DFI LANparty DK P45-T2RS/T3RS (Intel P45) - News|Info|Results

[tamashumi]

Some ppl get up to 600 with non-plus. But this is kind of challange ;]
If you squize of it 500 it will be really nice score. Check N-C's posts with settings in this thread.


edit: guys I'm just testing stability with 1.1950V (@bios, SG shows 1.16V!) for NB, FSB still 450MHz.
It's to early to say for sure, but it seems to be stable

edit2: stable after FastTestProcedure @ 1.1825V (@bios, SG 1.15V). I think it's quite good, huh? Check out the attachment

[dark22]

Don't know about quad cores but I recommend 08.12 bios, latest non-beta.
I had no difference between 08.12 and 10.03 (latest beta) on non-plus board. Which cannot be said about plus version (for which I would recommend 10.03).

Can you please give the download link?

[tamashumi]

Well it's as usual on producer site, but why not. Here you go: http://us.dfi.com.tw/Upload/BIOS/D245D812.zip

BTW. I'm posting under 8x500, not sure if it's stable yet

[DemonEyez]

@DemonEyez
I like "rock" stable myself. OCing for just give a try single run of SuperPi or 3dMark hoping that maybe it won't crash till the end of benchmarking and later spanking a monkey over the result really sux
For now I have many different settings to check, so tests have to be shorter. When I reach desired OC then I run some priming over a night (or day when I'm out of home).
Btw. you're welocme to try my test procedure for fast tests (f.e. when new board comes). SuperPi 32M up to the result while Priming with large FFTs single thread. It's heavy demanding, not much can affect stability if this one pass.
Which Prime preset do you use for testing?

.

For a quick impression if thing are stable I use SuperPi 32MB run, if it doesn't crash/error on me I use Linpack or the Intel Burn program and run it for 10 runs. That only takes 10 minutes to complete and some say that's it's equal to 24 hours Prime because of the much, very much heavier load. If Linpack doesn't error on me I run prime or Orthos for 8 hours

I run large FFTs with Prime and Orthos

[P_1]

For a quick impression if thing are stable I use SuperPi 32MB run, if it doesn't crash/error on me I use Linpack or the Intel Burn program and run it for 10 runs. That only takes 10 minutes to complete and some say that's it's equal to 24 hours Prime because of the much, very much heavier load. If Linpack doesn't error on me I run prime or Orthos for 8 hours

I run large FFTs with Prime and Orthos

That isn't always true, there have been many times where linpack would pass flawlessly for hours but occt fails within minutes. Here is what I suggest, start with prime to see if your problem is a gtl problem, usually prime errors quickly, tweak gtl a bit until it doesnt error. If tweaking gtl doesnt help the prime errors then start adding vcore and repeat with the gtl tweaking, continue to do this until prime doesnt error. Also just to note, only run prime on blend until it says Test 1 pass, Test 2 pass, and then move on, at this stage there isn't much reason to test longer. Once that is done, start using LinX to test, if LinX freezes or your computer reboots, it's still a gtl problem. Meaning even though the gtl you had was passing prime for a bit it might just be at the cusp of stability. So start tweaking the gtl again, once LinX doesn't freeze or error, start testing with occt. If occt errors then raise vcore, and test again, rinse and repeat till occt can pass upto atleast 10 hours on mix. That is my routine for finding stable oc's relatively quickly. Also be sure to run your ram at settings you definitely KNOW are stable, so that it doesn't factor into any crashing or errors you may encounter.

[DemonEyez]

That isn't always true, there have been many times where linpack would pass flawlessly for hours but occt fails within minutes. Here is what I suggest, start with prime to see if your problem is a gtl problem, usually prime errors quickly, tweak gtl a bit until it doesnt error. If tweaking gtl doesnt help the prime errors then start adding vcore and repeat with the gtl tweaking, continue to do this until prime doesnt error. Also just to note, only run prime on blend until it says Test 1 pass, Test 2 pass, and then move on, at this stage there isn't much reason to test longer. Once that is done, start using LinX to test, if LinX freezes or your computer reboots, it's still a gtl problem. Meaning even though the gtl you had was passing prime for a bit it might just be at the cusp of stability. So start tweaking the gtl again, once LinX doesn't freeze or error, start testing with occt. If occt errors then raise vcore, and test again, rinse and repeat till occt can pass upto atleast 10 hours on mix. That is my routine for finding stable oc's relatively quickly. Also be sure to run your ram at settings you definitely KNOW are stable, so that it doesn't factor into any crashing or errors you may encounter.

I never said that using Linpack guarantees stable hous running Othos. i said some people say, not me. That's why I run Orthos/Prime afterwards.

[has33]

Guys, as I posted before I got a weird board. Its the PLUS version with Rev.A2 mentioned in the corner

For some reason, it refuses to work with any ATI card I tried with 3 different ATI cards (Palit HD4870-512MB GDDR5, GeCube HD3870X2 and a HD 3650)...same issue with all the 3 cards. Diagnostic LED show no issues...posts fine throughout till it reaches FF only theres no display n there are these 3 beeps (1medium n 2short like beep bip bip) @ 26.

Works flawlessly with Nvidia cards... tried with 8800GT, 7150 & 9600GT

I tried both the latest bios (8.12 & 10.3)...no difference.

Don't know what 2 do...any body knows of any DFI forums I can post my problem ??? And how do I go abt the RMA if I ve to ??? Got the board through a cousin in Canada (NCIX) and I'm here in India.

[Tal Bar]

Try to remove battery for 2 min. and resetting all in BIOS, maybe it's help... It's known issue with 4870x2, but not with any other cards..

[DemonEyez]

Guys, as I posted before I got a weird board. Its the PLUS version with Rev.A2 mentioned in the corner

For some reason, it refuses to work with any ATI card I tried with 3 different ATI cards (Palit HD4870-512MB GDDR5, GeCube HD3870X2 and a HD 3650)...same issue with all the 3 cards. Diagnostic LED show no issues...posts fine throughout till it reaches FF only theres no display n there are these 3 beeps (1medium n 2short like beep bip bip) @ 26.

Works flawlessly with Nvidia cards... tried with 8800GT, 7150 & 9600GT

I tried both the latest bios (8.12 & 10.3)...no difference.

Don't know what 2 do...any body knows of any DFI forums I can post my problem ??? And how do I go abt the RMA if I ve to ??? Got the board through a cousin in Canada (NCIX) and I'm here in India.

Did you plug in that Floppy connector plug near the PCIe 16x slot?

[Viss]

Strange. 4870X2 is fine here without floppy conn. Tried second slot to see if that works?

[C-N]

For some reason, it refuses to work with any ATI card I tried with 3 different ATI cards (Palit HD4870-512MB GDDR5, GeCube HD3870X2 and a HD 3650)...same issue with all the 3 cards. Diagnostic LED show no issues...posts fine throughout till it reaches FF only theres no display n there are these 3 beeps (1medium n 2short like beep bip bip) @ 26.

Reference 3870 X2 & 4870 X2 working here.

http://www.bioscentral.com/beepcodes/awardbeep.htm

http://www.bioscentral.com/postcodes/awardbios.htm

[Revv23]

That isn't always true, there have been many times where linpack would pass flawlessly for hours but occt fails within minutes. Here is what I suggest, start with prime to see if your problem is a gtl problem, usually prime errors quickly, tweak gtl a bit until it doesnt error. If tweaking gtl doesnt help the prime errors then start adding vcore and repeat with the gtl tweaking, continue to do this until prime doesnt error. Also just to note, only run prime on blend until it says Test 1 pass, Test 2 pass, and then move on, at this stage there isn't much reason to test longer. Once that is done, start using LinX to test, if LinX freezes or your computer reboots, it's still a gtl problem. Meaning even though the gtl you had was passing prime for a bit it might just be at the cusp of stability. So start tweaking the gtl again, once LinX doesn't freeze or error, start testing with occt. If occt errors then raise vcore, and test again, rinse and repeat till occt can pass upto atleast 10 hours on mix. That is my routine for finding stable oc's relatively quickly. Also be sure to run your ram at settings you definitely KNOW are stable, so that it doesn't factor into any crashing or errors you may encounter.

Lol that is the fast way? what is the slow way?

[Ozzfest05]

So ordering mobo on friday rampage extreme but im aslo getting a p45 now should I get Biostar Tpower I45 or DFI DK T2RS+ the biostar is a bit cheaper what would you guys recommend this board will be used in DICE testing maybe some LN2 for ram and cpu overclocks.

[P_1]

Lol that is the fast way? what is the slow way?

It's the fastest way for a concrete stable oc... If you are just looking for suicide screens then look elsewhere.

The main point here is to target the most finicky values first(gtl) and work from there, because gtl has the highest influence on stability at high fsbs. So it is the quickest way to "fail" your oc if its unstable. If one were to just run prime for 12 hours for it to fail instead of doing the above, you would just be wasting your time. I achieved rock solid 470mhz fsb stability on my q6600 in around 2 days with the second day being mainly extra stability testing via occt.

If you have an even faster way of achieving rock solid stability please share with us.

[dz`]

The main point here is to target the most finicky values first(gtl) and work from there, because gtl has the highest influence on stability at high fsbs. So it is the quickest way to "fail" your oc if its unstable.

I'm going very slow with my board staring from square one...

With e8400 when I get to around 3600-3700MHz using 9 multi I lose stability. (so 395-415MHz FSB)
I've been using IBT with max memory for 5 loops as my general idea of "stability".
So far the only adjustments to regain stability in IBT is to lower NB GTL from 67x to 63x and add a bit of CPU VCore. I went from auto to +100mv probably didnt need that much of a jump from 3500 to 3800 but I'm trying to figure out what this board likes with regards to all these other voltages.
As I type this I've just passed a 32M Pi and the standard IBT test I've been using at 422*9=3800MHz.

CPU VID +100mv (1.17v)
SB/PLL 1.55v
NB Core 1.1825v (minimum selectable)
VTT 1.10v (minimum selectable)
VCore Droop Disabled
GTL 67x/67x/63x
1:1 FSB:RAM
10/03 Beta BIOS

End of Intel n00b post

[Eldonko]

What are you talking about with the floppy connector?

[dz`]

See the two floppy power connectors? one above each 16x PCI-e slot. For extra GPU stability I guess? Most likely unnecessary for single GPU setups. I don't have mine plugged in with my 8800GT but I haven't done any 3D either.

[DemonEyez]

...
Most likely unnecessary for single GPU setups. I don't have mine plugged in with my 8800GT but I haven't done any 3D either.

Wrong, the DK P35 also has the same connector located at it's PCIe16x slot (above the slot, next to the fan header), and that's a single 16x mobo.

picture of my old P35 board

This is just DFI's way to assure you get stable and decent power for the PCIe slot(s) containing your GFX(s).
So always plug it in. (btw, it also says so in the manual )
For a mere 8800GT one could do without it, but for 9800GTX and up or HD38xx and up it would always plug it in.

[dz`]

From the DK P45 manual:

"The FDD-type power connectors are additional power connectors. If
you are using more than one graphics cards, we recommend that
you plug a power cable from your power supply unit to the 5V/12V
power connector. This will provide more stability to the entire system.
The system board will still work even if the additional power con-
nector is not connected."

But I don't want to fight about it, its not really a big deal and I'd still wager that 80% or more don't plug/need to plug it in.

[Eldonko]

Oh gotcha, had no idea what you were refering to. JR board does not have those.

See the two floppy power connectors? one above each 16x PCI-e slot. For extra GPU stability I guess? Most likely unnecessary for single GPU setups. I don't have mine plugged in with my 8800GT but I haven't done any 3D either.

[decayed.cell]

using XFX 8800GT...no problems above 1.4v for NB

Using XFX 8800GT Alpha Dog Edition, on the 812 BIOS it will fail to boot with 1.4V on the NB here.

EDIT: Previously on the 0708 BIOS that it came with was fine with 1.4V NB (yes I know its a weird date, there isn't that release on the DFI website)

[cole2109]

From the DK P45 manual:

"The FDD-type power connectors are additional power connectors. If
you are using more than one graphics cards, we recommend that
you plug a power cable from your power supply unit to the 5V/12V
power connector. This will provide more stability to the entire system.
The system board will still work even if the additional power con-
nector is not connected."

But I don't want to fight about it, its not really a big deal and I'd still wager that 80% or more don't plug/need to plug it in.

But what if you have a GX2?

[DemonEyez]

Just use common sense and plug it in

Originally Posted by C-N
Let us know how you get on as it would be nice to work out whether the much coveted A3 revision of the P45 is actually a superior lower voltage chip than its A2 brother or not.

I did some futher testing. Is was looking for mixture of high performance but lower voltages for a 24/7 setup.
So my first step was to go from 9x500 with 1.32v to 8.5x500 with 1.29v
That's when I stumbled on the following.

Remember your tight timings for 8.5x500 with 1.54v applied on your NB, whereas I had to apply 1.40v for9x500 with same timings...
I'm talking about these ones

CPU Clock Ratio........................... 8.5x
CPU Clock................................... 500
DRAM Speed............................... 333/800

Enhance Data transmitting............. Fast
Enhance Addressing...................... Fast
T2 Dispatch................................. Disabled

Performance Level........................9
Read Delay Phase Adjust................All on Auto

Ch1 Clock Crossing Setting............ More Aggressive
Ch2 Clock Crossing Setting............ More Aggressive
CH1CH2 CommonClock Setting........ More Aggressive

... I just finished 9 hours of Prime on 8.5x500 with only 1.36v applied to the NB.
Seems that when you lower the CPU clock there's some less straining on the NB and it can do with less voltage.

[DemonEyez]

I've seen some people do a FSB of 600Mhz or more in this topic. I was wondering, does anyone of you have Prime or Orthos screens of let's say 8 hours?
If so, I would really appreciate your settings because I can't seem to get mine stable.
Allthough my E8600 primes at 9x500 for 8 hours straight , it can't do more than 5 minutes on 7x600. SuperPi wil run, that's no problem, but then agian i doesn't strain the system that much. So I was reading back into this topic to look for any info on stable settings for 600fsb and then I started to notice that all of the posted screenshots with 600fsb only have a SuperPi/PiFast run in it, but Prime or Orthios screens are nowhere to be found.
Is searching for stability in Prime with 600fsb like searching for the Holy Grail?

[C-N]

I did some futher testing..............

Looks good,

It was only really forcing nominal/aggressive to run at high fsb on the 266 strap that required the high voltage. My P35 board loved running on the 266 strap but the P45 or at least mine dislikes it. I also noticed the PL goes from 6 ~ 7 I think somewhere around 400fsb which is a bit weird my P35 stays on 6. One last thing remember I was referring to BIOS set voltage & you were referring to what SG read which is roughly .05v different.

I don't think I will bother getting the plus board after all as I think the UT P45 is going to be available real soon fingers crossed.

I haven't touched my board all week (no spare time) I've really got to get my better E8500 in & on water & a decent video card in before I find a nice 24/7 clock to settle on.

The only thing I've done interesting all week is make up/tweak Smartguardian using a skin a (DJSUB) posted up in the P35 thread & re arrange the .ini so it all fits in a box the size of core temp. I'll post it up later for anyone that want it.