Overclock.net banner

[OFFICIAL] AMD Radeon RX 7800 XT Owner's Club

1 reading
70K views 233 replies 35 participants last post by  fpsflow  
#1 · (Edited)
Here it is, the 7800 XT Owner's Club!

I just installed my Sapphire Nitro+ Radeon RX 7800 XT and activated AMD SAM (got an Ryzen 5600x on a super old X370 Taichi).

Have somebody tested undervolting the card?

edit:



This is a really good start to undervolt the 7800XT, I will try out things and give feedback here
 
#2 ·
Here it is, the 7800 XT Owner's Club!

I just installed my Sapphire Nitro+ Radeon RX 7800 XT and activated AMD SAM (got an Ryzen 5600x on a super old X370 Taichi).

Have somebody tested undervolting the card?
How is the delta between the GPU Edge Temp and the GPU Hotspot Temp?

I had a 7800 XT Nitro+ (briefly) but ended up returning it to the 30c delta in temps and even a repaste barely made any difference.

In the time I had it though I reduced the voltage to 950mv or so and that helped improve performance a bit with the power limit maxed out.
 
#3 ·
I have my hands on a Sapphire 7800XT reference. My son-in-law was contemplating a prebuild. I suggested we build our own, to save money and have more control over the parts. With a budget in mind, I provided a modest parts list. He bought the parts and had them shipped directly to my home. Sadly, other then building the PC, installing the OS and doing some tuning in the Bios over Thanksgiving day, I will not get the opportunity to benchmark the card, test thermals, overclock/underclock etc...
 
#6 · (Edited)
Can I ask something.
In Adrenaline the GPU reports between 990-1010 mv with the Voltage set to 1150mv? Is this correct?
I have the 7800XT Hellhound. Have only touched the fancurve and max freq to 2650 instead of 2560 something.
Is the MV slider just a max setting or some kind of offset, I dont quite figure.

I see people on reddit and elsewhere state they run at XXXmv but I cant figure if they mean the slider or actual value. For instance many users say to set the slider to 950mv but that gives me <850mv actual.

And another question:
Anyone else get hardlockups when playing games?
Aswell as random Driver crashing? Its crashed when viewing movies through madvr in MPC-HC. Aswell as hardlocked requiring a reboot when playing Cyberpunk for some time. Its completely random. Does not happen with previous GPU or a borrowed 4070ti. Have a RM850x V3 PSU.

Tried Adrenaline 23.9.3 and 23.10.1
 
#8 ·
Yeah wanting to see if it’s worth to watercool a Red Devil. Seeing some of the temps above looks like it
 
#9 ·
Looks like these GPUs are very popular.
 
#10 ·
Yeah wanting to see if it’s worth to watercool a Red Devil. Seeing some of the temps above looks like it
Doubt it, from what I seen of RDNA3 it doesn't scale well with OC/Performance gains. Buy a 7900xt instead of 7800xt and waterblock imo.
Nice cards though I'm tempted to stuff one in my itx build.
 
#11 ·
I guess that depends on what’s limiting the performance, voltage , core, temp etc.
MPT can help with voltage if it’s supported and waterblock will help with increased temps.
 
#13 ·
Plopped a Sapphire Pulse next to my 5600X and this the card is fantastic. The 2-fan cooling solution is more than adequate even with the maximum +15% power slider, which puts me at ~280W Total Board Power. I'm able to comfortably run a 2700MHz frequency at the maximum 1150mV. Never broke 50% fan speed on the board with 80F ambient.

I'm curious what the maximum voltage and power slider is for the Nitro+, because I know it will have more power available with it's BIOS.

I just installed my Sapphire Nitro+ Radeon RX 7800 XT
Hankmoody, can you post a snapshot of what AMD's software allows you to do for power slider and voltage?
 
#14 ·
Plopped a Sapphire Pulse next to my 5600X and this the card is fantastic. The 2-fan cooling solution is more than adequate even with the maximum +15% power slider, which puts me at ~280W Total Board Power. I'm able to comfortably run a 2700MHz frequency at the maximum 1150mV. Never broke 50% fan speed on the board with 80F ambient.

I'm curious what the maximum voltage and power slider is for the Nitro+, because I know it will have more power available with it's BIOS.



Hankmoody, can you post a snapshot of what AMD's software allows you to do for power slider and voltage?
1150mv at 2700 sounds like alot with +15% power?
My Hellhound runs at well under 990mv realworld, (Slider 1050mv) at 2700 with power at 0%

What are your mem temps in idle? Nobody wants to answer this question be it here or reddit it seems.
 
#15 ·
1150mv at 2700 sounds like alot with +15% power?
My Hellhound runs at well under 990mv realworld, (Slider 1050mv) at 2700 with power at 0%

What are your mem temps in idle? Nobody wants to answer this question be it here or reddit it seems.
To be honest I'm still fighting stability even at 1150mV. I need to probably start doing dedicated burn testing to get find out what's causing my hard crashes.

I can check on memory temps and report back.

What program/bench are you using for stability testing?
 
#17 · (Edited)
1150mv at 2700 sounds like alot with +15% power?
My Hellhound runs at well under 990mv realworld, (Slider 1050mv) at 2700 with power at 0%

What are your mem temps in idle? Nobody wants to answer this question be it here or reddit it seems.
Idle sensor readings with a 2650MHz OC Set. It looks like idle mem temps are pretty warm, but not like "omg 80C at Idle", warm. Average between the readings looks like 56C.
Image
 
#18 ·
Hello there. Long time no posting... :)
Have just got XFX RX 7800 XT Merc 319. My temps were around 56 for the core and 75 for the hot spot @ stock. Then I've decreased voltage with 10% and also am using a custom fan curve (using MSI AB). Now my max temps in stress tests are 49 for the core and 70 for the hot spot.

Have a question - what is a normal deviation between core temp and hot spot temp?
 
#19 ·
I think if it's about 15c you're sitting pretty.
 
  • Rep+
Reactions: Belkov
#21 ·
I have Asrock 7800 XT Challenger (2 fans), and used to have 23-24c delta at higher temps (66 and 90 hotspot), but after I repasted earlier today, it drops to around 19-20c at most and maxed out at 83c hotspot. Shows how much the difference between AIBs matter in purchasing a GPU.
 
#23 ·
This Merc model has outstanding cooling potential. With more agressive fan curve and undervolting with 12% my game temps are no more than 48 for the core and 66 for the hotspot.
 
#24 · (Edited)
Got a Gigabyte 7800 XT Gaming OC coming today. Hopefully it overclocks well.

Edit - 327W max seen

ASIC is 50.3%

Memory can't do 2600 (2614 in wattman)

30C delta between core and hotspot @ 325W.


2nd edit -

took rank 1 for 7800x3d and 7800 xt I scored 21 035 in Time Spy

Graphics Score 22 833
 
#31 ·
Could anyone check if they get errors when testing with MSI Kombustor (GL) msi-01 (first test from the dropdown menu), enable Artifact Scanner and see if there are errors logged even when running default clocks/voltages?
 
#33 ·
For easy overclock set to this:
Ended up with 1100mV undervolt (1075mV crashed even with 2700 max frequency cap), 500 min, 3200 max, +15 PPT, 2566MHz VRAM. I'm guessing Fast Timings don't bring much of an improvement?

As for MSI Kombustor (GL) msi-01, there seems to be a problem with this test and drivers perhaps, my GPU isn't the only one exhibiting such errors at stock.
 
#35 ·
I crash at 2700 max @1075mV :) 2600 VRAM artifacts when using vulkan_memtest
 
#40 · (Edited)
Could anyone check if they get errors when testing with MSI Kombustor (GL) msi-01 (first test from the dropdown menu), enable Artifact Scanner and see if there are errors logged even when running default clocks/voltages?
No, actually it doesn't error even with OC preset, as my VRAM clocks never go above 1538 [Maybe as i only have 1920*1080. I cannot see how 4k would be, because all data is off the screen]. Without OC they are even lower, most of the time at 909 mHz.
But i do hit 3150 mHz or so running it with OC (only 2750 or so without it) doing artifact scanner. Max frequency is not THAT much of an issue for boost algorithm it seems (it does take some time to reach it though).

Image

Will try to enforce max VRAM frequency (at cost of performance, of course).
-- Well, not consistent 2557 mHz reported, but no artifacts either. So... Are you sure it is not your OC that is unstable? For me unstable VRAM will just cause sudden hard reboot and throws single critical WHEA 18, even at idle occasionally. There is some range where vulkan_memtest will throw errors, sure (at around 2650+FT for me, while 2660+FT will just instantly reboot under load). But after multiple reboots (and VRAM reduction) atm my VRAM is stable at 2570+FT. It took actual WEEK of PC uptime for it to cause a last reboot at 2580+FT btw.

I also wonder how for you people 1050 mV and below can be even possible other for some very limited use cases? ASIC 69.0%.
Granted, i SPECIFICALLY have game that is absolutely brutal on GPU stability. 890-900 was fine to run Superposition, but Time Spy and Warframe won't even launch. 970 was fine to run TimeSpy, but Warframe will easily crash. At around 1060 i thought i managed to get away from crashes, but then i limited FPS to 144. And, oh boy... Right now i am at 1100mV already (1090 mV just crashed today on Archon Hunt). I know it would be able to run plenty of games at 1060 or slightly higher. But that is not my target. That is why i currently brutalize my GPU with Warframe with limited FPS. I should be close to stability right now, and please... Fate? Let 1100 mV be enough (1. Round number, 2. Why should it be higher?).

Btw, for information purposes. RDNA3 boost algorithm is instantaneous (meaning it doesn't hold frequency at value you see as reported, frequency is always wildly swinging from low to high and vice versa) and limited both by peak and average power consumption. Meaning on low enough FPS you can see either 30-2400 mHz core (no mistakes, it will report 30 mHz because load time is not enough for it to report any higher sometimes). Or it will report something like 3100+ mHz for example in artifact scanner. But that way of power management despite being efficient in saving power should have huge downside - transient response. Meaning if you don't account for peak to peak voltage drop, GPU won't become stable. And if you run higher FPS, or higher load stuff, it can actually be more stable than when you drop load to some extend, so GPU would idle in between.

---

Can you download and run GravityMark? After completing the benchmark (default settings are fine), leave it running in the background where it shows your results. In my experience, running 3000+ max clock will result in artifacts in that benchmark section.
GravityMark downloads like turtle from 20 years ago (62.5 KBps), when i finish downloading it, will try it out. I had some artifacting (in line of purple-ish corruption effect), but only in EXTREMELY specific place (on one title screen with 480 FPS limiter) when FPS dropped from single frame due to two games running simultaneously, and i cannot reproduce it right now because i cannot get frames to drop again. And that also happens on default settings, so, no... It is not OC issue.

Will report when i try.
UPD: No artifacting after GravityMark except shimmering from lots of details on small resolution and pop-in on some asteroid's textures. GPU autoboosted to 2970-->3025 mHz (first to 2970, then SLOWLY to 3025 within some timeframe). Result: 42799 pts with default settings.

Currently have:
Min_clock = 500 mHz
Max_clock = 3100 mHz
Voltage_curve = [1100 mV crashed] 1110 mV
Power limit = +15%
VRAM_freq = 2570
Fast Timings = enabled
 
#41 ·
So... Are you sure it is not your OC that is unstable?
I get an error even on stock settings, usually 1 artifact in 2h of testing. This is confirmed with one more 7800XT (Nitro+) running at stock as well. Seems very sporadic. Thanks for testing.

When you say you boost to 3150MHz, do you mean Front-end Clock or Shader Clock (as reported in HWInfo)? I'm boosting to ~3150MHz Front-end and ~2800MHz Shader Clock.

My GPU showed artifacts on GravityMark when the Benchmark results show, the background scene would cause artifacts, despite no artifact for the whole duration of the test. Needed to drop my Max clock to 3000-3050 from 3200 to fix artifacting in that specific scenario and currently testing 2950MHz @1100mV because of a random Fortnite crash.