Official Intel Gulftown (Westmere) Overclocking and Batch Tracking Thread

[crashmax]

Wow. Great Chip you have!!! Have the same Dominator GTX2 ordered.

Dou you use a Clssified 760 Mobo?

[MetalRacer]

Batch 3003B287.

[lawtq]

Exact same volts, batch and OC I have, great isnt it?! Especiall for 4.5ghz HT!!!!!

[mk-ultra]

Can we try to avoid turning another thread into a useless and rehashed "how stable is stable?" discussion?

i agree, but first post ask for prime or linx so idk
here's some action without using any retarted unneeded stressing tool, since the last few posts seem to be that way


Batch:3001B331
Watercooling testing

[Rubycon]

There should be a new thread altogether on stress testing these new chips. They are completely different than previous versions.

Also a 1024MB vprime or Vantage score does nothing to show that the o/c is stable enough to use for a real world use. Those 5GHz+ shots are fine but if the system cannot be used (after all if you have six cores you bought them for true MP use not gaming or surfing! ) in the real world.

I'm finding clocks that I THOUGHT were stable are hardly stable and bringing the speeds way down. With six cores a lot of the software (especially with HT ON) cannot properly stress these cores out like previous chips.

That said none of my chips are close to what the B batches are doing but then again my testing methodology is apparently much stricter as well. Kind of throwing the baby out with the bath water here.

[gosmeyer]

^^+1^^

[Sam_oslo]

Nobody is kicking on bad RAM and Uncore performance on these chips?

These chips have less read/write and higher latencies on RAM (compared to 45nm chips), and the lack of performance gets even more noticeable on Uncore.
These CPUs seams to need a high Uncore to perform at their peak, but the 12MB cache is making it really difficult to OC the Uncore. Anybody has found workarounds or ideas around this?

[Movieman]

There should be a new thread altogether on stress testing these new chips. They are completely different than previous versions.

Also a 1024MB vprime or Vantage score does nothing to show that the o/c is stable enough to use for a real world use. Those 5GHz+ shots are fine but if the system cannot be used (after all if you have six cores you bought them for true MP use not gaming or surfing! ) in the real world.

I'm finding clocks that I THOUGHT were stable are hardly stable and bringing the speeds way down. With six cores a lot of the software (especially with HT ON) cannot properly stress these cores out like previous chips.

That said none of my chips are close to what the B batches are doing but then again my testing methodology is apparently much stricter as well. Kind of throwing the baby out with the bath water here.

Some thoughts and I defer in the knowledge dept to many of you.
I've been running first A step 2530's then B step X5680's since last fall.
First on a SM X8DA3 board so no overclocking possible.
Never an issue, 10C cooler than the quad Gainestowns at the same speeds and app 50% more on the work end so scaling was darned close to perfect.
Then I got this EVGA board and had the ability to really see what the X5680's would do.
Excellent as long as you understand the limitations.
For normal usage( read that as air cooling) as long as you stay under 175 BCLK with a max of 1.4vcore,1.35VTT your fine.
Go over that and this is where you will start to kill these cpu's.
Not today, not in a week but you will.
Decide what you want. Long term excellent power or short term HUGE power?
To me this is too expensive a hobby not to use your smarts and work within what we are finding to be the known parameters of the chips.
I know everyone wants that magic 4GHz number for daily usage but the reality is unless your willing to buy the absolute top chip that 4GHz isn't there.
My X5680's have a 25 multi and for 24/7 100% load work I'm at 4152mhz..25x166..1.375vcore and 1.3VTT.
Yes, I can go higher and have but not for long term use.
Remember the differences between the 65nm Kentsfields and the 45nm Yorkies?
How many people killed Yorkies with high VTT added in with high vcore?
History is repeating itself with these.
Now before you all put my comments down to a old fart whose too conservative think again. ALL I care about is not seeing you guys lose a ton of money that you've invested.

Thanks for reading.

[Sam_oslo]

^^ You have several good points there.

Benching is always fun and good to push it to the edge, but it is not a good Idea to push the system to the limits for 24/7 use, specially for heavy duty systems, because you get some extra power-usage and heat overhead to deal with.

These 6C/12T CPUs are monsters even at moderate OC, and it is always a good idea to play safe until we get to know these chips, and what kind of torture they can take for 24/7.

I've been doing some overnight 3D-rendering on my X5650 @3.8Ghz 1.21v VVT=1.32v, and it is much faster than my good old w3250 @4.2GHz 1.20v. And now, the Load-temp is almost the same as the Idle-temp as before
These X56xx chips will shine in a double socket MB, and I'm waiting for EVGA RS2 before I seriously try to kill it.

[systemviper]

I've been doing some overnight 3D-rendering on my X5650 @3.8Ghz 1.21v VVT=1.32v, and it is much faster than my good old w3250 @4.2GHz 1.20v.

What would "it is much faster " mean, is it 10min faster an hour, takes half the time, i am just looking for some real world measures and it looks like your going down that path.

thanks in advance for the info

[Sam_oslo]

What would "it is much faster " mean, is it 10min faster an hour, takes half the time, i am just looking for some real world measures and it looks like your going down that path.

thanks in advance for the info

I will build a new system with my good old w3520 soon and then i can compare directly, but for now I can't compare them directly for real 3D-render jobs. I have to guess, my overnight 3D-renders seams to be running ~30% faster with my current "moderate" OC.
For now I can compare the CineBENCH 11.5 on the same MB and same RAM, X5650 @3.8GHz, 9.98 pts vs w3520 @4.5GHz, 7.7 pts.

[Rubycon]

Nobody is kicking on bad RAM and Uncore performance on these chips?

These chips have less read/write and higher latencies on RAM (compared to 45nm chips), and the lack of performance gets even more noticeable on Uncore.
These CPUs seams to need a high Uncore to perform at their peak, but the 12MB cache is making it really difficult to OC the Uncore. Anybody has found workarounds or ideas around this?

If you sacrifice BCLK speed you can run your uncore higher for lower latency and better memory throughput. Remember there is a limit in throughput whether its achieved through MHz or timings. Pick your poison and play as hard as you like. As long as you don't go outrageous on voltages you should be OK.

Another thing - voltages! NEVER take software readings for real. Fortunately on the Classified boards the e-leet software seems to show considerably HIGHER than actual (metered) results for VCORE and VTT - both which are crucial if you run at the limits.

As someone mentioned before we should have a dedicated GT thread and just leave this discussion to batch overclock results.

[Bodar]

Has anyone heard anything about this batch number???? cant find it anywhere? Thanks!
3005f671

[poke349]

Speaking on LinX not being able to use 6/12...
If there isn't a single stress test that can test everything... Maybe run multiple of them?


Say prime or LinX along with something more uncore/memory intensive such as HyperPi or MemTest of some sort.

[zalbard]

Speaking on LinX not being able to use 6/12...
If there isn't a single stress test that can test everything... Maybe run multiple of them?


Say prime or LinX along with something more uncore/memory intensive such as HyperPi or MemTest of some sort.

Yeah, since LinX only works well with 512MB... Could maybe try a few LinX instances at the same time.

[poke349]

Yeah, since LinX only works well with 512MB... Could maybe try a few LinX instances at the same time.

That's actually not a good idea.
As far as I can tell, Linpack is very carefully tuned for the size of your cache so running multiple instances is just going to thrash the same resources.
Result: less intensive load due to cache contention.

As a side note: Running max memory isn't exactly fair.
The more memory you use the longer each iteration is. (I tried running Linpack at max memory on my 64GB workstation once... half an hour later, still hadn't finished a single pass... )

[mbreslin]

3003b396: seems to need more voltage than other batches to be stable. :/

[The Water Dog]

Mine is 3003B396, and is pretty happy with it so far.
Havent tested much yet but i think its not bad. Just have to treat it like your woman

[zalbard]

Just to bump this thread... If you see a post with some valid info (that can be used in the post #2) somewhere, feel free to PM me or link it here, and I will add it to the list.
Thank you for your cooperation.

Oh, and updated!

[marcolecorse]

Hello

My 980X 3003B178

Test Wprime 32&1024 H2O

[Toolius]

i just got mine too
3005F584 batch..
any word on this batch ? seems the 3005's are the new batches out ..
cheers guys !

[Sam_oslo]

i just got mine too
3005F584 batch..
any word on this batch ? seems the 3005's are the new batches out ..
cheers guys !

Yeah 3005 is new. It is interesting to see if it can still keep up with earlier batch#s.
Keep us informed about the OC-headroom.

[needforspeed]

Yeah 3005 is new. It is interesting to see if it can still keep up with earlier batch#s.
Keep us informed about the OC-headroom.

I cound not resist and got a 3005f at Frys at the weekend.. Was not impressed.

Struggled to get 4500 with HT off 1.47v +/-.. one core was 10-12 deg Hotter than the rest.. I was up in to high 80s... wasnot a keeper

.. I promptly pulled it put and got a refund.. I hope others have better luck. it was my first gulftown experience and not too impressed. probably wont return

[Nostromo]

3005F are old I got a 3005F from fry's on day of release.

It does 4.5 HT on at 1.458 but a few of my cores read way low but when loaded seem to equalize with others.
Also 3916 uncore, memory 2100 8-8-8 timings vtt +150 (1.35 on classy) bclk 150*30
I am running it currently on classified 760..
I also have a second 3005F and 3003B still in boxes yer to try, also purchased within one week of 980X release.

[Sam_oslo]

^^ That 3003B is probably the best one, try it.

It doesn't matter when you've bought them, the 3005's are produced 2 weeks after 3003's, and it doesn't need to take more than 2 weeks actually, but these may sill be good too.
Sometimes it may wary among the same batch# too, but generally it is interesting to see how long it takes before the good batch#s disappear, and 2 weeks is a long time.