$9 Socket F Heatsink works on G34.

Qinsp

2[H]4U
Joined
Jan 7, 2011
Messages
2,154
ALL TEMP DATA IN THIS THREAD IS WRONG.

Wish it was easier to post pics.

I have a dual G34 system with the smaller (4U) Noctua dual fan heatsinks.

I was running CPU1 = 63°C and CPU2 = 53°C while folding in a warm room.

I didn't like that.

So I took a 1207 $9 2U copper heatsink w/fan (Tyan PN ?) and made it fit G34.

I stopped the machine mid-fold, swapped out CPU1, and ...

Are you READY???? I should have video'd the whole before and after.

CPU1 = 41°C and CPU2 = 53°C

Stainless steel is a CRAPPY heatsink. It look pretty but it sucketh.

Will post pics after I get the kids from school.
 
Last edited by a moderator:
I didn't even know that stainless steel was an option as a heatsync. :eek:
Aluminum or copper for life.
 
As been my experience with heatsinks, any form of steel that I've seen simply absorbs too much heat, and while it takes to to absorb it, it also takes a lot of time to dissapate it.

Copper is probably the best material to work with since it can absorb and dissapate quickly.
Aluminum is ok, it's just a bit cheaper and not quite as good as copper.

Have you ever considered using something like this in one of your rigs?
http://www.newegg.com/Product/Product.aspx?Item=N82E16835118019
 
Use imgur to host, don't need an account and after it uploads, it gives you URLs to share, even a bbcode one to paste right to these here forums
 
^ That's what I was starting to think.
Maybe the surface has flaws in it.
 
I suspect mounting issues.

On the copper Tyan 2U or the Noctua 4U?

Did I get the copper too tight? :D

Same procedure I always build to, and not my only Noctua's. Same Arctic Silver, same cleaning procedure, same seating, same sequential tensioning method. The Noctua showed a normal pattern on the CPU, center high, metal to metal.

Nickel-Plated Aluminum? Either really thick plating, or the strongest aluminum sheet I've witnessed, and I do satellites. PS - Stainless is high nickel alloy.

The fins on my old Noctua's aren't like these POS's. I thought they were shit when I opened the boxes, and I wasn't disappointed.
 
^ Ok, you probably wouldn't be able to fit the Zalman 110mm virticle heatsinks into the 2U, lol.
 
It just goes to prove the old adage, "If it ain't fast, then chrome it!"

EDIT - I put one of those Zalman's on my son's computer. It works good and is pretty at the same time.
 
Last edited:
It just goes to prove the old adage, "If it ain't fast, then chrome it!"

In the future, EVERYTHING is chrome!

chrome.jpg
 
Here's the finished pic:

dHKnmJB.jpg


After Temp:

DIar9D5.jpg


The room is 30C at the time.


This the CPU1 fan you see with a 1.125" x .500" x .125" strap extending the original 3.500" pattern to 4.125". Holes are .625" apart, one is .213" dia, and the other is .257". The existing hole in the heatsink is the correct tap drill size for 1/4-28 thd. Use a 1/4-28 x 5/16" cap screw. You will need a spring with a shorter compressed height. It took 30 minutes to make one strap with a hacksaw, a file, and a hand drill.
 
Last edited:
As an eBay Associate, HardForum may earn from qualifying purchases.
Even if it is nickel-plated aluminum (I doubt), the nickel insulates the blades. Inconel is a nickel alloy that is famous for rejecting heat transfer. But those fins are STOUT. Several of them were bent from the factory and took pliers to straighten them. Sure didn't feel like aluminum, so I have my doubts.
 
Well, whatever the reason, I'm going to build my 4P G34 GL system with these. Running 40C on air in a 30C room at max effort is good enough for me. And it will fit into the 2U case.
 
That's Tyan's 2U Socket F from what I can tell.

IIRC they also default to pull configuration (Tyan 2U cases have 4x 80mm Deltas on intake side).

EDIT:
img_1369.jpg
 
Did you double check TPC's node/socket mapping? It's not always intuitive... [just sayin']

One typically does it by:
1. Taking temp snapshot (reference)
2. Disconnecting fans on one HSF at a time and
3. Examining temps; node(s) with raised temps --> bingo
 
Did you double check TPC's node/socket mapping? It's not always intuitive... [just sayin']

Nope, I guessed.

Node 0/1 was 63, node 2/3 was 53.

So I shut it down, swapped the one closest to me to see what changed, then fired it up.

Now I know the one near the front of the case is 0/1. If the 53 number would have changed, then it would be the other way.

Sometimes I just block the air to one CPU to guess, but since it was already so hot, I didn't bother, since I couldn't run that way anyhow.

In the morning I will do the other, and start to overclock again.
 
Last edited:
So, you're saying that this 2U HSF took 23 deg off Noctua?
 
Tap this hole:

8lZYQI3.jpg




1/4-28 through. This puts the fins dead center on the CPU.
 
Last edited:
So, you're saying that this 2U HSF took 23 deg off Noctua?

In this instance, there is no doubt in my mind. It's two things. That Noctua 9 sucks, and the Tyan is really good. I saw it when I was using them for bench testing.
 
Material: Copper (base and heat-pipes)
aluminum (cooling fins)
soldered joints & nickel plating

Say again?

I don't think that tech sheet is right. These fins feel like steel when you bend them. My bigger Noctua's feel like alum. Aluminum doesn't spring back. But like I said, nickel plate is for pimps, cheap jewelry, and Harley's, it stupid for cooling fins.
 
Re-verify. Stainless is an expensive material and generally more difficult to work with in comparison of other metals. There are really no thermal benefit to use stainless over copper or aluminum.

I would advice taking more of a scientific approach in terms of the application of TIM, mounting method, pressure used to tighten heat sink, fans, and use a test wu to recreate scenario. The amount of TIM and method of mounting can create an improper contact pattern with the cpu. Ensure mounting is not too tight causing motherboard to cup slightly.
 
It was folding, I took it's temp.

I shut it off, then 10 minutes later restarted.

Then I waited 20 minutes and took the temp again.

Oddly Node 2/3 remained at 53C. No surprise since I didn't swap it.
 
PS - On a normal heatsink, you cannot over tighten. It's a shoulder bolt that bottoms out. Tension is done with a spring. If the Noctua's spring rate was too high, the center of the CPU wouldn't be bald when I pulled the heatsink, the edges would be.
 
On my Socket F Noctua 92s, I pulled the C-clip, removed the screw and spring, flipped the bracket a half turn, and put it back together. Mine were way too tight from the factory also.

On the G34s, I left them how they came and they are fine.
 
I use these Noctuas on my 6172 G34 and they work great, so this is surprising. 6172 should be higher wattage than these chips (right?) and I run about 35C on the front chips and 40C on the back (eating front chip exhaust) according to TPC. I had tried other HSFs before that were designed for 2P and the Noctuas cut 10C or more off them.
 
Dunno about the watts, they are 16 core ES chips at 2.6ghz, so they might be SE. The chips for my GL build are SE, so this is kind of a dry run. If they work well here when it hits 100F in the warehouse, then all is good. It's only 85F so far, but it's not summer yet.

In any case, the baby bricks seem to work well, and they are cheap, and it's easier to work on without the big fans in the center of everything.

Perhaps these Noctua's didn't seat well like my others did, but I doubt I'm going to go lower than 40C in a 30C room.

EDIT - Seems they are 140w chips.
 
Last edited:
Note to self:
This AM - 22.2°C inlet temp (NIST traceable digital WiFi thermometer place 6" in front of the machine), CPU1 = 37°C (Tyan) CPU2 = 50°C (N9)
 
Materials:

1 foot of 1/8"x 1/2" mild steel bar http://www.mcmaster.com/#catalog/119/3650/=maz0y1
9657K274 http://www.mcmaster.com/#catalog/119/1214/=mayxtu shorter spring.
1/4-28 x 5/16" socket head cap screw http://www.mcmaster.com/#catalog/119/3080/=maz0aw

Tools:
1/4-28 deep tap (not bottoming)
Felt Pen
Calipers
File
#3 drill
Letter F drill
Drill motor
Punch + hammer
1/8" pilot drill optional
Fine tooth hacksaw

Step #1 Blacken the first 1.5" inches of one end.
Set calipers to .250" and scribe centerline.
Scribe from end, .250" line (#3 hole location).
Scribe from end, .850" line (Ltr F hole location).
Scribe from end, 1.100 cutoff line.

Use punch to locate #3 and ltr F holes at crosshairs.
File end corners round.
Drill 1/8" pilot hole at both punch marks.
Round end drill #3 hole, next hole is ltr F.
Saw off at the 1.100"
File everything smooth.
Remove and discard the C-clip on the bolt opposite the fan. It is not a spring clip, and might fall off later. Single use.
Tap the empty hole 1/4-28 thru. Clean any chips off.
Attach strap to heatsink on top with 1/4-28 bolt. If bolt protrudes, use thin washer. I used 1/4 long bolts, but they aren't optimum.
Use new shorter spring for strap end.

Note: bolt holes are 6-32 thd.
 
Last edited:
DOH!!! Scratch all the temp data in this thread, none of it means anything except the inlet air temp. That is a fact.

TPC is not reporting correctly. Need other tools.

14°C in a 22°C room is not possible, but the BIOS health monitor reads OK.

Gotta go.

EDIT: I just shut the whole thing down for now until I load windows on it for testing.
 
Last edited:
I know you're heading back for more results, but I just wanted to add my experience with the Noctua 92s. Your noctua temps sound similar to mine....my 6172s load in the low 50s on the cool chips, high 50s on the hot chips.

But, this is with the fans set to their slowest setting and with the 5v cables attached. You have to put your ear up to my system to hear it running...the loudest fan on the system is the PSU fan. I bet the Tyan fans are 30db louder.
 
Well, semi-fail.

It's running 54/51 C all cores (TPC), so it isn't much of a solution other than being cheap and small. But seems to be stable. Do not know why I saw 14 C temps on one.

Spring rate is wrong on the McMaster P/N, so I added washers to increase rate. I'm confident this will cool in a 2U case with SE chips.
 
Code:
Temperature table:
Node 0  C0:51   C1:51   C2:51   C3:51   C4:51   C5:51
Node 1  C0:51   C1:51   C2:51   C3:51   C4:51   C5:51
Node 2  C0:46   C1:46   C2:46   C3:46   C4:46   C5:46
Node 3  C0:40   C1:40   C2:40   C3:40   C4:40   C5:40
Node 4  C0:53   C1:53   C2:53   C3:53   C4:53   C5:53
Node 5  C0:43   C1:43   C2:43   C3:43   C4:43   C5:43
Node 6  C0:47   C1:47   C2:47   C3:47   C4:47   C5:47
Node 7  C0:46   C1:46   C2:46   C3:46   C4:46   C5:46
 
http://youtu.be/FZlHUDWCAuU

Centerloading the Socket F active bricks allowed me to overclock to 3000ghz at 54°C max with 23°F intake temperature in a closed case.

The sound you hear is three 1U 4P servers with the fans on max underneath the dual G34 rig in a stack. The G34 is fairly quiet. That is nearly 1m PPD stack of crap.
 
Back
Top