35 replies to this topic

[Oderint dum Metuant]

Very brave

I want too, but im not prepared with the back up funds

Edited by DV McKenna, 24 December 2012 - 01:24 AM.

[Teh Rav3n]

DV McKenna, on 24 December 2012 - 01:24 AM, said:

Very brave

I want too, but im not prepared with the back up funds

Thanks It's nothing really, this isn't my first rodeo w/ popping an IHS off. It's just having the right tool for the job, and that's a standard shop knife blade. The rigid type also found in carpet knives. Do NOT use the flexible types, or you will gouge the mask on the board, and maybe even cut a trace or 2...hundred. Just the blade alone, not with the handle.

There's a trick to the procedure, and that's to lay the edge flat against the board, then slowly wiggle the blade under it. Once it's under, put a little downward pressure on the back side of the blade so that the edge never comes into contact with the cpu.

Note that the alignment indentions are oriented perpendicular to the die itself. Remember where they are, and how far you can go before touching the die, this is important. Most people screw up their cpu's because these 2 little details are either unknown to them, or forgotten. That's the recipe for a successful de-lidding.

Well, I'm 8 hours into the leak test, and all appears to be good. I'm just about to make the power connection for the Aquaero 5, and the permenant pump power connection, and then it's good to go.

Here's a little candy for you kiddies.




And the shortest run of tubing...ever.

[Sen]

WOOT! NEKKID PROCESSOR!!!!!!!!

/drools @ techno-****

I have to say, while I've never done a full closed loop system personally, I've looked at more than a few of them and daydreamed. . but I've never seen one that looked quite as. . . "industrial"? As what you're doing.

I swear the CPU block looks like a combo water block/peltier unit or something >.>

[Teh Rav3n]

Thanks Sen! Glad you like it Actually, the block is a "combo", but a pump & block combination. Basically, it's a Swiftech Apogee HD lower, bolted to a Swiftech MCP-35X pump, with a heatsink thrown on top for good measure. Along with a couple of Aquacomputer water temp sensors for the Aquaero 5 fan/pump controller (that's what your seeing on the inlet & outlet of the cpu block).

I've considered a Peltier many times over the years, but I just can't get over that power draw. It's far less expensive in the long run, to just go with a single, or mult-staged phase change cooling. And, more effective too.

I've got it up and running now as a matter of fact, but with one small problem it seems. The R7950 TF 3 is making all sorts of noise out of the vrm's/pwm. No, not coil whine, but worse than. I really don't even know how to describe it, except as something that shouldn't ever come from a vid card. Only under 3d clocks too, and it's damned annoying. Looks like I may have to rma it, which really sucks because noise aside, it's hitting a Furmark stable 1248 core/1500 mem. I've not even found it's max oc yet, afaik. Strangely, it wasn't doing that in Afterburner before, just in Trixx.

I'll say this though, image quality is definitely awesome. It makes my old 6950 & 6970 look washed out with the contrast & sharpness. Not to mention the vibrant colors. Huge IQ improvement over last gen, ridiculously huge.

[Sen]

Water cooling GFX card as well? If so, 100% sure the VRMs are getting cooled? If NOT, 100% sure the pads on the VRMs are installed properly?

[Teh Rav3n]

Sen, on 25 December 2012 - 03:10 PM, said:

Water cooling GFX card as well?  If so, 100% sure the VRMs are getting cooled?  If NOT, 100% sure the pads on the VRMs are installed properly?

Definitely on all counts. You can see the backplate for the gpu block in the above pics. I had someone suggest a possible fix, but right after I managed a (stable) epic oc on that very same card. I don't think it's maxxed out yet, but it's Furmark/3dMark11/Heaven/4 rounds of MWO stable @ 1300 core/1508 mem. If the fix doesn't work, I'm pretty sure I can tolerate the noise in exchange for the POWAH!

[Sen]

You want power add a 2nd liquid cooled 7950.

Should throw some general pics up, love to see the whole inside of this thing. . and O/C screenshots. . . I mean you may as well go all out. . and I'm sure I'm not the only one out there that appreciates the occasional bit of techno Pr0n

Edited by Sen, 26 December 2012 - 07:12 AM.

[Teh Rav3n]

Lol!  Not happening.  Firstly, I'd never get another 7950 that would oc any more than about 200 mhz below this one. If that. No one gets these oc numbers on less than DICE or LN2 cooling.   Second, it's a m-itx build, only 1 pci-e on the board.   Well, I also have a mini pci-e, but that won't fit anything other than the WAN card that's in it, or another SSD. :DPics you say? Si me Kapitan! Furmark burn in results. I also looped Uningine Heaven maxxed out for nearly an hour.  Only a couple of tiny artifacts every 8-10 loops, and that's way more stress than it'll ever see in an actual game. Catzilla Beta Benchmark @ 4.5 Ghz & 1300/1500 gpu core/mem.  Not to sure about this benchmark though.  They give much higher scores to GTX 670's & 680's, when the real world performance of the 7950 & 7970 are superior.  Methinks nVidia may have greased some dev palms.   It seems to place an untoward heavy weighting towards PhysX equipped cards. I forgot to screenie the results, but I managed a 10,500 and some odd run in 3DMark11 as well.  Granted, none of this is with my ram oc'd either.  Stock 11-11-11-28 N2 timings @ 1600.  I've bumped it to 2000 Mhz at the same timings, but haven't done any more benchies since.  Tired of testing & benching for one day.  Time for a little game action. The bad news:Unfortunately, my 3570k is dud for oc'ing. Temps are pretty awesome for a single rad loop w/ a cpu & gpu on it.  The loop reaches equilibrium at 67° C under full load even with the cpu smoking @ 1.495 Vcore.  The cpu boots stable at 1.2 Vcore @ a 44 multi, 1.38 Vcore @ 45 multi, and a downright horrible 1.495 Vcore for 4.6 Ghz to be stable.  It's still w/in the safe range with water cooling at 1.5 v, but I dropped it back to 4.5 Ghz to stay a bit safer. (1.55 is Ivy's max under water, 1.75 Vcore under DICE, and 1.95 Vcore under LN2/LHe3.  Intel's spec sheet says 1.52v, but...).  In any case, I'll just pop for the 3770k in a bit.  I'm sure someone else will be happy w/ 4.5 Ghz.  I otoh, want a 50 + multi, no compromises.Meh, no one wins the silicone lottery every time I guess. I struck gold on the gpu, and stepped in cr@p with the cpu.

Edited by Teh Rav3n, 26 December 2012 - 08:41 PM.

[Teh Rav3n]

Holy... wow, the forum really messed up the formatting of my post when I posted the edit. :/
Anyway, I just wanted to add that it's still a WIP.  Working on the last of the wiring, and modding the panels too.  Thanks for looking.

Edited by Teh Rav3n, 26 December 2012 - 08:46 PM.

[Sen]

Ahhhhhh. . nekkid tech in all it's glory I see what you mean about the back plate on the GFX card now.. I've never seen that before. 4.5 is a respectable O/C, Ivy was kind of a disappointment compared to sandy as I understand it. Benchmarks may be better at 5.0, but let's face it: You probably won't notice any real world performance over it, so don't be TOO discouraged [of course I say that with a 3930k that'll do up to 4.9, but likes to crash @ idle all the way down to 4.2 >< ]

And, for the record, I've READ that PCI-E GFX will work in PCI-E 1x slots. . . though you might lose some bandwidth or something like that [I have no way of testing that, just sayin'. . . .]

[Teh Rav3n]

T(h)anks Sen Have you tried disabling C1E, C3, C6, and Package C to see if their causing the low power state problem? Also try toggling Internal PLL Overvoltage, sometimes that can cause instability. Maybe in low power state, the boards under shooting Vcore and causing the crash. Are you running in offset or fixed Vcore mode?

You're correct about the PCI-E 1x slot to PCI-E cards, it's just a bit bandwidth loss. This on the other hand, is a mini pci-e, the type typically found in notebooks/laptops and m-itx boards.



About the only thing you can mount in them is WAN cards & specialized SSD's (which are much more expensive than even the mSata SSD's, & by far more pricey than full size 2.5" SSD's. Surprisingly enough).


Oh, it's not a bad oc on the cpu. But the average de-lidded Ivy will hit about 4.7 Ghz, and on the top end, 5.4 Ghz is as high as they'll go under water. Mine ranks in the bottom third percentile unfortunately. Luck of the draw I'm afraid. At 5 Ghz, Ivy's more than a match for Sandy, much less at 5.4 Ghz. Especially with a higher Tj max than Sandy's capable of.

Really, the differences lie in the way that Intel chose to attach the heat spreader. Ivy isn't soldered in place like Sandy is, they used an adhesive and a really bad grade of TIM, with the intent of (I believe) crippling Ivy a bit to make SB-E look better for moving units.

Remember when the 990X hit the market, and Sandy was whooping it in single threaded performance? 990X's didn't sell too well because...well why buy a $1300 chip that performs slower than the $400 chip in nearly every task? Client side wise, not many people had a use for a cpu that would only perform better in an multi threaded application that could take advantage of the 2 extra threads. Think professional CAD/Rendering/Vid editing & conversion, and you can see immediately how small the advantage of the 9**X chips actually was. That would be why they intentionally crippled Ivy, or at least in my best estimation.

Back to the topic at hand. I upped the voltage to 1275 mV and it ceased artifacting in Heaven. I then managed to eek another 250 Mhz out of the Vram overclock (now at 1300 Mhz core/1750 mem), and in doing so, running the Uningine Heaven 3.0 benchmark, bested the scores of every single gpu system with a 7950, GTX 670, 680 & lower on Overclock.net. In fact, the only 2 single gpu systems that scored higher, were 2 7970's that scored .5 & 1.3 fps more than my 7950 TF 3 respectively. Core clocks were the same on those 2 systems, and my mem oc was better, but the additional shaders the uncut 7970's have available for that 5% performance increase made a small difference. If only AMD hadn't tossed mine in the 7950 bin...



http://www.overclock...mark-3-0-scores

...and mine:



Not bad eh? Obviously below the top performing CF & SLI equipped systems, but...

----------------------------------------------------------------------------------------------------------------------------------------------------------
For reference, these are the settings to run Heaven 3.0 on to compare.

Render: Direct X 11
Mode: 1680x1050 fullscreen
Shaders: high
Textures: high
Filter: trilinear
Anisotropy: 16x
Occlusion: enabled
Refraction: enabled
Volumetric: enabled
Anti-Aliasing: 8x
Tessellation: extreme

Driver mods or hacks are not allowed.

------------------------------------------------------------------------------------------------------------------------------------------------------------
That's why I don't trust that Catzilla benchmark. Faster in Heaven 3.0 by a good measure than a GTX 680, yet a GTX 670 beats out my massively clocked 7950 in Catzilla, with a far lower clock speed? Not likely, and the physics scores are why it's like that in Catzilla. Heavily weighted indeed.

Edited by Teh Rav3n, 27 December 2012 - 01:39 PM.

[Sen]

Word of Blake! 1300/1750?! I can see why you don't wanna let go of it Definitely hear ya about the ivy bridge internal TIM, I know I'd seen a bit about it after ivy was released. . . I figured I'd "do it right" just once and build a 2011. . . max out one build in my life. I'm still messing with it on and off, I have the internal C3/6 off, but I'm leaving it capable of downclocking itself to TRY to save a few watts here and there. I could probably up the idle vcore and lower LLC [it'll pretty much do 4.9 @ 1.35v @ full load]. . . but all I ever do with the thing is MwO. . . it's almost not worth hammering out that kind of performance.

and ROFL at the mini PCI-E . . I can't say I've ever seen one in a desktop motherboard before! That's kinda neat You're SURE you can't fit another 7950 in it? LOLOLOLOLOL!!

Now get back in there on that GFX O/C. . . let's see 1500/2000!!!!

[Teh Rav3n]

Yup, that 7950 is a bad ***, no doubt. I'm right at the limit w/ her oc though, I pushed it up past 1350 mV last night, and the core didn't scale any further, nor did the memory. Albeit, I pushed the mem up about 30 more Mhz when I did so, so it's possible a few more Mhz may be possible. The core clock otoh, is maxxed completely. Even 1 Mhz more on the core and it starts running into the PowerTune brick wall and downclocking slightly, losing fps in the process. Temps were fine though, with the loop reaching equilibrium with the core @ 66° C. Good enough.

I'd have to get someone to write a special benching bios to take it any further. Frankly, I really don't need it to go any higher, as even running Radeon Pro's Adaptive V-sync (which was off for all of the benching), it was starting to screen tear in a few spots. No artifacts, just a screen tear mid screen.

Edited by Teh Rav3n, 27 December 2012 - 07:43 PM.

[Teh Rav3n]

Well, I figured out what the problem is w/ my 3570k. It's just a defective bunk chip completely. If you guys recall, I dropped the oc & Vcore down a few days ago. Settling on 4.4 Ghz @ 1.35 Vcore, then it starts crashing out of the blue. Bumped Vcore a little, and again a ctd. Dropped everything to stock clocks, still crashing. First in programs, now just full on bsod's. 0x0A codes. Ran memtest 86 & Prime 95 (2x each) for 12 hours on my old 1090T with the Sammie mem in it at the same settings, and zero errors.

Popped my known good SuperTalent Project X sticks (thankfully they'll run on 1.45 V) into the AsRock mobo with a friends 2500k, and did the same. Everything passed, with the only variable being the 3570k. Looks as though the 3570k's IMC is gone. That pretty much sucks, being that my mem was at 1.35 V 1600 Mhz. Really hadn't even had the opportunity to start oc'ing it yet.


Amazon's shipped the replacement already, and it will arrive on Thursday. Thankfully, they're pretty good about cross shipping on defects. Phew *wipes brow* Hopefully I have better luck w/ the new 3570k on the oc front.

Edited by Teh Rav3n, 01 January 2013 - 12:07 AM.

[Teh Rav3n]

Happy New Year everone!

I totally forgot to add these earlier, but some revised benches before the ctd marathon:

[ASSASSYN X]

Move over gents.