[WORKLOG] Silly Bends? Hard Tube Build - O11D-XL with G1 Distro

Small update here.

I did pickup some caustic soda juzz86, but sadly didn't get a chance to use it Sold the 2080 Ti + WB above here on the forums, as I managed to land a 3080 & 3090 (sold the 3090, don't need two/one already in main rig). Oh well sure I'm sure that it'll come in handy for something down the line.

Next step now is PSU.

Not sure if the trusty Corsair HX750i sits well with me if I'm aiming to give the GPU a full 450w (if stable) and the CPU currently spiking to 180w @ 5.0Ghz PBO (avg 150W at load). I know its rare/never be at 100% CPU + 100% GPU at the same time on this machine, but could be 70% CPU & 100% GPU in some sim games, might be a bit close for comfort...? Also don't mind having a tad more headroom in case next round of GPU's get stupidly power hungry.

Stupidly I dissembled & packed it up in the box, but as I'm writing this I should've tested total power consumption at the wall, if only for curiosity. Will get to that tomorrow.

FanControl and OpenRGB also setup, great pieces of kit.

- Have reduced pump speed. Whilst the very tiny whine @ 30% duty (2350rpm) on the DDC 3.2 was very tolerable, happy to take a 1-2 deg hit for now for dead silence until I plumb in the GPU/re-evaluate whole loop again then.


- Very comfortable with the dead silence now, fans/pump completely inaudible in normal use, idle temps I'm ok with for now (ambient temp 19 deg); but the important stuff will be with the 3080 plumbed in @ load obv.

- But bloody hell: Getting the Lian Li UniFans (connected to Lian Li proprietary fan/rgb controller) to sync with the Noctua fans (connected to mobo pwm) was a royal pain in the ass.

There's no way for the LL fan controller to ramp up/down based on a sensor (or custom temp input); only CPU or GPU temp offered in their software (and Fan Control doesnt detect or support the LL controller).

So the solution:
1. Connect Lian Li fans directly to mobo fan header
2. Create an rpm range & fan curve in FanControl for the Lian Li fans connected to mobo
3. Disconnect fans from mobo, then reconnect to Lian Li fan controller

and voila- the Fan Control profile for the UNI fans can now be sent across the Lian Li fan controller USB header! Now the fans (not connected to mobo) ramp up/down based on whatever sensor set in Fan Control. And more importantly, all fans ramp up/down together.

Unfortunately though- the USB header doesn't offer rpm so Fan Control reports 0 rpm monitoring from the LL controller. Not a major deal as its pretty close to the Noctua fans. Will eventually change to all Noctua's/Lian Li, pending how much this RGB shits me

OpenRGB is great, but can be a tad buggy.

Some RGB below prior to ripping out the PSU. Just toying around here, still undecided on it all. Sometimes I look at it and go 'thats nice & pretty', other times I laugh thinking what a wanker!

RAM RGB only, matching PSU cable colour:



Shots 3 & 4 above in the middle are the same but #4 is how the CPU block actually does look- a rich green without the white in the middle. Phone camera can't capture the CPU block correctly without lowering exposure (blows it out).

So next:
- PSU test, prob PSU upgrade
- Have a Bykski flow meter to test
- GPU block, then plumb the GPU
- Wrap up this RGB!

 

Looks great mate, awesome work!

The caustic can go down the drain if you ever get into trouble with furballs - it's the base of Drano, works very well

If it were me, I'd be finding a new PSU - at ~580W sustained load you're over the efficiency curve for the 750W unit anyway, but it depends where the draw will sit for the majority of the rig's life - if you turn your PC off all the time, fire it up for some games, do work/browse the web and then shut it down, well you'll want to aim for the sustained load being ~50% of the PSU capacity. If you leave it on 24/7 and it idles mostly with the odd game sesh then your sustained draw is much lower, and you want to target that, but have the capacity for the gaming load spikes. The HX750 is more than likely up to the task - overcurrent shutdown is usually +10% of rated or thereabouts, but you've got to factor in noise too. If the stars align and most of your rig's uptime will be the higher sustained load, I'd be aiming around the 1kW mark - for noise and efficiency. It'll be passive while you're working, and low on the fan curve while you're relaxing.

Very annoying that the UNI controller is proprietary - the lack of integration is something that really gives me the shits especially this far along into RGB. The headers for speed and lights are established, just go with that! They can usually be broken out to the standard, but you shouldn't have to do that either. Thermaltake another offender, although at least their later gear gives you motherboard passthrough now. At least it's not NZXT!

Coming along nice mate

 

Ok so finally got around to some quick power tests. Nothing special here, only wanted to gauge worst case power draw.
- Cranked the CPU to 1.5v, 4.7Ghz all core. Easy boot first go into Windows, finally something easy!
- Next up Afterburner: sliders all the way --> right.
- Fans/pump/led's all the way up for full powaaa.
- Launched 3Dmark & AIDA64 (AIDA to put 100% load on cpu & ram too) and clicked Run.

It started... I watched.. 500w. 600w. 650w.. ok, I need a quick glass of water.

Came back to the chair and almost fell out before gettin' in.
The water was for thirst, not in the event of a fkn fire!



First reaction was wtf happened here, until I looked up at HWInfo and saw this:



Was a tad concerned to be honest, obviously never intended to stress the PSU that hard. At ~90% efficiency (prob a tad less at that pointy end of the curve), thats a cool ~710w. Gobsmacked the CPU was drawing 200w+, that's proper out of order! As for the GPU- 450w and it was screaming for more, with GPUz reporting it was power limited the entire time, bonkers.

There are early signs I like here:
- The cpu (half decent silicon..?)
- The gpu able to max out its 450w Vbios first go
- No stability issues running AIDA64 + Timespy at the same time for 10 min (PSU handled well enough too haha)

But goddamm those power numbers and temps, hell no.

Gonna be a big day of fuckery today. Swapping PSU and a boatload of testing/benchmarking. Wasn't too keen on OC'ing the CPU yet until the GPU was plumbed in (to see what the complete loop temp was like, and then target something).. But this damn CPU now calling me to tinker.

Given the PSU goes in, may aswell rip that damn Lian Li controller out, its of zero use other than RGB.
Looked it up earlier- they do have a cable kit allowing you to bypass their silly hub, but that's $22 of bollocks i'd like to say no thanks too

Agree with you Juzz- 5yrs on with perfectly fine RGB/standardised headers, yet everyone has their take on it (for very few/if any good reasons). Calls for an open RGB consortium. I vote to ditch blue and elect yellow, and call it ORGY, it pretty much is a fuckfest at the moment.

The power bill is going up a tad I think, stupid idea to sell the 2080/deviate from the original plan

-----

EDIT: Sorry I forgot to screenshot- but water temp from memory was 32-33deg 41deg (fans obviously doing their work @ full duty/1950 rpm).

Question for you guys (its been 10yrs/possible i've lost my ways!):
- Iirc a half decent delta diff between ambient and loop temp was max 10 degrees, anything over we're talking inefficiencies in the loop. Still rings true or was that only at idle? Or maybe I'm gettin old.. Only asking as ambient temp was/is 20deg.

And as always thanks for the continued input/nuggets

 

Corsair HX1000 PSU and 3080 are in:



- Again EK cpu block isn't that blown out (2nd pic of each is correct). Got a giggle out of a breathing red/HAL type look. Temp based so green < 35deg, red >=35deg in example above, oh the silly things these RGB's can do

- Not sure if you had the same issue with your build juzz86, but I also had OpenRGB constantly loading the wrong ram profile (previous colour) on startup (1st pic). Strangely, re-created the startup entry with flags loading a newly created profile, and now it bloody works/saves

- Clearly see the GPU bracket sagging, with arguably one of the biggest 3080 AIB stock coolers going (GB 3080 Master). Will see how the GPU block sags but a small bracket from behind should fix if need be.

 

Nice work on the power draw testing. I had to scroll up and check but was that just a normal (non Ti) 3080 drawing 450W? Or did you flash the bios on it to get it that high?
The water temps you're seeing are pretty good, but that was with only the CPU adding heat to it right? Adding another 450W from the GPU will obviously make it jump up a bit and i'm curious as to how it handles it, mainly as I'm contemplating an upgrade when the 4090's launch and have a similar case/layout as you.
The 5900x's are known to run pretty hot and it looks like it only peaked once during your testing rather than running at the limit the whole time, but yeah, left unchecked, it will take all the juice it can. Plenty of options to play with here to get most of the performance out of it without all the heat though.

I'm not sure that the 10 degrees delta holds up anymore given the amount of power/heat components are adding to loops these days, and have seen ITX systems running water temps up into the mid 40's without too many issues. Approaching 50 degrees is asking for trouble though as some plastics (including the 'soft' hard tube, not PMMA) start to deform at those temps. I'd say you're pretty safe up to 40 degrees so you've got some headroom in the system. My 5800x and 6900XT max's out at 36 degrees for reference which is probably a 15 Degree delta from room temp.

 

Thanks mate. It is a normal 3080- Gigabyte Master 10GB (rev 2.0 with 3x8pins, rev 1.0 initially came out with 2x8pin).

And yep spot on, did have to flash the bios. Aware the performance improvements are minimal (groovetek actually demonstrated few days ago); but since this build is still experimental/secondary, figured it'd come in handy for testing (clocks/temps/noise/limits of water) etc.

The card came running GB's factory F40 bios (370w). After a few hours of researching- still not sure why GB released a dozen bios revisions for these; but only 2 are 450w variants (F50 & F71). Running the latter. VBIOS update was straightforward, took a couple mins.



Just double backed on my previous post re CPU running > 200w @ water temp 32-33deg <-- sorry that was incorrect. Checked the HWInfo logs, looks like that was after I restarted and relaunched. During that load test water temps were actually 41deg!

Agree, perfectly fine for here (running acrylic) and for a CPU pumping out > 200w (still scratching my head on that 230w spike though!). But 86deg CPU temp is far too high for my liking.

Ideally I'd like to target 60-65deg on this 5900x, with a water temp at low 30's - max 35deg under load. Not sure I'll get that with the GPU looped in, but will find out this week when the block arrives. Really ok with 70deg on the CPU too, but keen to see performance difference vs temp increase. By most accounts I suspect all minimal unless clocks are extreme; and that'll mean extreme temps too

So yep- at the moment solely with the CPU and VRM in the loop, GPU isn't under water yet. In a way its kinda nice- forced me/isolated how far I can take the CPU on water. Not chasing every last frame, so definitely not interested in a 200w CPU! And definitely not at the expense of loud fans & high temps. A half decent clock/something modest and she'll be purring all day (although I will try for 5Ghz all core haha. Highly unlikely, but gotta find its limits!)

I don't mind what AMD have done with PBO past few gens; seems very sensible to use it in my scenario as the PC will be doing some background tasks (light modelling/algorithms) when not used for gaming, so doesn't make sense to OC & run all core all the time.

Hopefully getting all 12 cores to boost > 5Ghz should be doable, given she did 4.7Ghz all core (at a pretty high 1.5v).

Just about to start playing with CPU now with PBO, will post back findings. Thanks for the heads up on delta, good common sense Despite remembering it as a good rule of thumb, was a tad unsure from testing I'd seen so far under load.

 

Beautiful build mate. Didn't notice this thread at all before now.

 

Haha cheers bud, gettin there slowly, 400 posts later!
It is a bit silly, a bit of a pain to get my fingers in places i.e fan headers etc. But once done wont have to touch it.
Still no idea what kind of stupidity unfolds with the GPU loop/tubing..

and now I've got thoughts of putting another rad in there/possibly make use of an old D5 I found in storage.
(facepalm)

 

Got a nice window next 24hrs to do some temps testing & CPU OC’ing, so back onto the build this w/e. Some quick Fans/Pump/Case Temp Tests with this 3080 in, before the CPU which I've started on:



FANS:
- Lower (intake) case/rad fans are +200RPM higher than the top exhaust as per juzz's suggestion early in the piece for some +ve air pressure. Tested with some incense sticks, smoke definitely blowing away from the case

- Set the fan curve to ramp on water temp from 25-35deg, using 20%-80% duty linear curve, targeting avg 1200rpm on full load.

PUMP:
- Decided to see if I can run this pump lower, reduce the tiny bit of noise I can still hear. Impact of pump at 2350rpm (50%) vs 1300rpm (~25%) is +1deg to water temps, +2deg to CPU. I’ll take that over whine/noise, so think I’m happy to settle on 1300rpm.

CASE:
- Won’t be running the case open, but wanted to see how good/poor the case temps & airflow is closed, relative to when open. Clearly not great with the case closed! Half suspected this going into the build reading others experiencing mixed results. Considering all the tempered glass, not the end of the world.

- Whilst I’m ok with closed case CPU temps, think can do better here as AMD’s stock 5900x voltage runs higher than needed.

- Absolutely NOT impressed by the 3080 temps and all the heat it dumps into the case & resultant loop temps; especially when I compare back to the RTX 2080 that was in there.

 

Feel your pain mate - how people run their GPUs on air and live with it (without headphones) is beyond me - the 6900 roars like a banshee to keep itself at 78. Straight back to water ASAP, recommend the same for you too. I looked over your tables and smiled because they're also reflected this end.

Roll on thermal pads, can't bloody wait

 

Insanity innit? And the whole time GPU-z reporting power limited, voltage throttling, and thermal throttling meaning they just want more and more lol.

105 deg?
90 deg? just no.
even 80 deg doesnt sit well with me

and the noise is comical

 

Some CPU overclocking and temps running Cinebench to stress the CPU/VRM.

- Started off with a quick all core OC to get a feel for where CPU limits are. Highest I could go was 4750Mhz all core @ 1.350v. Any higher MHz it would crash, even with more juice. Same goes with lower volts.

- Clocking down to 4700Mhz did allow for much less voltage; and whilst I could get Cinebench to pass @ 1.263v; OCCT large data sets would throw errors everywhere. Best I could get @ 4700 all core was 1.275v.

Some numbers (allowed 5mins between each run @ max fans for coolant temps to settle):


Spreadsheet for you number nerds HERE if easier (will be adding to this).

Obviously ones that are fair to compare are highlighted in red box (top 2 stock just for out of box reference).

Interesting:
- With this RTX 3080 in, IDLE temps are also now much higher vs when I did a quick test back in June with a 2080 over HERE, but CPU load tests are about right (given +2deg temp increase in ambient).

- Stock voltage inefficiency on AMD 5000 chips already well documented, but all core OC temps/power above give me a good idea to compare playing around with the CPU curve now.

This Crosshair VIII Formula BIOS doesn’t have the DOS (Dynamic OC Switcher) the Dark Hero and Extreme mobo’s do- which allows auto switching between all core OC & PBO based on load; which for my purpose I think is the best of both worlds when OC’ing these Ryzen chips.

I spent a bit of time on ClockTuner on a 3700x years ago so about to start with HYDRA now.

Anyone played with it before I go down the rabbit hole?

 

How come you water-cooled everything except for the GPU?

 

Intention was always to put the GPU on water right from the start (here).

Had a 2080 Ti with block ready to go few pages back, and then found this 3080 for a great price but didnt want to spend another $400 on an EK block; especially with next gen GPU's around the corner.

Looked at the Bykski and Barrows 3080 blocks for this Gigabyte Master- price is reasonable but I do like the EK blocks; sadly they're impossible to find 2nd hand for these Gigabyte Masters/Xtremes. By pure chance I found one in New Zealand (Auckland), and have a friend arriving tomorrow and he's done me a massive favour picking it up and bringing here.

In the end its given me some time to test the CPU/VRM and rest of the loop on its own, before adding the GPU & dumping more heat into loop.

Done a lot of testing last few days! Now I've got delusions of grandeur to rip the whole thing apart, put in another 360 rad and a D5 pump. Almost turning into jjjc_93's worklog thread

As much as I'd like another rad and a D5 in there, think its wise to leave till later and just get the gpu looped in first. If not happy with temps will reconsider then.

Happy to hear any thoughts/criticisms

 

Ok think I'm done playing with HYDRA. Overall impression- its excellent, far more polished than I remember CTR to be. Takes a bit of time to run CPU curve offset tests (~4hrs)- ran all tests (single core, multi & PBO); and spent a good chunk of yesterday testing/breaking/tweaking.

Given this 5900x does 4.7Ghz all core @ 1.275v, decided to use that as my base; targeting max ~70deg 100% load temps / max 250w PPT.

Should mention going into this:
- Never been a fan of optimisation other than for a bit of data & stress testing; as like all systems in life- over optimised (curve fitted) models work fine when optimised for conditions a, b and c... But then all of a sudden break/fail/crash when scenario y or z pops up. So have left some on the table re frequencies/voltage for stability. Loose pants fit all

Here's what HYDRA spat out (+200Mhz Boost Override on top of these numbers):



After some tweaking of CO (and 50 restarts later!), settled on these profiles all stable in AIDA, Cinebench, OCCT & 3DMark/no errors. OCCT gave me the most headaches with errors during testing.



Voltage a little higher than the 4.7Ghz 1.275v all core, but still well within & below AMD default. Increased voltage needed here for PBO2 otherwise HYDRA would simply deactivate, and some random PC crashes once PBO kicked in at these frequences. Happy with all cores boosting > 5Ghz, and the switching between all core & single core profiles is awesome, seamless.



With all core on HYDRA:
- Real voltage is actually 1.27v accounting for Vdroop,
- Maxing the CPU out at ~210w / 78deg ( Tad high temp wise but I'm ok given won't be anywhere near this in day-to-day/gaming)
- Water temp 30 degrees.



Running last set of real benchmarks now (games) on temps/power/performance between stock vs factory PBO vs HYDRA OC.

Not expecting huge gains in terms of performance tbh, but hopefully half decent improvements in temps & some baseline numbers before the GPU sees water tomorrow/Thur.

 

Did you see he has released HYDRA 1.2D? (Or am I reading it wrong and it's not out yet?)

HYDRA 1.2D PRO - August 31 (BIG update) | Yuri Bubliy on Patreon

As we share the exact same mobo I'd be interested to see your bios setup for factory PBO, I fiddled with my CPU so long ago I can't remember anything!

Might give HYDRA a go, I presume you have to join the Early Access tier for at least a month to get access?

 

Yeh did see that thanks bud.

Caught up to date on version log over the past week. The move to Hydra v1.2x has predominantly been for RDNA2 overclocking (V/F curve), and preparation for Zen 4. Did sign up only for a month, but went back to last ver of 1.1 (1.1H3) as don't have any use for all the RDNA2 stuff, and alotta guys still raving on about 1.1H3 stability.

Will move this over to Hydra thread (hope you dont mind!), shoot you mobo settings there

 

Final set of HYDRA OC'ing game benchmarks/test results:

- Just picked 5 games I’ve played in the last 12mo that have built in benchmarks (Assetto Corza Competitzione I created a benchmark as it doesnt have one built in. Not perfect but reproducible), as well as some synthetics.
- All benchmarks maxed out settings @ 4k, no quality option turned off. Settings used in SHEET. Ran each test 3 times, took avg.
- Included some Ray Tracing and DLSS for CP2077 and ACC
- I am not a reviewer
- Blue: Stock Ryzen 5900x 3.70Ghz / 4.95Ghz PBO (105w TDP, ~145w max)
- Orange: Hydra OC 5900x 4.75Ghz All Core / 5.15Ghz PBO (targeting ~200w PPT) - all core voltage profile settled at 1.325v, Single thread max 1.385v

As expected, no improvements on the GPU front.
At 4K I'm GPU bound, GPU numbers (power, temps etc in 2nd image) also reflect this. Obvious 1080p runs would've shown improvements here but since I'm never on anything < 4k was pointless for me.



On the CPU, temp and power improvements do however look healthy:


- CPU Temps: ↓ ~10 deg on avg (15%)
- CPU Power: ↓ ~20w on avg (18%)

(only looking at gaming. In Cinebench/AIDA numbers are obviously higher as all core OC using much more power)

Rest of the loop & mobo temps have also improved:


- VRM: ↓ 1 deg
- DRAM: avg ↓ 2-3 deg drop
- Chipset: ↓ 2 deg drop
- Water: ↓ ~1 deg drop

All in all with the OC seeing ~10deg drop in CPU temps, couple of degrees in VRM/DRAM/Chipset, for same single core performance, but better all core.

For the stat nazi's, SHEET updated



I did actually manage to get all core stable @ 4.8Ghz + PBO with LLC = Level 4, but temps far too high for little gain

That wraps up some boring CPU number crunching, got some probes to quickly check some inlet/outlet temps (really should've done this before all this testing hah but they just arrived today). Then onto GPU.

Sorry for the long/gnarly posts!

 

Ok quick update folks, sorry have been quite busy/just got around to this.

Re-padded/pasted the GB 3080 Master with the EK waterblock and plumbed in:

Stock:


Repadded with Gelid Ultimate pads & Extreme paste:

(yes I removed the Gelid plastic hah)


(forgive the poor camera lighting, gpu block massively overexposed!)

You're probably thinking wtf- soft tubing!?? Given this is a secondary rig and I know I'll be swapping out/potentially trialling RDNA3/4080/4090 soon, figured this was the best of all evils; rather than having to redo hard lines again.

Was rather pain free to install too, not having to flush the loop. Just laid it on its side (distro flat on table side/ports facing north).

I will indeed do a hard line once I've finalised what GPU goes in here (or whether the 3080 suffices).

To say I'm happy with temps is understatement , some bloody huge improvements over stock HSF. I'll redo all the above benchmarks again this weekend but for now, a quick run of 3DMark Timespy:

Ambient Temp: 18deg, on stock 370w GPU power draw (no GPU overclocking/undervolting etc yet).

BEFORE = CPU & mobo plumbed in, stock GPU HSF on air
AFTER = CPU, mobo & GPU EK wb plumbed in

BEFORE/AFTER:
CPU Temp Avg: 55.8 deg / 57.3 deg
CPU Temp Max: 82.3 / 83.1
GPU Temp Avg: 64.9 / 40.1
GPU Temp Max: 76.6 / 49.8
GPU Hot Spot Temp Avg: 82.9 / 53.5
GPU Hot Spot Temp Max: 99.3 / 64.5
GPU VRAM Temp Avg: 78.3 / 43.1
GPU VRAM Temp Max: 90.0 / 50.0
Water Temp: 32 / 31



and peacefully quiet EDIT: Just checked those numbers again on another run, yep, 40deg drop in mem temps

 

Ohh those temps mate

Looks the goods. I like the blend or hard and soft tubing - coolant brings it together really tidily, looks great from here!