7900 XTX Seriously lacking

[tabascosauz]

Isn't AGESA part of the BIOS update? I have the newest BIOS. I've tried at least the last 4-5 BIOS versions hoping one would work better. Currently the newest one from 3/2024 I think is installed.

Each BIOS release is based on specific AGESA release versions. Somewhere around 1.2.0.7-1.2.0.8, there seems to have been a big change in voltage curves for a lot of 5800X3Ds that flat out causes them to perform worse in all aspects. All the AGESAs since then have not further changed anything much. Unfortunately, the only pre-change alternative is 1.2.0.6 (anything earlier does not properly support X3D), and unlike later AGESAs that one is prone to the fTPM lag issue, if you use TPM.

Did you run R23 and get some HWInfo data for CPU and GPU load?

 

[Solaris17]

Each BIOS release is based on specific AGESA release versions. Somewhere around 1.2.0.7-1.2.0.8, there seems to have been a big change in voltage curves for a lot of 5800X3Ds that flat out causes them to perform worse in all aspects. All the AGESAs since then have not further changed anything much. Unfortunately, the only pre-change alternative is 1.2.0.6 (anything earlier does not properly support X3D), and unlike later AGESAs that one is prone to the fTPM lag issue, if you use TPM.

Did you run R23 and get some HWInfo data for CPU and GPU load?

With this I also dont understand any reccomendations to OC or change voltages. I have seen a few posts requesting or reccomending changes they or others utilize during OC. However if im understanding the issue, OP cant even compete against OTHER STOCK setups. If im right OCing, voltage tweaks, etc are only going to get him to BASELINE. Which is where I think we should be focusing.

IMO there is an issue elsewhere in the system, be it temps under load or an OS issue. Whatever changes that may have been made to the bios need to be reset. I would reset CMOS uninstall all OC software or otherwise, DDU the drivers and do a clean run, then another run with GPU-Z and like HWmonitor/info/we running so we can see the overall performance of the machine in general.

This thread might be about and focusing on a "card" problem, but I dont think we are dealing with a card problem, I think we are dealing with a problem that manifests as a card problem.

The OP has posted like 5 times and we dont have any concrete numbers or screen shots, we really need to start (properly) gathering data if he is as far off course as he thinks he is.

 

[Veseleil]

Timespy CPU score is alright if stock. -30 CO, tuned ram gets about 12500-13000 range (Timespy highly affected by ram) I'm afraid it's on the GPU. Maybe also use Radeon OSD while testing to see if there are sudden drops or any unusual values while running (GPU clock, power, usage, voltage etc).

I just think that it's easier to first confirm his 5800X3D behaves as intended, then it's left to isolate other parts of the system and eventually the GPU.

Isn't AGESA part of the BIOS update? I have the newest BIOS. I've tried at least the last 4-5 BIOS versions hoping one would work better. Currently the newest one from 3/2024 I think is installed.

Each BIOS version is based on a particular AGESA update. 5800X3D works best with 1.2.0.6c

With this I also dont understand any reccomendations to OC or change voltages. I have seen a few posts requesting or reccomending changes they or others utilize during OC. However if im understanding the issue, OP cant even compete against OTHER STOCK setups. If im right OCing, voltage tweaks, etc are only going to get him to BASELINE. Which is where I think we should be focusing.

IMO there is an issue elsewhere in the system, be it temps under load or an OS issue. Whatever changes that may have been made to the bios need to be reset. I would reset CMOS uninstall all OC software or otherwise, DDU the drivers and do a clean run, then another run with GPU-Z and like HWmonitor/info/we running so we can see the overall performance of the machine in general.

This thread might be about and focusing on a "card" problem, but I dont think we are dealing with a card problem, I think we are dealing with a problem that manifests as a card problem.

My thoughts were to first eliminate 5800X3D by making sure it runs as it should, after that RAM, storage, PSU and GPU the last. But more info is necessary. Off to bed, good night.

 

[DonnieDangerous]

Each BIOS release is based on specific AGESA release versions. Somewhere around 1.2.0.7-1.2.0.8, there seems to have been a big change in voltage curves for a lot of 5800X3Ds that flat out causes them to perform worse in all aspects. All the AGESAs since then have not further changed anything much. Unfortunately, the only pre-change alternative is 1.2.0.6 (anything earlier does not properly support X3D), and unlike later AGESAs that one is prone to the fTPM lag issue, if you use TPM.

Did you run R23 and get some HWInfo data for CPU and GPU load?

Here is the results from running R23

I took some metrics for you all to look over.

Some taken while running Uningine With the 7900xtx OC in adrenalin (see settings pic) the highest gpu clock speed i saw was 2611. Highest hotspot was 66c. total board power average was 345-355 with a high spike of 525. The gpu memory clock speed seemed to hold steady at 2487.

Some specs are from running cpu-z benchmark

Here is also a link to the cpu-z benchmarks results https://valid.x86.fr/zhfc2k

 

[Solaris17]

Im going to go out on a limb. I am not saying this is the main or even a large contributer. As someone however that did run 3x 4k displays until recently.

This is going to drop your overall scoring in everything. It doesnt matter if your game/movie/benchmark is only running in fullscreen mode (as you should be) on only 1 monitor, your GPU still needs to push the pixels for 3.

 

[tabascosauz]

Here is the results from running R23

I took some metrics for you all to look over.

I am curious as to why you are so resolutely opposed to using HWInfo. No need to keep all these pointless windows up to show not a lot of info, just run it minimized, and you'll see why we're all telling you to use it once you see HWInfo's sensors window for the first time.

Not much to say about the R23 results other than it's just pretty low. Since you didn't include temps/clocks/volts for the run. ST is maybe ~50pts off of a decent score so it's not the absolute worst for stock. MT score is just low, even for stock. 69C in CPU-Z is about expected but doesn't reveal much either, it's a short and lax test compared to most others.

Some taken while running Uningine With the 7900xtx OC in adrenalin (see settings pic) the highest gpu clock speed i saw was 2611. Highest hotspot was 66c. total board power average was 345-355 with a high spike of 525. The gpu memory clock speed seemed to hold steady at 2487.

Some specs are from running cpu-z benchmark

Here is also a link to the cpu-z benchmarks results https://valid.x86.fr/zhfc2k

So at 350W your edge temp is 39C and your hotspot is 64C on the 7900XTX. Neither is throttling but even knowing Navi31's general propensity for big edge-hotspot delta, I might have expected better from a waterblocked XTX.

Clocks seem to be quite low considering this is an OC'ed and waterblocked card. Just a hair over 2600MHz seems more befitting of a reference MBA card.

HWInfo also has very long and detailed sensor section for Navi31 cards. Give it a try and see if the data shows anything off.

Radeon software does not report Navi31 GDDR6 temps, but HWInfo does. Probably best to check it and make sure you installed those memory pads properly.

 

[thesmokingman]

So at 350W your edge temp is 39C and your hotspot is 64C on the 7900XTX. Neither is throttling but even knowing Navi31's general propensity for big edge-hotspot delta, I might have expected better from a waterblocked XTX.

It's more a sign of poor block fitment.

 

[DonnieDangerous]

Not sure how to upload this but I had HWinfo running in the background and ran the Uningine 4k benchmark with the default settings in adrenalin. If the link does not show, please let me know how to upload it correctly.

 

[freeagent]

With a solid curve R23 should be at 15K and change.

 

[DonnieDangerous]

test

 

[tabascosauz]

Not sure how to upload this but I had HWinfo running in the background and ran the Uningine 4k benchmark with the default settings in adrenalin. If the link does not show, please let me know how to upload it correctly.

Just take a screenshot.

HWInfo reports current, min, max and average over a specific polling period. You can reset the period by hitting the clock icon in the lower right of the sensors window - do that right before you run a test, and once you check afterwards you'll have HWInfo data for only that test period.

With a solid curve R23 should be at 15K and change.

Agreed, but not without any CO. Still, without CO it still should not be dipping below 14k.

imo the GPU is the big issue here with clocks just being too low overall even though GPU temps are fine.

@DonnieDangerous what card specifically is this? What block did you use? If it wasn't a factory waterblocked card, did you have any frame of reference for the stock air-cooled card in terms of clocks, power and temps? Is it a reference MBA card?

 

[DonnieDangerous]

Just take a screenshot.

HWInfo reports current, min, max and average over a specific polling period. You can reset the period by hitting the clock icon in the lower right of the sensors window - do that right before you run a test, and once you check afterwards you'll have HWInfo data for only that test period.



Agreed, but not without any CO. Still, without CO it still should not be dipping below 14k.

imo the GPU is the big issue here with clocks just being too low overall even though GPU temps are fine.

@DonnieDangerous what card specifically is this? What block did you use? If it wasn't a factory waterblocked card, did you have any frame of reference for the stock air-cooled card in terms of clocks, power and temps? Is it a reference MBA card?

Hopefully this is the info you need to see

Just take a screenshot.

HWInfo reports current, min, max and average over a specific polling period. You can reset the period by hitting the clock icon in the lower right of the sensors window - do that right before you run a test, and once you check afterwards you'll have HWInfo data for only that test period.



Agreed, but not without any CO. Still, without CO it still should not be dipping below 14k.

imo the GPU is the big issue here with clocks just being too low overall even though GPU temps are fine.

@DonnieDangerous what card specifically is this? What block did you use? If it wasn't a factory waterblocked card, did you have any frame of reference for the stock air-cooled card in terms of clocks, power and temps? Is it a reference MBA card?

Not factory, I have an alphacool on it

Im going to go out on a limb. I am not saying this is the main or even a large contributer. As someone however that did run 3x 4k displays until recently.

View attachment 345700

This is going to drop your overall scoring in everything. It doesnt matter if your game/movie/benchmark is only running in fullscreen mode (as you should be) on only 1 monitor, your GPU still needs to push the pixels for 3.

I get the exact scores in 1080 or on my other monitor that is a 2560x165hz

 

[Solaris17]

I get the exact scores in 1080 or on my other monitor that is a 2560x165hz

im going to assume by that you mean its far closer than say the gap you are getting now, because that literally isnt possible otherwise. Not only from personal experience, but years of collective internet understanding.

At any rate, I would be curious to see if the scores change or improve at all with like a blank SSD and a fresh format no extra software. It might also be worth running some 3dmark runs in your current situation to see how you stack up instead of beating on cinebench and unigen.

 

[DonnieDangerous]

Im going to go out on a limb. I am not saying this is the main or even a large contributer. As someone however that did run 3x 4k displays until recently.

View attachment 345700

This is going to drop your overall scoring in everything. It doesnt matter if your game/movie/benchmark is only running in fullscreen mode (as you should be) on only 1 monitor, your GPU still needs to push the pixels for 3.

when gaming or testing i only use one monitor. I was using 2 for some of the tests above before running the HWinfo but the scores were no different than when I had 2 or 1 monitors.

im going to assume by that you mean its far closer than say the gap you are getting now, because that literally isnt possible otherwise. Not only from personal experience, but years of collective internet understanding.

well, i never get exact for any test but nothing noticeably. Maybe 50-200 point difference when scores are 18k-19k

im going to assume by that you mean its far closer than say the gap you are getting now, because that literally isnt possible otherwise. Not only from personal experience, but years of collective internet understanding.

At any rate, I would be curious to see if the scores change or improve at all with like a blank SSD and a fresh format no extra software. It might also be worth running some 3dmark runs in your current situation to see how you stack up instead of beating on cinebench and unigen.

Timespy is the 3dmark? I think I posted one pic from time spy. It is the only test that had the gpu in the normal range for 7900xtx around 30k-32k

 

[Solaris17]

when gaming or testing i only use one monitor. I was using 2 for some of the tests above before running the HWinfo but the scores were no different than when I had 2 or 1 monitors.


well, i never get exact for any test but nothing noticeably. Maybe 50-200 point difference when scores are 18k-19k

but you had 3 plugged in. it doesnt matter if the other 1 or in this case 2 other monitors did nothing other than display a picture of a duck its still pixels. The CPU-Z validation showed you had 3 monitors connected.

It doesnt meant it isnt true though. Like I said in the quote you took, it doesnt need to be the primary reason or the biggest influence. If your score truly isnt changing at all even with the other 2 disconnected physically it simply makes it even more probable you have a platform centric issue (mobo/cpu/gpu).

Timespy is the 3dmark? I think I posted one pic from time spy. It is the only test that had the gpu in the normal range for 7900xtx around 30k-32k

What do other runs tell you? Are they the same? what about broken down into the GPU and CPU scores? Which one varies the most compared to others online?

 

[DonnieDangerous]

but you had 3 plugged in. it doesnt matter if the other 1 or in this case 2 other monitors did nothing other than display a picture of a duck its still pixels. The CPU-Z validation showed you had 3 monitors connected.

It doesnt meant it isnt true though. Like I said in the quote you took, it doesnt need to be the primary reason or the biggest influence. If your score truly isnt changing at all even with the other 2 disconnected physically it simply makes it even more probable you have a platform centric issue (mobo/cpu/gpu).



What do other runs tell you? Are they the same? what about broken down into the GPU and CPU scores? Which one varies the most compared to others online?

have have no Idea why it says three are connected, I disconnect them in display settings and they are physically powered off.

Timespy has average scores to others online

Unigine and Passmark far lower than other uploaded results.

 

[tabascosauz]

Hopefully this is the info you need to see


Not factory, I have an alphacool on it


I get the exact scores in 1080 or on my other monitor that is a 2560x165hz

Ah, reference. That's fine. Do you remember what kind of edge-hotspot temp delta you were seeing on the stock MBA cooler?

27C delta at full load - that block should really be remounted. Assuming the block doesn't have unlucky machining defects causing the problem, that's just a straight up bad paste job/mount.

You don't have to show the HWInfo summary pages - the sensor pages are more important. ie. show the first few pages with CPU and mobo reporting when running CPU tests

Not seeing anything that is an obvious problem. Only other thing I can think of is whether you have clean installed Windows anytime recently. I see the mini displays inside your case and am curious just how much background processes are running. On old and dirty Windows installs I did lose bench perf noticeably on both 4070 Ti and 7900XT.

 

[Super Firm Tofu]

If performance is inconsistent, PCIe riser/cables issue possibly?

 

[DonnieDangerous]

Ah, reference. That's fine. Do you remember what kind of edge-hotspot temp delta you were seeing on the stock MBA cooler?

27C delta at full load - that block should really be remounted. Assuming the block doesn't have unlucky machining defects causing the problem, that's just a straight up bad paste job/mount.

You don't have to show the HWInfo summary pages - the sensor pages are more important. ie. show the first few pages with CPU and mobo reporting when running CPU tests

Not seeing anything that is an obvious problem. Only other thing I can think of is whether you have clean installed Windows anytime recently. I see the mini displays inside your case and am curious just how much background processes are running. On old and dirty Windows installs I did lose bench perf noticeably on both 4070 Ti and 7900XT.

before photos

Ah, reference. That's fine. Do you remember what kind of edge-hotspot temp delta you were seeing on the stock MBA cooler?

27C delta at full load - that block should really be remounted. Assuming the block doesn't have unlucky machining defects causing the problem, that's just a straight up bad paste job/mount.

You don't have to show the HWInfo summary pages - the sensor pages are more important. ie. show the first few pages with CPU and mobo reporting when running CPU tests

Not seeing anything that is an obvious problem. Only other thing I can think of is whether you have clean installed Windows anytime recently. I see the mini displays inside your case and am curious just how much background processes are running. On old and dirty Windows installs I did lose bench perf noticeably on both 4070 Ti and 7900XT.

after photos these were while runnin gpu benchmark only

Ah, reference. That's fine. Do you remember what kind of edge-hotspot temp delta you were seeing on the stock MBA cooler?

27C delta at full load - that block should really be remounted. Assuming the block doesn't have unlucky machining defects causing the problem, that's just a straight up bad paste job/mount.

You don't have to show the HWInfo summary pages - the sensor pages are more important. ie. show the first few pages with CPU and mobo reporting when running CPU tests

Not seeing anything that is an obvious problem. Only other thing I can think of is whether you have clean installed Windows anytime recently. I see the mini displays inside your case and am curious just how much background processes are running. On old and dirty Windows installs I did lose bench perf noticeably on both 4070 Ti and 7900XT.

Background processes

Here is before and after a quick cpu-z benchmark test.

 

[tabascosauz]

before photos


after photos these were while runnin gpu benchmark only


Background processes

Here is before and after a quick cpu-z benchmark test.

How are your individual GPU hotspot readings? Go into the dropdown.

I can kind of see the "new AGESA" effect: 116W in CPU-Z for just 4.3GHz effective. CPU-Z is barely even a workload, and 116W is pretty obscene power draw for such light work for a stock X3D.

Are you sure your X3D is completely stock?

But I still don't think any of that would account for a problem as obvious as what's going on here (ie. GPU clocks and performs low)

When does this current Windows install date back to?

 

[DonnieDangerous]

How are your individual GPU hotspot readings? Go into the dropdown.

I can kind of see the "new AGESA" effect: 116W in CPU-Z for just 4.3GHz effective. CPU-Z is barely even a workload, and 116W is pretty obscene power draw for such light work for a stock X3D.

Are you sure your X3D is completely stock?

But I still don't think any of that would account for a problem as obvious as what's going on here (ie. GPU clocks and performs low)

When does this current Windows install date back to?

I have literally done at least 10 fresh installs of windows over last few weeks. When I mean fresh I mean erasing the ssd through the motherboard, removing all drives but main c drive, clearing cmos, fresh windows install setup as a new pc so nothing rolls back over. Many of the first test screenshots on my initial posting of passmark, timespy and uningine were right after a fresh install with zero other programs installed only updated drivers and was done so within past 2 weeks.

I have literally done at least 10 fresh installs of windows over last few weeks. When I mean fresh I mean erasing the ssd through the motherboard, removing all drives but main c drive, clearing cmos, fresh windows install setup as a new pc so nothing rolls back over. Many of the first test screenshots on my initial posting of passmark, timespy and uningine were right after a fresh install with zero other programs installed only updated drivers and was done so within past 2 weeks.

How are your individual GPU hotspot readings? Go into the dropdown.

I can kind of see the "new AGESA" effect: 116W in CPU-Z for just 4.3GHz effective. CPU-Z is barely even a workload, and 116W is pretty obscene power draw for such light work for a stock X3D.

Are you sure your X3D is completely stock?

But I still don't think any of that would account for a problem as obvious as what's going on here (ie. GPU clocks and performs low)

When does this current Windows install date back to?

I will run the Cinebench tomorrow with HWinfo running in the background

I have literally done at least 10 fresh installs of windows over last few weeks. When I mean fresh I mean erasing the ssd through the motherboard, removing all drives but main c drive, clearing cmos, fresh windows install setup as a new pc so nothing rolls back over. Many of the first test screenshots on my initial posting of passmark, timespy and uningine were right after a fresh install with zero other programs installed only updated drivers and was done so within past 2 weeks.

X3d is completely stock. I purchased it from Amazon and installed it. I looked into OC-ing but quickly found that was not something I wanted to risk being not like other easily OC'd cpu's

It's more a sign of poor block fitment.

I'm not sure how I could do it any better. I followed the directions exactly and watched numerous videos before doing so. I assumed average Temps of 39 and hotspot of 66 was pretty good compared to non water cooled I've seen idle around 50 and over 100 sometimes for hotspots. No? Not an expert and why I seeked help. I'm a mechanic and I see this as I can do all my own maintenance but ripping apart the engine is beyond my scope if that make sense? I built the dang thing and can do some troubleshooting but this goes beyond my capabilities or knowledge.

 

[tabascosauz]

I'm not sure how I could do it any better. I followed the directions exactly and watched numerous videos before doing so. I assumed average Temps of 39 and hotspot of 66 was pretty good compared to non water cooled I've seen idle around 50 and over 100 sometimes for hotspots. No? Not an expert and why I seeked help. I'm a mechanic and I see this as I can do all my own maintenance but ripping apart the engine is beyond my scope if that make sense? I built the dang thing and can do some troubleshooting but this goes beyond my capabilities or knowledge.

Good that you've clean installed, at least rules that out.

If your MBA card was an early sample, it sounds like yours was afflicted with the vapor chamber defect. Anything beyond about 30C delta would have been under that recall. The water block temps right now aren't nearly that bad, but on air 25C+ deltas would be an immediate justification to repaste.

I can understand the reluctance to do it though, it looks like it would be a pain.

 

[thesmokingman]

I'm not sure how I could do it any better. I followed the directions exactly and watched numerous videos before doing so. I assumed average Temps of 39 and hotspot of 66 was pretty good compared to non water cooled I've seen idle around 50 and over 100 sometimes for hotspots. No? Not an expert and why I seeked help. I'm a mechanic and I see this as I can do all my own maintenance but ripping apart the engine is beyond my scope if that make sense? I built the dang thing and can do some troubleshooting but this goes beyond my capabilities or knowledge.

It's common with most blocks out of the box. It will take you having to check your clamping pressure vs pad thickness. This is why I avoided all that by going with the Aqua card which doesn't suffer from a big delta. There's another thread here where someone tweaked their block to shrink the delta. I didn't follow that thread much be definitely recall it.

 

[DonnieDangerous]

Good that you've clean installed, at least rules that out.

If your MBA card was an early sample, it sounds like yours was afflicted with the vapor chamber defect. Anything beyond about 30C delta would have been under that recall. The water block temps right now aren't nearly that bad, but on air 25C+ deltas would be an immediate justification to repaste.

I can understand the reluctance to do it though, it looks like it would be a pain.

Yeah it sucks, have to drain the whole system, remove the card from the block and so on. I love my setup and spent a crap ton of money and time on the cooling setup. Kind of wish I just would have bought the 4090 from the get go. I switched to water because of the vapor issue. But I need to figure out how to fix it since AMD d**** don't want to help me.

 

[thesmokingman]

But I need to figure out how to fix it since AMD d**** don't want to help me.

Yea, throwing that block on means you're on your own now. Which is why I wrote that ya need to validate the card before going to a loop. Btw, looking at your pics dump the drain valve, they're useless. You want at least one QDC somewhere, ideally around the gpu so you can yank it out easily. Drain valves are dumb cuz it's redundant with QDCs and all the water will never drain out for obvious reason. With a QDC open you can use a datavac to push the water out.