7800X3D new build, help please?

I had a display driver crash earlier today while watching youtube, I don't know if it is related to the new RAM and timings. I'm running OCCT now to check stability.

Also, damn, the temps on the 7800X3D are not what I was expecting. I'm used to my old i7 5960X which, overclocked, was probably drawing more power than the RTX 4090. I have this system setup on the motherboard box right now for testing, using a very old Hyper212 with some thermal paste only, and while running OCCT its sitting at only 58C. WTF.
 
Ah FFS. Something has gone terribly wrong. I loaded up EXPO and everything seemed fine. Installed OCCT and ran the memory test and it passed a one hour test no problem.

I figured it was time for some more tinkering so I had a play with precision boost, set a negative curve of 20, restarted and when I ran OCCT the system just shut down. WTF, I haven't seen this behaviour before, even with the old RAM installed.

It seems like the OCP is triggering on the PSU. I reset everything back to default (load optimised defaults in BIOS, then loaded EXPO again) and the same thing happened. It will not run EXPO any more. I dropped the ram speed all the way down to 5200 and it shut down again after a little longer.

I wouldn't be surprised if it was the PSU. It is an old Silverstone 1KW I have lying around and can deliver 80A on the 12V rail (when new) but I was always a little suspicious about just how capable the silverstone PSUs were so swapped everything out for Seasonic. It might be time to assemble this thing in its home with the Prime 1200W Platinum.
 
Deimos said:
Ah FFS. Something has gone terribly wrong. I loaded up EXPO and everything seemed fine. Installed OCCT and ran the memory test and it passed a one hour test no problem.

I figured it was time for some more tinkering so I had a play with precision boost, set a negative curve of 20, restarted and when I ran OCCT the system just shut down. WTF, I haven't seen this behaviour before, even with the old RAM installed.

It seems like the OCP is triggering on the PSU. I reset everything back to default (load optimised defaults in BIOS, then loaded EXPO again) and the same thing happened. It will not run EXPO any more. I dropped the ram speed all the way down to 5200 and it shut down again after a little longer.

I wouldn't be surprised if it was the PSU. It is an old Silverstone 1KW I have lying around and can deliver 80A on the 12V rail (when new) but I was always a little suspicious about just how capable the silverstone PSUs were so swapped everything out for Seasonic. It might be time to assemble this thing in its home with the Prime 1200W Platinum.
Click to expand...
7950x, ASUS X670E-E and a ASUS 4070 OC run comfortably on a Seasonic 850. Full tilt Prime95 I'm seeing 425 Watt draw at the UPS, gaming load maybe another 130 peak for around 550ish in Doom at 2k 2560x1440. Furmark @2k adds 210 Watts over desktop so that's probably peak for this gpu. Running around the last mission I'm seeing maybe 230W average, 300W peak at the UPS running Doom at 2k.
 
Last edited:
Deimos said:
Ah FFS. Something has gone terribly wrong. I loaded up EXPO and everything seemed fine. Installed OCCT and ran the memory test and it passed a one hour test no problem.

I figured it was time for some more tinkering so I had a play with precision boost, set a negative curve of 20, restarted and when I ran OCCT the system just shut down. WTF, I haven't seen this behaviour before, even with the old RAM installed.

It seems like the OCP is triggering on the PSU. I reset everything back to default (load optimised defaults in BIOS, then loaded EXPO again) and the same thing happened. It will not run EXPO any more. I dropped the ram speed all the way down to 5200 and it shut down again after a little longer.

I wouldn't be surprised if it was the PSU. It is an old Silverstone 1KW I have lying around and can deliver 80A on the 12V rail (when new) but I was always a little suspicious about just how capable the silverstone PSUs were so swapped everything out for Seasonic. It might be time to assemble this thing in its home with the Prime 1200W Platinum.
Click to expand...

Time to put that bitch on the pizza box and test! I know older PSUs do NOT play well with the 4090, the 7800X3D has integrated graphics so you can try using the onboard and see if she boots up again.
 
Deimos said:
Ah FFS. Something has gone terribly wrong. I loaded up EXPO and everything seemed fine. Installed OCCT and ran the memory test and it passed a one hour test no problem.

I figured it was time for some more tinkering so I had a play with precision boost, set a negative curve of 20, restarted and when I ran OCCT the system just shut down. WTF, I haven't seen this behaviour before, even with the old RAM installed.

It seems like the OCP is triggering on the PSU. I reset everything back to default (load optimised defaults in BIOS, then loaded EXPO again) and the same thing happened. It will not run EXPO any more. I dropped the ram speed all the way down to 5200 and it shut down again after a little longer.

I wouldn't be surprised if it was the PSU. It is an old Silverstone 1KW I have lying around and can deliver 80A on the 12V rail (when new) but I was always a little suspicious about just how capable the silverstone PSUs were so swapped everything out for Seasonic. It might be time to assemble this thing in its home with the Prime 1200W Platinum.
Click to expand...


Well the only way you can rule out the PSU is by trying a known good one in this case. You might also have to swap them back and forth a few times to rule out a transient issue. Have you been checking your SOC voltage just in case by the way? I kind of think that a high quality 850W should be able to handle a 4090 and 7800X3D though. The 7800X3D sips watts and the 4090 doesn't use much more than 3090. I had a 3080 Ti with a 5950X running on an 850W Seasonic just fine for quite a while. Right now I'm using an EVGA P2 1200W, and it's handling it just fine.

FWIW I assume your G.Skill sticks are Hynix and set at 1.4 volts out of the box. I've heard that's fine, but I set mine to 1.39v just in case because I've been told that 1.4 is sort of at the edge of "not quite good for them", even though that's what the XMP/Expo/whatever on mine was as well.

https://www.patreon.com/posts/low-effort-rank-77403831

You can also give these timings a shot, if you want. I'm running them on my own 2x32 and it's been working great. Pretty stable. I set the SoC to 1.26 and the ram voltage to 1.39.
 
StoleMyOwnCar said:
You can also give these timings a shot, if you want. I'm running them on my own 2x32 and it's been working great. Pretty stable. I set the SoC to 1.26 and the ram voltage to 1.39.
Click to expand...

The biggest culprit of instability I've found so far is one nobody talks about, VDDP. Its the on-board I/O controller for the DDR5 modules, and anything over 1.000V is very bad for it. Standard AMD voltage is 800mv. I run mine at 850mv with overclocked RAM and it helped tremendously, for example my Asus board was pushing it to match VDD/VDDIO at 1.4V when left on auto and it was unable to even POST most of the time.
 
After playing for a few months with the 7900x CPU. the most stable for me (with an overclock for best performance) is the Asus AI overclock and PB set to 0. EXPO2 profile and of course the latest BIOS. That's it.
 
LigTasm said:
The biggest culprit of instability I've found so far is one nobody talks about, VDDP. Its the on-board I/O controller for the DDR5 modules, and anything over 1.000V is very bad for it. Standard AMD voltage is 800mv. I run mine at 850mv with overclocked RAM and it helped tremendously, for example my Asus board was pushing it to match VDD/VDDIO at 1.4V when left on auto and it was unable to even POST most of the time.
Click to expand...


Interesting. My computer has been incredibly stable, but I wonder what my VDDP is set at. I let MSI control it automatically when I tweaked everything earlier. I'll need to look at that when I get home.
 
StoleMyOwnCar said:
Interesting. My computer has been incredibly stable, but I wonder what my VDDP is set at. I let MSI control it automatically when I tweaked everything earlier. I'll need to look at that when I get home.
Click to expand...

I only know for ASRock and Asus, I did not check the MSI and Gigabyte AM5 boards I've built with for other people recently. ASRock was sending about 1.100V which is still high. If its working good I would probably leave it alone lol
 
I really appreciate all you guys weighing in. I read all your posts and I'm way ahead of you. I have been checking voltages thoroughly since I got the system becuase I was aware of the burnout issues with the 7800X3D when it came out. All the voltages are running within recommended settings.
I took note of the VDD comment though, and have dropped it to 1.395V. IO voltage was bumped up to 1.2V when I installed the new ram kit and it worked fine initially.

I'm 99% sure the shutdown was PSU related. I have another identical silverstone which I just switched out and I'm back at EXPO settings running OCCT. It hasn't powered off yet. While this is testing I'm preparing to switch it out for my Seasonic. The only issue is that I got cablemod cables and the ones I have aren't what is needed on this new system so I have to pull out the old cables and test them to make sure I have the right cables, I have a few Seasonic PSUs and while the cables fit, they have slightly different pinouts between them.

And yes, I have this on a box for testing. I would have to be a little insane to build my custom WC loop without testing the hardware first...
 

Attachments

  • 20231113_140843.jpg
    20231113_140843.jpg
    421.8 KB · Views: 0
  • 20231116_102424.jpg
    20231116_102424.jpg
    921.7 KB · Views: 0
Deimos said:
I have this system setup on the motherboard box right now for testing, using a very old Hyper212 with some thermal paste only, and while running OCCT its sitting at only 58C. WTF.
Click to expand...
What exact part of OCCT are you running? (and with which settings?). I would like to run the same thing and see what my temps are for my 7800X3D, with a Noctua NH-C14S.
 
OK so that was unexpected. The system restarted while running OCCT, an improvement over the complete shutdowns, so I switched to the seasonic after making sure the pinouts were correct. I suspect these Silverstone PSUs have moisture damage.

I hooked up the Seasonic and the system POSTs and boots much faster now. Am I crazy? Running OCCT now. I'll report back later.
 
chameleoneel said:
What exact part of OCCT are you running? (and with which settings?). I would like to run the same thing and see what my temps are for my 7800X3D, with a Noctua NH-C14S.
Click to expand...
I'm running the memory stability test at the default 80% setting. CPU right now is below 60C.
 
Deimos said:
OK so that was unexpected. The system restarted while running OCCT, an improvement over the complete shutdowns, so I switched to the seasonic after making sure the pinouts were correct.

I hooked up the Seasonic and the system POSTs and boots much faster now. Am I crazy? Running OCCT now. I'll report back later.
Click to expand...
Could have been too much V-droop on those Seasonics Silverstones. Especially since the issue first happened after setting -20 in Curve Optimizer.

My own 7800X3D actually won't work below -20 with Curve Optimizer, but does work perfectly at -20. I have a Corsair SF750 power supply. Which is generally regarded to be excellent quality/performance.
 
Last edited:
chameleoneel said:
Could have been too much V-droop on those Seasonics. Especially since the issue first happened after setting -20 in Curve Optimizer.

My own 7800X3D actually won't work below -20 with Curve Optimizer, but does work perfectly at -20. I have a Corsair SF750 power supply. Which is generally regarded to be excellent quality/performance.
Click to expand...
Blasphemy!

Too much v-droop on the SILVERSTONES.
 
Well, the EXPO settings are definitely not stable. I don't really have time right now to continue testing and tweaking so I'm just going to leave it at 5600 for now and finish my assignment.
 
Deimos said:
Well, the EXPO settings are definitely not stable. I don't really have time right now to continue testing and tweaking so I'm just going to leave it at 5600 for now and finish my assignment.
Click to expand...
I run buildzoid's timings which he provided to Hardware Unboxed, as a first round of tuning:


View: https://youtu.be/MOatIQuQo3s?si=6Stb8gr2eSrYmeIp&t=66

And here are some updated timings from Buildzoid. I will post my final timings later. But, they aren't far off.
 

Attachments

  • Zen4timings.png
    Zen4timings.png
    137.9 KB · Views: 0
Deimos said:
Well, the EXPO settings are definitely not stable. I don't really have time right now to continue testing and tweaking so I'm just going to leave it at 5600 for now and finish my assignment.
Click to expand...

6000 EXPO on a dual rank kit is difficult, you need to get real lucky. Most of that kit you bought are A-die, but some are M-die - you can look at the lot code on the module and the letter it ends with will tell you. Either way, the buildzoid timings should work, just set the tRDRDSD/DD and tWRWRSD/DD to 8 instead of auto.

Like this is a quick way to get my system running today when I swapped back to this board and I haven't had time yet to tune it fully but it runs 100% stable. One thing I've learned with these rigs through half a dozen boards and CPUs and even more memory kits is that less is more on the voltage. It gets more unstable the higher you go, and its not the usual crashes or bluescreens but funky stuff like network adapters dropping out or refusing to load pages, or USB devices randomly disappearing.

ZenTimings_Screenshot.png
 
Last edited:
LigTasm said:
6000 EXPO on a dual rank kit is difficult, you need to get real lucky. Most of that kit you bought are A-die, but some are M-die - you can look at the lot code on the module and the letter it ends with will tell you. Either way, the buildzoid timings should work, just set the tRDRDSD/DD and tWRWRSD/DD to 8 instead of auto.

Like this is a quick way to get my system running today when I swapped back to this board and I haven't had time yet to tune it fully but it runs 100% stable. One thing I've learned with these rigs through half a dozen boards and CPUs and even more memory kits is that less is more on the voltage. It gets more unstable the higher you go, and its not the usual crashes or bluescreens but funky stuff like network adapters dropping out or refusing to load pages, or USB devices randomly disappearing.

View attachment 613720
Click to expand...

Interesting, so what my MSI board did wasn't quite set it up to parity with the memory voltage, but it did definitely raise it well over 1.0V.

1700129539364.png


It set it to.... uh... 1.1484.

Now, noteworthily I do have my context restore turned off, so my board does its training every time I boot. So maybe that helps. I don't reboot much and it only takes about 1-1.5 minutes anyway, so I don't really care.
 
StoleMyOwnCar said:
Interesting, so what my MSI board did wasn't quite set it up to parity with the memory voltage, but it did definitely raise it well over 1.0V.

It set it to.... uh... 1.1484.

Now, noteworthily I do have my context restore turned off, so my board does its training every time I boot. So maybe that helps. I don't reboot much and it only takes about 1-1.5 minutes anyway, so I don't really care.
Click to expand...

If it works, it works. Although you may want to change your tRC to 70, I'm not sure of MSI boards but generally most BIOS will ignore this if it isn't the same as tRP+tRAS.
 
LigTasm said:
If it works, it works. Although you may want to change your tRC to 70, I'm not sure of MSI boards but generally most BIOS will ignore this if it isn't the same as tRP+tRAS.
Click to expand...

Does it default to some stock or really high value, or does it just set it to 70 itself?
 
StoleMyOwnCar said:
Does it default to some stock or really high value, or does it just set it to 70 itself?
Click to expand...

I don't know, probably just goes with a default value which I think is 96. For a long time people were setting the tRAS to 28 because Intel can do this, but it turns out Ryzen will not respect it and uses 30. I think buildzoid made a note of that in his page there. Anyways, its nitpicky crap, if its working good for you I wouldn't worry about saving the 3 nanoseconds or whatever lol.
 
Enable Power Down Mode then report back. Ever since agesa 1.0.0.7 power down has been pretty much required for stable memory.
 
OK so this is wild. Did any of you guys know that RGB control software can brick your RAM? I found out the hard way.
https://www.gskill.us/forum/forum/p...-z-rgb/166671-trident-z-neo-rgb-corrupted-spd Check out post number 4. Holy shit.

RGB control on memory sticks is handled by the spd module. Any time you tell the spd module to do something with the RGB, it writes to the spd flash. If anything goes wrong, the spd gets corrupted and the ram no longer works. There is some software (Thaiphoon burner) that can reflash the spd, but it only works on DDR4.

I bricked a stick of ram while messing around with open RGB. The wild thing is, I removed all other RGB software first to make sure they wouldn't conflict and it was an openRGB plugin, rather than the base software that caused it. Had to RMA my ram (again). Sigh.
 
I ended up getting a refund on the 64GB kit and going for 32GB. It has been running flawlessly at 6000Mhz EXPO, although I haven't done any actual testing yet. I have to submit a draft of my dissertation by tomorrow so have been working to get that finished. I did get around to installing it in my case. I'll post pics when I get the lighting finished, I'm just waiting on a cable to connect the case RGB to my motherboard. OpenRGB is pretty cool with being able to set the colour to the CPU and GPU temps. I haven't been able to get fan control working though.

The CPU and GPU blocks are amazing as always. The CPU has occasional spikes up to 85C+ but that seems to be unavoidable with the 7800X3D. Temps overall are very low with low fan speeds.
 
RGB link cable for my Phanteks Enthoo elite went missing, new one should be arriving today and I'll post some pics once it arrives.

First try at overclocking I have just put a mild negative 10 curve all core to see what happens and did a one hour stress test. Temps dropped about 4C but I'm not sure if the boost clocks have been effected. HW Monitor reported clock spikes of upto 5619 Mhz (I did see 5900mhz prior to tweaking) I have no idea what that means. I don't think that is being sustained at all.

Coming frrom an old i7 5960X and this system is just wild. The performance uplift is obscene from my previous system and with this -10 HWMonitor is reporting a max of 80W during stress test and my PC is cold to the touch. these three massive radiators don't even seem to notice the CPU under full load. I don't know how intel can justify the crazy wattage on their CPUs when this is the competition.

I setup OpenRGB to control the RGB on my GPU and CPU blocks to indicate temps which is real nice but onfurtunately I can't seem to get OpenRGB to control fans connected to the motherboard. Not sure what is going on there.
 
I'm finishing up with an assignment before I move on to the next one and have a little time for tweaking. I saw something on yt today about DDR5 'high-voltage mode' and the guy was running up to 1.8v, like holy shit man, thats DDR2 levels of voltage.

I pushed mine up to 1.5v, 6200 CL28, 101BCLK and 2033 infinity fabric for shits and giggles on the m-die hynix and it boots. Running OCCT stability test right now. Ram temps are sitting at 62-63C while testing which seems acceptable I guess?
1702991891469.png


I also had a go at the PBO and did a minus 30 all cores but I have no idea if it is doing anything. PBO is in three different places in this Gigabyte bios, and turning on one doesn't turn on the others. No idea WTF I'm supposed to be doing but I'm seeing a consistent 5.1Ghz all cores and temps seem pretty good most of the time. HWMonitor is reporting occassional boost clocks over 6Ghz but I have no idea what that actually means either. Could be for a nano-second for all I know.
 
Last edited:
NOPE, Hard fail. I've gone back to my previous settings with the BCLK set to 101Mhz to see how that goes. Baby steps I guess.
 
Deimos said:
NOPE, Hard fail. I've gone back to my previous settings with the BCLK set to 101Mhz to see how that goes. Baby steps I guess.
Click to expand...

I don't remember if the Aorus Elite AX has external clockgen, but its best to leave that alone on the X3D chips. Also, try and get to 2067mhz for your FCLK. Best performance is found at FLCK=MCLK/1.5
 
LigTasm said:
I don't remember if the Aorus Elite AX has external clockgen, but its best to leave that alone on the X3D chips. Also, try and get to 2067mhz for your FCLK. Best performance is found at FLCK=MCLK/1.5
Click to expand...
Tweaking this system just doesn't seem worth it.

Buildzoid mentioned that FCLK gives best performance at 2033 which I have been running. Here are my results so far on Time Spy with these settings:
101 BCLK, 2067 FCLK: 30486
101 BCLK, 2033 FCLK: 30620
Auto BCLK, 2033 FCLK: 30638

Auto BCLK + 2033 FCLK is definitely optimum based on Time Spy but the difference is like 0.1 - 0.01%.

The only real difference I have noticed with PBO is a slight drop in temps.

I'm running 'no effort' ram timings from buildzoid on this 32GB Hynix-M kit. Tweaking and overclocking ram and timings just results in instability.

Even though I'm watercooling my 4090, I dialed in settings used on review websites to overclock air-cooled cards; +225 on the core and +1250 on the mem with max power limit. GPU and mem temps are around 55-60C

I've gone from 28227 on time spy to my current best of 30638. I am curious about how someone with the same CPU+GPU is able to get 31000 but not that curious. I think I'm good.
 
Overclocking AMD is a waste of time.
Also overclocking the 4090 is whatever. Yes you can get maybe 2-4 fps higher at 4K but the amount of heat the card puts out is not worth it. If you are in a loop then maybe. I like my default setup. Plays everything and does everything whilst being stable.

I know not a very hard post but seriously, the amount of time I have spent chasing benches is no longer my priority.
 
I said I was good but I couldn't resist after I couldn't reproduce my best score.

I ended up playing with a whole lot of PBO settings which all did nothing or made things worse. The only setting that made any difference was the all core negative offset. Everything esle set to auto seems to be best with the exeption of the PBO liimit which I set to motherboard. I ended up setting offset to -39. The PBO enhancement seemed pointless. Also, an extra 1mhz on the BCLK is beneficial and I'm now getting a consistent 5100mhz all core when combined with the -39 offset resulting in a PB of 30 716 in time spy.

I might keep playing if I feel the urge...
 
Deimos said:
Tweaking this system just doesn't seem worth it.

Buildzoid mentioned that FCLK gives best performance at 2033 which I have been running. Here are my results so far on Time Spy with these settings:
101 BCLK, 2067 FCLK: 30486
101 BCLK, 2033 FCLK: 30620
Auto BCLK, 2033 FCLK: 30638

Auto BCLK + 2033 FCLK is definitely optimum based on Time Spy but the difference is like 0.1 - 0.01%.

The only real difference I have noticed with PBO is a slight drop in temps.

I'm running 'no effort' ram timings from buildzoid on this 32GB Hynix-M kit. Tweaking and overclocking ram and timings just results in instability.

Even though I'm watercooling my 4090, I dialed in settings used on review websites to overclock air-cooled cards; +225 on the core and +1250 on the mem with max power limit. GPU and mem temps are around 55-60C

I've gone from 28227 on time spy to my current best of 30638. I am curious about how someone with the same CPU+GPU is able to get 31000 but not that curious. I think I'm good.
Click to expand...

Anything over 55C on DDR5 is usually unstable. I bet if you ran karhu you’d only make it 30 seconds. 1.5v is a LOT for 6200, that’s like 7600+ voltage. CL is the big culprit, trying to run 28 is basically useless because the refresh time is too slow anyways. Drop it to 32 and you’ll be able to run 1.35V and cut 20C of the temps.
 
-39 offset works? For me even going from -15 to -20 led to slower performance in Cyberpunk 2077. Can you try this and let me know how you bench maxed out in Cyberpunk?

Also for some reason I thought -30 was max offset.
 
-39 was unstable. I was able to get a timespy score but soon after the system hard rebooted.
 
I've found that offset is very much a mix of silicon lottery, AEGSA version, and motherboard allowances. It's been a long while since I last looked at the more "advanced" portions of my motherboard's UEFI BIOS options, but if memory serves the X670E Ace allowed for an extra wide positive or negative offset option. Other boards may also go beyond the basic Ryzen Master "spec."
 
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.
 
You must log in or register to reply here.
Back
Top