7800X3D new build, help please?

Deimos said:
I'm planning on rebuilding this thing for a few reasons:
Even after I did my best to clean my loop, I'm seeing debris in my CPU block, so it needs a clean.
I've heard that GPU PCBs are cracking so I bought a new vert PCIe4.0 GPU mount.
I want to update my storage while I have this thing apart so I bought a 4TB 990Pro to install while I have it in pieces.

Has anyone lapped a 7800X3D? I'm tempted to have a crack at it while I have it apart to take maybe 1mm of the height of it. I'm not particularly impressed with the temps on this CPU and how quickly it climbs to max temp. I've never seen a CPU temp spike so quickly in this loop and I suspect it is the super thick IHS to blame. My GPU sits at around 50C while gaming but fan speed is based on CPU temps. This usually isn't an issue for gaming as CPU temps remain fairly low but for something dumb like looking through youtube the temps spike and fans start ramping which is absurd.
Click to expand...

AM5 CPUs almost always shoot right up to max temp under any kind of load. ALL of them do it. It's a deliberate design decision by AMD. Der Bauer did a whole thing on it back in early AM5 release days. Lapping the CPU die is unlikely to make much difference for you, but he had great success de-lidding the CPU and going with direct-die cooling. De-lidding a non-X3D AM5 processor is pretty straightforward, but I am unsure how much of a complication the 3D cache adds to the process (if any). Might be worth looking into if this is a thing you are very intent on fighting without sacrificing any performance.

Alternative (and what I do): Use PBO to limit the CPU to 105W or 65W. I limit my 7950X to 105W, and the performance difference is only 5% give or take a few points depending on the task, but temps and power draw sit a bunch lower. I know you can't use PBO to OVERCLOCK an X3D CPU, but I have not heard if it can be used to power limit those CPUs or not.
 
Jays2cents lapped almost 1mm off a 7950X3D and saw almost 10C temp difference. Thats not insignificant.

I usually see about 85W under load so I don't think a limit of 105W will do anything.

There is no way I would delid a CPU again. I killed a few xeons doing this. The last successful delid I did lasted a few years until I tried to rebuild it and without even removing teh CPU from the board it refused to work again. It started out not detecting all the ram and the more I tried to fix it, the more it failed. My last attempt was a i7 7700k which was a massive, expensive failure (dead CPU).

Lapping is fairly low risk for a potential 10C improvement. That sounds tempting to me.
 
I started the rebuild on Friday (NZ time) and it still isn't done. I have a warning if anyone is contemplating doing this.

I started with a goal of a 2mm 'shave'. 120 grit sand paper wasn't cutting it so I went back to the hardware store and got a few pieces of 80 grit. There really is no choice, you have to wet sand if you are doing this by hand otherwise the CPU gets very hot very fast. Also, sanding 2mm of material by hand is a BITCH. I gave up after almost 1mm and started the slow process of polishing. Even after a great deal of effort it is impossible to get all the scratches out and my polishing kit gave out anyway.

I got more or less 1mm shaved off. The CPU thickness started out at 5.80mm and I got just under 4.8 at the end.

This is where the warning starts. It turns out that the retention mechanism is a chunky boy and after 1mm shaving the top of the IHS is pretty much level with the retention mechanism. I didn't realise this when I started so I'm damn lucky I stopped when I did. The water block is making good contact with the IHS and I booted it up without filling the loop to make sure I hadn't killed the CPU. I saw 50C with the water block only before I shut it off.

I also noticed some discolouration on the wafer which is why I booted it without water. I was worried that I had killed it, thankfully everything checks out and all the ram is detected too.

Another thing I would warn about is cleaning. You have to clean the CPU. I used our water flosser filled with isopropyl alcohol and it did the trick.

While I was at it I noticed that my loop was a little green so I have flushed the rads with cleaning vinegar, as well as pulling apart both blocks for cleaning. What a mission.

I have to run almost all new tubing so I'll probably have to spend a week or so bending to finish it off.

I'll let everyone know what kind of temps I get.
 
Finally finished the rebuild. I'm reasonably happy with the new bends. The only problem is that I did something really stupid. I cleaned everything and flushed the radiators for hours. After I got done with the bending I contemplated flushing the loop once more for good measure but figured I had done enough flushing. I was totally wrong and now I have copper oxide all through my loop from the radiators. A total waste of $50 worth of EK coolant. Fuck. Also, this monster loop takes 2.5 litres so I didn't even have enough to fill it as I only bought 2 litres.

Anyway, it looks like it is at least 6-8C cooler after lapping the CPU while running OCCT. I'll do some testing and see what kind of spikes I get.
20240301_161428.jpg
 
Without waiting for the loop to warm up I'm seeing a peak of 77C while running prime95. I don't expect to see a big difference even after a few hours as we're only talking about cooling 90W. with three massive radiators I'd expect to see maybe 1-2C more after running this for a few hours.

I'm calling this a success but I'm not so sure about saying that lapping a whole 1mm was worth the effort.
 
Deimos said:
Without waiting for the loop to warm up I'm seeing a peak of 77C while running prime95. I don't expect to see a big difference even after a few hours as we're only talking about cooling 90W. with three massive radiators I'd expect to see maybe 1-2C more after running this for a few hours.

I'm calling this a success but I'm not so sure about saying that lapping a whole 1mm was worth the effort.
Click to expand...
Help me out and tell me your P95 temp before lapping the CPU and flushing your loop?
 
I didn't check it sorry, the rebuild was done in a rush as I'm back at university this week. I was seeing peaks of 89C before lapping. The most I've seen so far is 82C. I also feel like there is still room for improvement if I lap the water block as well. That might be something I will consider when I have it apart for the next clean.

As an aside, no matter what I do I can't get any better clock speeds tinkering with PBO, however if I simply set it to 'enabled' I get consistently higher clock speeds by about 50mhz or so.
 
I just checked my 3dmark results and the difference is -3C +22Mhz CPU and -4C +38Mhz GPU +45Mhz mem in timespy.

https://www.3dmark.com/compare/spy/46123499/spy/45543407#

This is completely unscientific though. I'm not entirely sure if the BIOS and overclock settings are the same between runs as the BIOS in my mobo reset after removing the CPU. I have no idea why that is a feature, its pretty annoying having to enter all my sub-timings again.
 
I've been running prime95 all night and it peaked at 83C over night. It is sitting on average of 77-78C.

The reason I wanted to lap the CPU was the wild occassional spikes. The way I set up my fan curve meant that my fans were ramping while doing mundane shit like browsing youtube. The lapping defintiely reduces the temp spikes.

The loop has cleared somewhat but there is still some cloudy water in the top of the res. I'm not sure what is going on there.

20240302_095549.jpg
 
Last edited:
Deimos said:
I've been running prime95 all night and it peaked at 83C over night. It is sitting on average of 77-78C.

The reason I wanted to lap the CPU was the wild occassional spikes. The way I set up my fan curve meant that my fans were ramping while doing mundane shit like browsing youtube. The lapping defintiely reduces the temp spikes.
Click to expand...
if you set your fans to ramp up after the spikes it will be much quieter. the heat sink will handle the blips fine.
 
My air cooled 7800x3D machine in Torrent has no "curve" at all. All 5 case fans and two CPU cooler fans are all tied to a single flat 40% PWM "curve" until 85C, then they go to 60% PWM. For gaming and most everything the system is silent. If I run an encode or some other 100% multicore load is the only time I hear the fans ramp up at all.

The worst thing in the world to me is the sound of fans ramping up and down constantly in direct relationship to CPU temp. That jumps around so much these days that unless you get real complicated with FanControl and hysteresis, you're going to hear it.
 
criccio said:
My air cooled 7800x3D machine in Torrent has no "curve" at all. All 5 case fans and two CPU cooler fans are all tied to a single flat 40% PWM "curve" until 85C, then they go to 60% PWM. For gaming and most everything the system is silent. If I run an encode or some other 100% multicore load is the only time I hear the fans ramp up at all.

The worst thing in the world to me is the sound of fans ramping up and down contently in direct relationship to CPU temp. That jumps around so much these days that unless you get real complicated with FanControl and hysteresis, you're going to hear it.
Click to expand...
i do flat 30% until like 2c over whatever the normal spikes are (55 when it spikes to 53) then i set fans to 90% at max temp, like 85. then the fans only ramp up when the cpu is actually working on a sustained load.
 
criccio said:
My air cooled 7800x3D machine in Torrent has no "curve" at all. All 5 case fans and two CPU cooler fans are all tied to a single flat 40% PWM "curve" until 85C, then they go to 60% PWM. For gaming and most everything the system is silent. If I run an encode or some other 100% multicore load is the only time I hear the fans ramp up at all.

The worst thing in the world to me is the sound of fans ramping up and down constantly in direct relationship to CPU temp. That jumps around so much these days that unless you get real complicated with FanControl and hysteresis, you're going to hear it.
Click to expand...

pendragon1 said:
i do flat 30% until like 2c over whatever the normal spikes are (55 when it spikes to 53) then i set fans to 90% at max temp, like 85. then the fans only ramp up when the cpu is actually working on a sustained load.
Click to expand...
Percentages don't say a whole lot, what RPM are you running your fans at?

I'm running my 120mm Vardars at 585rpm and my 140mm Noctuas at 792rpm. Dropping the fans speeds lower than this doesn't seem to make any difference to fan noise. I changed my fan curve to go up gradually by 4% from 65c to 84C which is barely noticable, and a steeper 20% slope from 84C to 95C. So far I haven't seen it go over 80C momentarily during normal use.
 
Deimos said:
Load up HWMonitor and it should tell you fan rpm...
Click to expand...
Its and unreliable factor when all fans are different. PWM percentage is a universal factor. (for the most part)

I also have fans of different RPM's connected to the same hub, TBH i dont' even know which fan the tach signal is being pulled from. It doesn't matter.
 
Is there not an option for setting the curve based on coolant temp if you build your own loop or does it depend on the pump/controller? I know a lot of AIO's have that option, which makes a lot more sense than transient cpu temp given the fan ramp won't immediately cool the cpu as it has to lower the coolant temp first. I have mine set to keep coolant temp between 30-35C, which means fans only ramp on sustained loads and it's a very gradual ramp. 30C vs. 35C coolant only gives roughly a 2-3C temp difference on the cpu itself...
 
bobzdar said:
Is there not an option for setting the curve based on coolant temp if you build your own loop or does it depend on the pump/controller?
Click to expand...
You are correct in that when cooling with water, fans are there to cool the water, not the CPU itself.

As for triggering off water temperature, that is very much so a function of the AIO and its accompanying software (if it has any). Corsair AIO's (at least some of them) through iCUE can report water temp and allow you to base your curves off of that. Command Pro (and possibly Core) have probe connections where you can hook up external sensors and, for example, have one on the exhaust side of your radiator and trigger your curves based on that.
 
I have to finally admin defeat, I have no idea how PBO functions or what settings produce a good result. I messed with every setting and nothing seemed to change. The only setting that had any repeatable effect is curve optimisation.
BCLK tweaking makes the system behave weird. Even just 1mhz up and the system exhibits weird quirks like youtube videos stalling and system hitching, while seemingly stable. It also has very little effect on boost clocks overall apart from single core boosting.

After my latest round of rebuild/lapping -39 all cores seems to be working. I'm seeing consistently higher all core boosting to just under 5Ghz stable, and is 100+Mhz higher than any other setting I tried. Everything else is set to auto for PBO. Also, -39 offset actually results in a higher package temp while boosting.
 
Just wondering if anyone else can verify, I dropped my SOC voltage in BIOS to 1.08v. The bios reports 1.11v but all monitoring tools in windows, including Ryzen Master reports it as 1.2V. It never seems to change either and is the only v that reports exactly the same every time. Every other v is usually a few hundred mV off the bios setting and fluctuates.
 
You must log in or register to reply here.
Back
Top