This post is more than 5 years old
1 Rookie
•
17 Posts
1
25108
August 2nd, 2017 18:00
Precision T3600 / T5600 / T7600 heatsink alternatives
There have been many threads about sourcing a heatsink alternative for the DELL T3600 / T5600 / T7600 Workstation. Having acquired several used T5600 workstations at a very low cost, I had the idea to make all of them dual processor machines. There appeared to be two obstacles to this idea of buying a CPU, putting on a heatsink and loading up the ram. The problem was the heatsink is different. Very different.
PHOTO: Left is the Intel heatsink and right is the DELL heatsink. Notice the different bolts and springs.
The two main heatsink issues
1/ The latched fan connector
2/ The Independent Loading Mechanism (ILM) is not the most common variant. The ILM is the pressure latch and lever arms holding the CPU onto the socket. Various ebay sellers describe it as a latch, socket, holder .. in fact it is an ILM.
Lets go through this one step at a time.
Heatsink fan and its connector
The connector DELL use on the PWM Heatsink fan is a standard connector. Due to the cyclonic forces created by the DELL heatsink, the resistance of the fan connector latch is the only force capable of resisting the gusts of wind driven by 15-30 cubic feet per minute (cfm) of volume.
However, the resourceful sellers of ebay / newark / digikey / element14 have a solution. For the non-DELL heatsinks using the flimsy PWM connector in a 4 pin housing there are two options.
1/ Cut off the standard connector and crimp on the new connector , or
2/ Purchase an adaptor / converter cable.
Even though I have the crimping tools / pins / housings I opted for the second option. This meant I didnt have to waste much time. My main reason was that I had many of these to do and I only had 20 crimp pins. These adaptor cables are called "dell fan latch" (works as a keyword bundle) on eBay
PHOTO: The converter cable I bought from eBay.
Caution: There are actually cables which allow you to use a DELL workstation fan WITH a standard non-DELL motherboard. You probably dont want this cable.
Do validate what you are connecting at the motherboard end. Also bear in mind that you want 4 wires and not the older 3 wire version.
If you must assemble it yourself, then search for SL crimp housing positive latch, in a site such as Digikey. The associated pins will be easy to find once you narrow down the housing.
Molex Part Lists:
Molex 50-57-9405 SL Housing 5 Position 2.54mm Pitch 70066
SL Crimp Terminal, Series 71851
SL Crimp Terminal, Series 70058
Heatsink Alternatives
The supply of LGA2011-0 heatsinks is rapidly dwindling. For the Generation-1 E5 CPU the correct socket is the LGA2011-0.
When you inspect this socket you notice that the corners where the heatsink attaches are "missing". In other words, compared to other types of motherboard (non-DELL) the lugs are not above the motherboard but rather underneath.
This is because DELL has opted for an industrial ILM where the attachment posts are not held with the ILM but rather heavily reinforced bolts are press welded into the actual chassis underneath. Please refer back to my accurate and scientific description of the cyclonic forces generated by the heatsink.
Due to this variation the bolts for "normal" heatsinks may not reach or may involve such enormous forces (to resist the heatsink springs) that you may well risk damaging the thread on the chassis mount or heatsink.
There are 3 types of LGA2011-0 ILM (X79 or C600 chipset) with regard to types of heatsinks.
* Narrow: TE 2134439-1
* Wide TE 2134439-2
* Wide and Deep (DELL) - without ILM heatsink studs. TE 2134439-4
PHOTO: The original ILM on my T5600 motherboard.
An ILM is attached to the supporting backplate by 4 x T25 / T27 Tor-X screws.
The good news is that Digikey sells the 2011-0 Wide ILM cheaply. Part number A108323-ND. This is a TE part number, there is a Molex version. Search LGA2011 ILM.
PHOTO: TE 2011-0 R-Socket LGA2011 replacement Wide ILM with cover. Backplate not shown
Take a moment to appreciate the elegant simplicity of having the heatsink lugs above the motherboard. This is why the DELL screws are so much deeper than the standard 2011 heatsinks.
Now before you go rushing into this, two things.
1/ Before you remove the ILM make sure the machine is inverted so that the backplane doesnt fall off into the void. Then its going to be a matter of removing the motherboard or having the patience of a saint.
2/ During modification your LGA socket is totally exposed and unprotected. VERY RISKY
Carefully attach your new ILM to the backplate behind the motherboard. Do not over-torque the backplate bolts.
Heatsink options
There are two immediate heatsink options for the total power dissipation (TPD). I was using TPD around 95W in the CPU and the cooling products I was looking at are both rated at a TPD of at least 130W. This means I have adequate cooling capability for the CPU I am using (E5-2640).
* Intel Thermal Solution STS200C TDP=150W
* Supermicro SNK-P0048AP4 TDP=145W
These are 2U active server heatsinks and they are both able to clear the CD cage at the front of the machine above CPU-0.
In brief, to get around the DELL T3600 / 5600 / 7600 heatsink issue you need the following parts
* Replacement ILM
* TORX tool kit
* Molex pins + housings OR 4 pin PWM to Molex 5 pin Latch adaptor cable
* a 2U server LGA-2011-0 heatsink with a 4 pin fan.
Results
Using Arctic silver thermal paste, the thermal performance is indicated as stable and reliable. When installed as a DELL + non-DELL dual CPU combination the thermal performance of the second heatsink matched the DELL heatsink and did not suffer any ill effects in normal usage. This experiment does not validate intensive applications such as 3D modelling as we do not have 3D software. Core Temp by ALCPU was used to measure the temps.
All performance benchmarks such as Cinebench R15 ran correctly.
My motivation for doing this change is mainly to address the issue of making the workstation more powerful and breathing new life into an old workstation. The Dual CPU E5 solution I am using with 32GB ram cost me less than a new i7-7700K processor.
However, this solution is not valid for any CPU which has a TPD close to or above the TPD of the heatsink. Some of the most powerful CPU options available do in fact exceed the capacity of the heatsinks mentioned above.
It is not worth risking safety and reliability if you plan to use a E5-2687W CPU as this CPU at full speed will draw 150W and saturate the heatsink which will throttle back your performance. I deliberately selected the E5-2640 because of its thermal load / cpu core ratio (watts per core).
Future
A letter explaining the problem has already been sent to Supermicro. They were willing to consider the issue because they have a heatsink where the heatsink bolts are on a removable plate which makes the same heatsink for wide or narrow in one box. Hopefully they will consider making a spare part for Wide and Deep to fit with this style of engineering.
If Supermicro decides to make a spare part which addresses this issue then you will not need a change of ILM. Basically all that needs to happen is the screws need to be longer and the springs need to be shorter or softer. I have been looking on eBay for springs and screws to see if I can make my own hybrid. An example below of a removable mounting plate from Supermicro.
PHOTO: A Supermicro part showing the bolts are exchangeable via a removable plate
Comments
I agree that some users have had no issues using the Supermicro heatsink directly on the original DELL ILM (made by Foxconn) but I am not willing to take the risk of causing damage due to the high amount of pressure caused by the springs being used under the bolts holding down the heatsink. I am travelling down this road because it represents a long term option to make the motherboard more standard.
If someone knows how to buy the missing ILM studs, then for a few dollars the issue would be resolved.
Caveat
As all of my machines are well outside warranty I have done this to assist with installation and refurbishment in situations where the DELL option is not available or easily provided. Some of the steps mentioned here are risky and may permanently damaged your CPU socket if an accident occurs during assembly. Do not attempt to install non-DELL heatsinks using the standard DELL ILM as the pressure and forces may exceed the tolerances of the screw threads and permanent damage may result to your chassis. Do not exceed the TPD of the heatsink with a CPU option. This method is not valid for LGA2011-1 or LGA2011-3 CPU sockets. Please refer to the specific model machine you have before attempting to make any changes. Modifying your motherboard may void support agreements. Perform at your own risk. I do not work for Digikey and I have no affiliation with or benefit from suppliers.
mjur3191
1 Rookie
•
17 Posts
0
August 7th, 2017 07:00
A photo of the new ILM installed with the old backplate. Heatsink studs on each corner. (Wide model)
There is almost no strain on the mounting studs of the ILM given the distances and threads match perfectly.
Final result, fits under the CD tray too. NEW ILM and Intel Heatsink with DELL adaptor cable.
Verdict
In terms of flexibility and options. I would recommend this mod for applications where you want to use your own specific cooling solution up to and including a TDP of 130W. Above 130W it becomes a totally different problem and I would suggest staying with the original DELL products at extreme heat levels unless you have the resources to go beyond, such as water cooling.
Arctic silver heatsink compound (or better) is highly recommended and performs better than standard heatsink compounds by a margin of 5C temp in most cases. I would recommend removing the grey factory paste in favour of more thermally efficient products.
speedstep
9 Legend
•
47K Posts
0
August 6th, 2017 09:00
Digikey in thief river falls has been around for many years.
Your post is long and likely TMI for a mere mortal.
There are specific sources for the fan conversions.
http://www.rexususa.com/cableadapt/3-pin-cable.html
Compatibility
Dimension 4300, Dimension 4550,
Dimension 8200, Dimension 8250,
Dimension 8300, OptiPlex GX240,
PowerEdge 400SC
The standard Fan connectors are not the same as Dell connectors.
P/N: CB-PWM-D5
https://www.amazon.com/adapter-connector-converts-proprietary-sleeved/dp/B005G50C6Q/
The TE plate has a very long lead time.
https://www.digikey.com/product-detail/en/te-connectivity-amp-connectors/2134439-2/A108323-ND/4022435
You also didn't mention
Is this the heatsink you are referencing?
https://www.digikey.com/product-detail/en/delta-electronics/DHS-B9292-05A/603-1748-ND/3621529
mjur3191
1 Rookie
•
17 Posts
0
August 6th, 2017 16:00
a) The 3 pin PWM cable is not ideal for the T3600 / T5600 / T7600. That is the "previous" design needed for computers I dont use.
b) There are 156 TE ILM in stock at the moment. Took me 3 days to get one, half way across the planet.
c) Backplate is standard issue for an ILM. Using the old plate is acceptable. Buying a new backplate is optional.
d) Heatsink is the passive version of the Intel package. The Delta part number, has the area to install the fan. But the heatsink you refer to has no fan.
Does anyone sell the ILM studs ?
Scuffers
6 Posts
0
August 19th, 2017 09:00
Hi,
Interesting thread, Looking to do something similar, I embarked on fitting a Coolmaster Seidon 120V water setup on my T3600 and it's E5-2687W.
My motivation is that with the std cooler, under stress, temps would rise to the point that the CPU starts to step down to control the temps - typically this seems to happen at ~90C core temps (using HWmonitor), so I was looking for a more efficient cooler, the 2687W's 150W rating looks just too much for the std setup.
Physically fitting this was easy enough - once I realised that Dell use M4 threaded heatsink mounting bolts.
Brought a couple of fan lead adapters, one to go from Dell 5 pin on the MB to std 4 pin PMW, then one to split this to 2x 4 pin (one with PMW, one without).
Plan was to use the non-PMW output for the pump and the PMW one for the fan.
All went together OK, plugged up and turned on, to be confronted by a Bios message to tell me the CPU0 fan had failed (even though it's spinning round just fine).
So, after a couple more cold-boots, decided to press 'ignore' and carry on with booting it.
All worked fine, using HWmonitor, temps were well down on using the std Dell heatsink, and almost totally inaudible (even with the side cover off).
So, next I tried stressing the CPU, this system is used for Handbrake a lot, so fired it up and gave it a 4K UHD file to crunch.
Temps climbed a bit, but was well down on the std air setup, cores were all under 70C, BUT - and this is where it get's hinkey, the CPU started to step down speeds again?
Played about thinking that the fan failure warning may well trip something in the Bios, I reattached the OEM Cpu fan into it's usual socket and use the HDD fan output for the coolmaster, bios passed the CPU fan, but then failed the HDD one! (clearly does not like the fan), however, the same behaviour, temps all look great, but CPU's still stepping down.
So my questions are:
1) what's the BIOS criteria for fan failure warnings?
2) what's the driving criteria for the CPU speed stepping?
Any comments/advice/info gratefully received!
Scuffers
6 Posts
0
August 19th, 2017 10:00
Thanks for the input.
Currently running A14 BIOS (see A15 is out but looks to be a security fix only).
CPU is not original fit, I gut it as the best option available for the board (as in fastest), and with the 635W PSU seemed like a good option.
For completeness, full spec:
T3600 with 635W Psu
Nvidia Quadro P600
Dell Perc H310 Raid card
64GB memory (4x16GB)
4x Samsung SSD 850 evo 500GB
2 x ADATA Premier Pro SP900 SSD 128GB
Optiarc BD Rom BC-5500S4 Optical Drive
RE-power - good point, sharing the pump with the fan may not be such a great plan, I'll try separating them - adaptor leads all look good, checked the pin assignment and all good.
Currently got it back on the OEM air cooler, and with enough airflow (ie, jacking up the cpu fan speed to the max) it works just fine if you can ignore the noise - temps stable at 80-90C core temps, no speed-stepping.
More idea's welcome!
PS. is there any decent way to control the fan speeds as BIOS seems *** at it and all the fan speed applications I have tried either run at slow or flat out, nothing in between.
PPS. just seen your edit, 130W... Humm.... that might explain the std setup not cutting it .
Interesting, HWmonitor shows it running core power up to ~151W when loaded, and the CPU can run at max speed if kept cool on air...
mjur3191
1 Rookie
•
17 Posts
0
August 19th, 2017 10:00
I have to wonder if you are using the fan header to drive the pump. That's not going to be enough current coming out of the board to make it work well. But if you are suggesting the PWM fan in your cooling kit is not interfacing well with the DELL Motherboard, then thats purely a pinout or fan issue.
I never saw a fan fail with the Intel and Supermicro solution I posted. I am going to assume from this arrangement that my converter cable and fan were a good match.
Second thing that springs to mind is the CPU exceeds the power rating for this motherboard. Having the most power hungry CPU in the E5-2600 family might be an issue for this motherboard. So I have to wonder if your bios is the latest version.
I was deliberate in trying to marry the E5-2630 / E5-2640 with the T5600. My logic was that I wanted fantastic value for money and low heat. If i needed more cores and more performance, I would tend towards a V2 processor and a T5610 motherboard. As each E5 version comes out, you see more cores and less heat. Since the V4 and V5 technology is premium priced, I stay away from the cutting edge.
It seems to me that you have tripped a temp setting buried deep inside the board. The airflow you get from a fan assisted heatsink also cools other parts on the motherboard. In the past I have seen temp readings which were not immediately available from a "generic" tool. This included all kinds of additional readings including the power supply, motherboard and various FET temps.
It is not uncommon for DELL to release better spec machines in the second half of the product life and for the early spec sheets to not actually reflect these options. Can you confirm your CPU was a factory option because the spec sheet I read could be a very early version?
In summary, it could be as simple as BIOS, missing readings, lost airflow, excessive power demands for this board (130W limit) to the CPU demands (150W)
mjur3191
1 Rookie
•
17 Posts
0
August 19th, 2017 21:00
I dont always follow the CPU guidelines myself as far as processor selection goes but I would stay within the power consumption guidelines.
It seems that you have two problems.
1/ Powering the liquid coolant pump with a 2A source. The PWM output is 1A in total. I have used PWM splitters in the past and they work amazingly well but I was concerned to stay under the power output limits. I have seen the Corsair kits and they usually come with a "SATA Power to Pump" cable header. Only the most recent motherboards come with a pump header for liquid cooling. It is clearly marked as a fixed voltage header as coolant pumps do not respond well to variable voltage under the PWM controller. The issue is that running the pump slowly will cause a temp spike.
2/ Looking at other possible components reaching T-max. For this you might need a laser thermometer from eBay. In the type of config you are using I would consider motherboard chipset cooling and other cooling locations. You are effectively over-clocking the motherboard above the factory max.
Overclocking tends to heat up the entire motherboard and requires heatsinks in locations the factory board does not have heatsinks or requires much larger heatsinks.
Cooling is the main issue as I am considering the possibility of other components causing the stepping back on the CPU speed. Keep in mind that PWM can also sense the MAX RPM of the cooling solution and if your liquid system does not have the fan RPM the DELL Heatsink has, then the system can also step back to the "assumed" max performance of the fan under heavy load.
In other words, if you are getting 2000 rpm and the DELL heatsink is 2250 then your top CPU speed will be a variation of the cooling curve so that your effective limit is 2000/2250. You will not be able to reach top speed with the assumptions integrated into the motherboard unless your fan has the RPM and the CFM to handle the cooling profile expected by the BIOS.
The only way around this is not to fall outside the expected performance levels for RPM / Heat / Power in the motherboard sensors. Because you are effectively overclocking using a high-spec CPU.
Scuffers
6 Posts
0
August 20th, 2017 09:00
Good info...
Will have another play and let you know what I end up with, current thinking is that it's fan power/speed related issues.
mjur3191
1 Rookie
•
17 Posts
0
August 21st, 2017 21:00
I took a closer look at your problem. See the attached photo. Look at the regulators and chips which are air cooled by the heatsink airflow. This might be where the power regulators are overheating when using a fluid cooling solution.
Would be interesting to put a laser thermometer here and see the temps when the CPU steps back.
If this turns out to be the main issue then you might need to look at adding heatsinks or some other cooling process.