The [H]arbringiner is alive

P

Patriot

[H]ard|DCer of the Month - March 2011/June 2013/De
Joined
Dec 15, 2010
Messages
2,502
Meet [H]arbringer

2012-10-30%252000.19.27.jpg


Specs
48cores @ 3.48ghz
refclock 240, nb 2160

SM GL w/ OCNG4 (thank you everyone who had a hand in this...)
16x 2gb G.skill cas 6 ddr3 1600 @ 1600
Antec HCP1200w ( a real trooper... looking to get it some help soon... 1432w AC est 1267DC)
2x RX360 rads
1x RS240 rad
8x GT-15s
Laing-d5 swiftech rebranded
4x DT G34 snipers (skripka told me to ask DT... and thus the G34 sniper was born)

32gb samsung ssd
Ubuntu 10.10 (destined to change)





Performance

Fah performance: [Classified: Need to know] :)

Cinebench 11.5 45.3
I haven't bothered updating hwbot to the new WR yet.
Partly because I already own it.


Sponsors: (Also known as the best folding team ever)

For:
TPC ,OCNG4 , The Kraken, Langouste, d3sak, and all those random scripts...
Tear, musky, dsee firedfly and the rest of the faithful testers of tears handiwork...
Thankyou!



Revision History
  • 0.8= 2.6 ghz serial wc 1333 cas 9 ecc ram 59c load @ 2.6
  • 0.9 = 2.8 cores 50-53c
  • 1.0 = final overclock target 3.1ghz (3ghz achieved)
  • 1.1 = ripjawsX cas 6 ram
  • 1.2 = spd hack to cas 5
  • 1.3 = fan upgrade GT-15s
  • 1.4 = replace xspc pumps with Laing D5 and RS240 with RX240 or larger radiator vertically in front of case Combined with 1.9
  • 1.5 = 1500w or greater platinum psu Canceled
  • 1.9 =Swap to SM GL for OCNG and 3x 12v aux plugs for cpu voltage stability

Future planned revisions
  • 2.0 = IL ES chips aiming for 3.5ghz-4ghz clock

Current parts list:
  • 4x AMD Magny Cours ES chips 2ghz suggested clock.
  • Supermicro H8QGL-iF+
  • 16 x G.skill ripjawsX ddr3 1600 cas 6
  • hitachi 500gb sata drive
  • 1x laing D5
  • 2x XSPC RX360
  • 1x XSPC RS240
  • 4x DT G34 Snipers
  • < 7' primochill tubing with anti kink coil
  • 8x GT-15s
  • Enzo copper ram sinks (not going to count)
  • Antec HCP 1200w
  • Xigmatek Elysium (now heavily modified)


Revision Walk-through, Original build log coming soon....


Rev 0.8
Limited to 2.6ghz because of the serial wc loop could not keep temperatures comfortable for the last cpu in the line.

IMG_0444.jpg





Rev 0.9
wc loop redone to have 2 cpus per radiator, 2.8ghz temps 50-53c
enzo copper ramsinks added to vrms (donated by Rtype)​

IMG_0477.jpg



Rev 1.0
Final overclock 3ghz nb 2200 (has been pushed beyond but not stable for folding past 3ghz)


Rev 1.1
RipjawsX ddr3 1600 cas 6 ram (generously donated by Dsee)

IMG_0487.jpg




Rev 1.2
spd hack to cas 5 1333 (board recognizes as cas6, early Guinea pig of tears d3sak with him guiding me though it)



Rev 1.3
Fan upgrade GT-15s, I had originally planned on doing this in the summer but 1 of the fans on the rs240 started to fail so I went ahead and swapped all the rad fans out. I may or may not order another set in the summer and do push pull. Temps mid 50s for 3ghz load.

2012-04-04%252020.50.59.jpg



Rev 1.5
Postponed/scrapped.


Rev 1.9
Swapped to a SM GL board, gained 3rd 12v Aux plug.
Added an RX360 radiator and 3 more GT-15s.
Swapped to a Laing D5 pump.
Lots of little case mods required for the switches and yes its upside down ;) ...
Swapped ES chips with R-type so that I have 4x 2ghz chips now.

New overclock tbd ... more cooling, more power, better chips. Time to see what she has got.

2012-10-30%252000.19.27.jpg
 
Last edited:
What if you run tube from 2nd cpu to dual rad then go back to 3rd cpu ? would that help last cpu cooler ?
 
And this is the answer to the question: how bad do you want the #1 spot back?
 
sbinh said:
What if you run tube from 2nd cpu to dual rad then go back to 3rd cpu ? would that help last cpu cooler ?
Click to expand...

as noted... the picture does not include the retubing... I was too tired at 1:45 to take pictures... the picture is also dated the 19th...

in quadrant style (pump > 360 rad to Q1 cpu > Q3 cpu >pump 240 rad > Q4 cpu >Q2 cpu >)

Its quite a crisscross but cools much much better...

I will of course take a picture and show you 0.9
 
Last edited:
are those RAM coming my way ? lolz ... Darn .. you guys are folding [H]ard .. :D
 
Glad to hear the revised tube routing is working better for you.

3.1ghz with spd hack is probably going to set the TPF records that will hold until they make the client scale well over 48 cores.
 
Mother of God....

What is that? Could I get some hardware details please? :D
 
So what kind of PPD will this rig produce. Will you run 2 instances of bigadv or one 48 core mega-bigadv-monster?
 
My guess:

850K ppd @ 6904 CAS5.
Maybe we should make a poll
 
My guess in at around 817K PPD on 6903 last time I played around with the numbers.
 
Untitledone said:
My guess in at around 817K PPD on 6903 last time I played around with the numbers.
Click to expand...

playing around with the numbers 960k @ 3.1 6903 I am hoping cas 6 will boost it over 1m
 
What? Nobody asked how it plays Crysis?

(my apologies for the cliche) :D
 
For the EK AMD block, I used a metal file to file down 1-2 mm of the acetal around the left and right motherboard mounting screws, which allows the block to be properly aligned with the chip. Probably doesn't make a big difference in block performance -- mostly asethetics and stability in your case -- but something to consider. In my situation, this was required given the design of the custom mounting brackets -- there's basically no play possible with the blocks once the brackets cover the blocks.

Is cas5 working for you with that MB? This nut hasn't yet been cracked with the supermicro quad -- appears to be a bios limitation. I'm still running at 6-7-5-20-1T @ ddr3-1440, even though the modules have tested fine at 5-7-5-20-1T at equivalent frequency on an SR-2 system.

Somebody mentioned FAH scaling limits beyond 48 cores -- scaling beyond this already works fine and is basically linear.

You should have the fastest single system currently -- though a couple of others have the same chips -- unless somebody sneaks out a "-smp 160" 8x E7-8870 (...or even bigger system...).

best AMD TPF for me is 6903: 11:45, 6904: 16:00. These are retail chips.


Patriot said:
Meet The [H]arbringer
^ that is a picture of rev 0.8 it ran at 2.6ghz and because of the serial cooling of cpus the last cpu was on fire if ran over 2.6... 59c @ 2.6
  • 0.9 runs currently without tweaking @ 2.8 cores 50-53c

  • 1.0 = final overclock target 3.1ghz
  • 1.1 = ripjaws cas 6 ram
  • 1.2 = spd hack to cas 5
Click to expand...
 
Last edited:
You must log in or register to reply here.
Back
Top