12-31-2018 05:46 AM - edited 01-04-2019 06:21 AM
Thanks to some of our pioneers in overclocking and modding the Z620s, here's my progress on the project:
Z620 + 1650 V2 (stable at 4.5 Ghz). Max temp - 49 degrees Celsius.
A few things to mention:
- I wanted a single-processor config, because I need all the PCI ports I can get.
(GPU + NVME + FrontPanel card reader and usb 3.0 + quich charge USB for Android Phone)
- I am shooting for the 1680v2, at the maximum OC it can take (probably 4.8)
- I have quited the Z620 system fans, by replacing them as follows:
1x Arctic F9 PWM (rev2) in the back, hotwired to use the 6-pin from the original double 90cm HP Fans
1x Xilence 120mm fan in the front, 3-pin, replacing the HP original 4-pin - this runs at full RPM all the time, but it's a low-noise, low RPM fan anyway 🙂
I did this because HP has a custom fan curve for the case fans as well (probably BIOS hardcoded) when the CPU goes into full load.
Also - the fan grille for the Corsair H115i was needed... I have a 3-month old cat running around the house, don't want any accidents 🙂
BIG THANKS to:
Javato (my original source of hack-mod inspiration)
Brian1965 (check out his OC guide for Z620)
DGroves ( very good "security & proofing" tips, not to mention excellent wiring skills! )
12-31-2018 02:50 PM
nice cooling mod,and great performance ...... only things i would try to do is
1. make some brackets which connect to the back of the case to hold the external radiator in place i would make a bottom plate that extends outwards from the back of the computer and the use two side plates (that are aprox one in wide (one on each side) at aprox 35 degrees from bottom end of rad up towards where hoses exit and then two more one in plates from top of rad to back of case
2. protect the water lines exiting from the back of the case using perhaps AC Line foam insulating coverings
3. lift the case/rad aprox 6in off floor using or making short legs and a base for the system/rad to sit on to prevent all of the dust that exists in this area from entering the rad/case fans
01-01-2019 08:50 AM
An excellent external cooling solution- very well done!
I'd made a big pile of designs for external units:
> before deciding to buy an Alphacool Eiswand integrated unit which has 6X 120mm fans in push /pull at constant speeds to reduce noise.
I didn't buy the complete unit, only the radiator/fans/pumps/reservoir unit and upgraded the other components as compared to the kit, for example an all-Copper CPU block instead of Copper / Acrylic and metal quick disconnects instead of plastic.
It would be possible on your unit to, as in the Eiswand, to bolt another two fans on the other side and run them in the opposite direction, improving CFM and the fan speeds will be lower and the unit quieter.
Also, given the cat situation- and cats are heat-seaking- consider adding a base for the radiator/ fans for stability. This doesn't need to be elaborate, for example:
Another grille could be added on the other side with a spacer so Fluffy won't rub against the bare and hot radiator.
It would be interesting to see more temperature results in HWMonitor or similar and especially interesting to see Passmark results for the system at 4.5GHz including the single thread results. These are the Passmark results for z620_2 with the E5-1680 v2 8C @4.3GHz on all cores:
The critical score in this system, given that 90% of the CPU-intensive work is 3D CAD, is the single-thread performance. My goal was to have 2300+ where the average for the E5-1680 v2 is 2102, so the 2368 is more than satisfactory. Our friend Brian1965's z620 is running 2500+ and that level is my next aspiration. As z620_2 is not entirely happy at 4.3GHz on the z420 liquid cooler - a couple of random shut downs of XTU, is the motivation behind adding the Alphacool cooler. I'm also considering a new Xeon W-2145 system or building an i7-9800X system, but it's certainly more fun to hot rod a z620.
Again, excellent work and I for one would enjoy seeing some more temperature and CPU performance results.
01-03-2019 03:27 PM
Thank you sooooo much for the free brain-picking 🙂
The H115i already has cable insulation, I've also ran a few (3) layers of duct tape on the cut sides (after also sanding them a bit, to remove any chips / blatantly sharp areas):
It's that kind of weaved fiber, like on shoe-laces... only out of some pretty stiff nylon. works great!
@BambiBoomZ, as requested:
What I can notice from comparing the screenshots :
For single threaded performance - looks like the 1650 is the bang-for-buck of the crowd. Still gets you REALLY close to 2500 score.
For multi-threaded performance - clearly the 1680 v2 is the champ!
Fortunately, I also need single-thread most of the time when actively using the computer (Unreal Engine, Lightroom, Photoshop)
For Unreal Engine, when I need to build lights - which is mutithreaded.... I have an ace up my sleeve:
WORKSTATION is my Z620
D4R4 is a home-server, built on a s2600cp motherboard (bought for cheaps from ebay). The home server currently sports a single 2670 v2, but I'm about to drop in the second one soon. Cooling a 2670 v2 with a H45 is ludicrous:
By the end of this year I should run:
WORKSTATION - 1680v2
D4R4 - dual 2667 v2 (to get some sense of worth out of those H45 coolers)
Last but not least, thanks for the clear graphic explanation of what DGroves was also suggesting - I REALLY need to build that base for the cooling unit!
Looks like more screenshots will come your way after I get it done!
PS: the benchmarks I've run on the 1650 v2 and all the temps listed so far are on QUIET mode! I literally have an HDD cooler in the D4R4 home-server which is noisier that that
08-10-2019 07:28 AM
Congratulations! Thanks for the credits! awesome job!
... BTW, in the meantime, I had to change my H45 chillers. Failed (both at the same time) around 2 years after they entered "into service"... Bottom line: do not buy the cheapest again. now with H60s 2018 version.
08-11-2019 07:58 AM
Levels and Maps,
Thanks for the update; the system performance appears to be very good. Congratulations on the 2483 single-thread - well done!
Yes, given the springiness of the cooling hoses, a simple base for the radiator should be a help.
Since writing last time, z420_3 has had upgrades, mainly as I was having screen tearing when using it with a Samsung 40" 4K monitor and decided to have a stronger backup system.
The specification now is:
HP z420_3: (2015) (R11) Xeon E5-1650 v2 (6C@ 4.3GHz) / z420 Liquid cooling / 32GB (HP/Samsung 4X 8GB DDR3-1866 ECC registered) / NVIDIA GeForce GTX 1060 6GB (EVGA SCC) / Samsung 860 EVO 500GB + HGST 4TB / ASUS Essence STX / Logitech z2300 2.1 / 600W PSU > Windows 7 Professional 64-bit (HP OEM ) > Samsung 40" 4K
A 2500+ Single Thread Mark is still the goal with z620_2 (currently 2364) but the 2384 is the highest of any office system thus far and was a bargain; the E5-1650 v2 cost $90. The clock speed limitation is the z420 liquid cooler, which idles at 33-36C. For some reason, Its not as effective as the same cooler in z620_2 which idles the 1680 v2 at 29-33C. I'm going to remount it as a first measure. I hadn't intended to change the E5-1620 v2, nor the GTX 660 ti, but there was slow 3D modeling and navigation on the large projects and some screen tearing on the Samsung 40" 4K monitor. It's also useful to have a two systems that, if necessary can substitute for the other in the event of problems.
I still have the Alphacool Eiswand external cooler in the box, purchased 6 months ago to achieve the intended 4.6Ghz, as I'm reluctant to disturb z620_2 when so near to finishing a complex project.
The prices for E5-1680 v2 in the US at least have dropped very significantly, and that is an great processor, rated to 85C instead of the usual 72C. However, I've noted that in rendering on a 100% load, the all-core speed is still 3.4GHz, whereas the E5-1650 v2 is all core at 3.7GHz. There is a concern that going significantly over 1.5V, regardless of the temperatures may degrade the processor memory controller.
I'm looking forward to the next report.
08-11-2019 04:40 PM
If you're feeling brave... maybe try some Grizzly Conductanaut.
Liquid metal thermal paste, it will drop you some 5-7 C just by improving thermal transfer between CPU lid and fan.
ALWAYS drip the paste out in a tray, far away from your PC and then smudge it on the CPU and fan plate. Copper will change color in time, on the fan plate, but you will not get depth corrosion or anything like that.
Yes, nail polish and/or duct-tape around the CPU lid IS an option 😄
I did this for a laptop, but I didn't have to resort to it for the HP, because I've cut the case to put the external beast-cooler...
08-11-2019 06:28 PM - edited 08-11-2019 06:31 PM
Liquid metal thermal compound such as Grizzly Conductanaut have been proven effective, and as suggested, a bit difficult to handle and can stain metal parts including the CPU lid- but not destructively. I have considered it. There is a very good YouTube site, Gamers Nexus, that posts with a quite high level of engineering knowledge, methodical reviews and is very instructive as to high performance PC's. That site very strongly recommends Grizzly Conductanaut. There is a surprising cross-over of gaming and workstation hardware as well as more WS users using liquid cooling and overclocking techniques borrowed from gaming.
With z420_3, I'm beginning to believe that the elevated CPU temperature is due to, first, it's being overclocked, but also significantly, the use of a dual fan, open enclosure GPU, an EVGA GTX 1060 6GB SSC, instead of a blower style GPU such as the MSI Aero GTX 1070 Ti in z620_2, which pushes the hot air our the back panel mounting plate. It's certainly not as effective as the large, multiple fan, open coolers, and has to run at a higher RPM, but they do keep the interior of a WS case cooler. I'm not terribly worried as the temperatures are never near the limits, but given that the E5-1680 v2 in z620_2 runs so much cooler, I would like to understand and improve the situation.
10-24-2019 10:57 AM - edited 10-24-2019 01:31 PM
I realise I'm a little late in posting, but I thought I would share some images of my final liquid cooling mod on my HP Z620, (both my GTX 1080Ti and E5-1680 v2 are liquid cooled);
Temperature/fan controller board attached to the top of the memory shroud and powered from the spare Molex connector (behind the DVD drive);
Custom made PCI bracket installed in PCI slot 1;
Thermocouple attached to the back of the CPU waterblock;
Custom made PCI bracket with liquid and electrical pass-throughs;
PCI bracket CAD File;
Removed the fan from the HP stock air cooler and fitted to the memory shroud, (to cool the CPU voltage regulators);
A spacer nut was required to clear memory shroud fan;
Custom fabricated chassis made to mount radiator, reservoir and pump;
Chassis CAD file;
No-leakage, disconnects used on the cooling lines;
Custom made PCI bracket;
The final results;
I am extremely lucky since I have access to a large CO2 laser at work and fabricated everything from 0.040" (1mm) thick 316 stainless steel sheet. With this set-up I can easily remove the cooling unit without the need to drain the system.
The temperature/fan controller board;
The 3x 140mm fans are all connected to the FAN1 socket on the temperature controller board using a 3-way fan cable;
The pump is connected to the FAN2 socket. The potentiometer adjacent to the FAN2 socket allows manual speed control of this output. This is set at 100% for the pump to ensure constant maximum coolant flow in the loop.
The thermostat display on the water reservoir is powered from the FAN3 socket as this also has a potentiometer for manual speed control. Again, this is set at 100% = constant 12v DC output, NOT pulsed.
With regards to the temperature controller board, here are the settings I'm using;
TFL1 = ON (FAN1 output minimum speed set at 40%)
TP1 = OFF, TP2 = OFF (Accelerate from 40% speed at 35°C to 100% speed at 45°C), e.g. 70% speed at 40°C
BF1 = ON (Alarm if fans stall)
BF2 = ON (Alarm if pump stalls)
The reason I have the fans set at a minimum of 40% is simply because the Corsair ML (Magnetic Levitation bearing) fans are extremely quiet so I might as well)
The reason TP1&2 are set so low is because the temperature probe is mounted on the back of the water block, i.e. not in direct contact with the CPU, and there is a thermal lag when the CPU starts to heat up.