (190x21), 6GB, HT enabled, guess what!:

http://img218.imageshack.us/my.php?i...12vht6grf0.jpg

Yep, same result pretty much. I didn't do long term Prime blend testing yet but the vCore is verified for sure.

I love this chip I got!!!!

I just tested with exact same setting of yours. Did 10 runs of 17000. It passed.

Only thing is the temperature was 87/85/86/79 after the run. On a completely closed Case (I shud have kept one side open) Room ambient was 25C.

[anzial]

How about the Google Spreadsheet Data sheet for Sharing everyone I7 Overclock : http://spreadsheets.google.com/ccc?k...dzXcJAOw&hl=en

Any way to paste the info into it? I've got excel spreadsheet with with the info from this thread + some updates.

I've been experimenting with my 3836B043 under water. 3.8ghz (21x180) seems to be stable @ 1.22vcore (4 hours of prim95). I'm using Biostar mobo with default bios, should prolly try new beta bioses and see how it goes since 4ghz and above require much more tweaking than leaving everything except vcore and vdimm at auto/default setting lol

[norim_13]

I was not waiting for OCs like these on Corei7 CPUs! These are great results!

Congratulations!