PC Graphics

[theolu]

As a preface, my PSU is a 9 year old Gigabyte Odin 585, motherboard also 9 years old ASUS P7H55D-M EVO, and I have since experiencing this issue tress-tested with my old GTX460 1GB (dual 6-pin power connectors) on the same PSU+motherboard for testing purposes, without any issues, although to my knowledge that card cannot tax the PCI-E 'rail' my PSU may have as heavily as the RX570 could.

I bought a used (I'm the third or more owner thereof) ASUS RX570 4GB Expedition which I received on Monday. By 8PM I had it up and running and was testing some games on it. There were some performance issues and power consumption seemed to be limited to up to 129W peak, around 100W average, where I was under the impression the card could easily draw almost 200W under load, especially if that load is a heavy one. Temperatures easily rose to around 79C, and there were performance issues in games (erratic framerate drops/chopping even at settings which my GTX750Ti 1GB would do better at). This was with using settings like 'Chill' and/or the AMD framerate target setting, and without.

I removed the cooler and redid the still-stock TIM, which was thoroughly dry and judging by the ease with which the heatsink detached, no longer cohesive.

I reassembled, installed the card, ran some games tests and immediate impressions were that temperatures were substantially better. The computer was running fine for about 10 minutes, by which time I put the side-panel on, very lightly touching the rounding corner of the protruding PCI-E power cable. 10-15s later, my screen went black; on my G13 I could see the CPU and internet connection were still active, and there were no sounds from Windows as though something had been disconnected.

I pressed the power button for Windows to shut down, which it did (HDD activity and all).

Upon turning the computer back on, it began reporting there was no GPU attached.

I kept in mind I'd lightly scraped the power cable's side with the sidepanel and reconnected that; no go.

I completely reseated the card and the cable, after blowing out the port with a lens cleaner's blower; the card booted and I got into Windows. About 20-30s into being in Windows, while programs were still loading, the screen went blank again. This time, Windows hanged after about 5 seconds and eventually the computer switched itself off. Restarting, it once again reported no GPU.

I can, reliably, get the card to start up and display a picture for anything between 10-30s if it's not been running for a while and I try starting it up; but inevitably its fans switch off and on subsequent start attempts it'll either briefly power the card or won't even try.

I've already redone its TIM twice again just in case the cooler was somehow not making contact with the die, but the area behind the GPU and the heatsink remain quite cool, so I don't think the GPU is necessarily overheating... What I can say is that with the first application I evidently applied just a teeny bit too much of a highly-viscous one (Cooler Master's MasterGel as supplied with with their MA410P cooler), and a bit of it came into contact with the resistors(?) present on top of the GPU package. The entire package has since been thoroughly cleaned using absolute alcohol. I haven't gone anywhere near the VRMs and haven't taken off the heatsink there either.

So my question is, could this be little more than an issue of the PSU being too dated to supply the card with the juice it wants? Voltages in the two 2min OCCT runs I was able to do looked normal, though performance figures weren't with the framerate starting at 300fps for the first 20s and then dropping off to around 230fps thereafter in both runs. Voltage and current obviously not being the same thing, and OCCT doesn't report to me what the consumption was, I can only say what the AMD monitor told me (100-120w, with occasional 129W peaks)

I'm going to have the opportunity to test on a relatively-new PSU (Cooler Master GX II 750W) with four dedicated 6+2pin rail this weekend, so I'm *hoping* I'm dealing with little more than a PSU issue - but just in case there's some other reason the card might work on that system I'd like to know whether the sort of behaviour I'm experiencing could be PSU, PCI-E slot power supply or motherboard power supply in general related otherwise. I mightn't have the opportunity to test that PSU on my system overall to see whether there's a difference in that regard, I'll have to see how things go. If I can I'd like to forego buying a new PSU which mightn't solve my problem anyway, as I'm already out about $100 equivalent having just bought this card..

Thanks for any input, and sorry for the lengthy post.

NB: I note that in the OCCT tests the huge FPS drop (300 -> 230fps) seems to correspond with the CPUTIN and overall temperature going up markedly, however the core clocks across the board (X3460) were basically the same from the beginning of the test till its end, so I don't think the first 20s were it running a higher turboboost clock on some of them and then dropping to no TB...

[elstaci]

The recent AMD Drivers have had some issues with GPU Fan controls.

First try installing a older previous 18.xx driver and see if the GPU starts to overheat? 

Also have you tried running OCCT PSU Test and see if it crashes immediately? If not what are the PSU Outputs when you tested the GPU card in OCCT? The 3.3vdc, 5.0vdc, and the 12 vdc should all being within 5% tolerances when under stress. If your 12 vdc is 11 vdc or under is indicating a weak PSU, as an example.

Also the RX 4xx and RX 5xx needs to be configure in AMD Wattman to prevent some of the problems you are having: here is how to configure your RX 5xx in Wattman from kingfish‌:

  Re: Wattman fan settings don't work

  

       kingfish     Jun 4, 2018 10:08 AM  (in response to snapperorgan)   

Your still changing the fan speed from automatic to manual and you haven't moved your Power Limit from default which throttles your card. It's explained here if you read it > AMD graphics performance

Before giving up on the card..why don't you try the settings exactly like I suggested? Set Wattman back to factory defaults first. Then change only the manual temperature settings and the Power Limit..click apply. Leave everything else at default...which is Automatic.

One more thing to check. Since you purchased that RX 570 used, it is possible that one of the former owners flashed the original vBIOS on the card. You can check by running a free program called GPU-Z and going to this website and see if the vBIOS version is the same for your GPU card. You can check from here: VGA Bios Collection | TechPowerUp 

[theolu]

Thanks, but as I said the card cannot boot, or when it does it's only active for 10-30s from when I press the power switch, so I can't get into Windows to get any of that done. It'll turn itself off even if I'm just sitting in the BIOS or have paused the POST tests.

In the meantime I've pulled up the specifications of the PSU again (figures I'd be so lazy as to not open the case to have a look on its sticker) for the 12V1 and 12V2 ratings, which are 16A and 18A respectively, or 192W and 216W.

By my understanding, the 12V1 is used for the mainboard power and auxiliary power connector there, while the 12V2 is used for everything else.

There are currently a 3.5" 7200rpm HDD and a 2.5" SSD attached to the 'other' 12V cables; of note is that the SSD will occasionally upon a cold start seemingly not start up, irrespective of which port it's connected to on the motherboard. A quick power down and up or a restart usually alleviates that problem. I suspect there's as much of a chance of there being an issue with the drive controller where that's concerned, though; I know my brother in law's more-recent motherboard required a firmware update to reliably detect his SSD at POST, even with AHCI mode enabled.

I also know my CPU can easily draw 130W or even more if overclocked on its own AND that my overclocks for the CPU have more recently become unstable, specifically when the CPU would try to enter a lower-power state rather than a higher-power one.

This only leads me to believe even more my PSU, which is certainly no spring chicken, is on its way out... Hopefully I'll gain some clarity in that regard this weekend.

[elstaci]

Yes your PSU is very old, But I would try your RX GPU Card on another computer and see if it does the same thing. It is possible the USED RX is causing all the problems. Just to eliminate the GPU Card. If the GPU card works normally in another computer then it may be your PSU.

Try using you old previous GPU card and see if it does the same thing? IF it boots up than check your PSU to see if it is failing by using OCCT PSU Test.

I reread your post and with the old Nvidia GPU card it works normally. It uses the same amount of power as the RX 570 according to this website: PSU REQUIREMENTS - RealHardTechX

So most likely it is your RX 570 is defective.

[theolu]

Tested the card on the other mobo+PSU, presented with the same symptoms. I ended up pulling the heatsink for the MOSFETs on the PCI-E power connector side (the motherboard one, not the auxiliary one) and noticed half of the bank of M3085M units only had about a third covered by the thermal tape, while the M3054M units which the heatsink was evidently meant to make contact with didn't ever have /any/ contact with the tape, the entire heatsink assembly essentially tilting off of those as it pressed down on the center-most column of M3085M units. See pic:

Leftmost are the M3054M units, pairs of 2 in 4 sets are the M3085M units.

I need to give the card one last test, but it seems it's reached the point where it altogether refuses to boot now. I'm beginning to think one or more of the MOSFETs on the board may have been damaged from a surge/short when I scraped at the PCI-E aux power cable. I can't notice that any of them get hot during attempts to get the card running, although it's obviously very difficult to gauge that without temperature probes seeing as the units can cool down to ambient before I can get the GPU's heatsink pulled off again, so I can't tell whether one of them is perhaps reaching a critical temperature and then causing power to the board to cut.

What I do know is they're relatively inexpensive (a handful of dollars for packs of 5-10 pieces per model type) and with a reflow station would be easy for an amateur like me to replace also. As such, I'm thinking of getting a full replacement set of all the MOSFETs for the board and just redoing the lot. The worst that can happen is I'll have spent $10 or so (reflow station aside, I'll need to find one to borrow or just give in and finally buy myself one, which I have wanted for a while) and I'll have not fixed the card. Best that can happen is I'll have fixed the card and have either learned to A) never touch the damned PCI-E aux cable ever again in my life just-in-case-it-bricks-the-frigging-card or B) that ASUS' engineers did a really shoddy job of designing this card's power solution's cooling, possibly having baked it (notice those dark patches between the 3 sets from the top? those were there when I pulled the heatsink off)

As for the page you linked, thanks, it'll be a useful resource for me in the future.

While the GTX460 has a similar overall power requirement, its peak consumption on any models ever tested was apparently up to around 175W, but more importantly all models released drew that over 2x 6-pin connectors whereas this card relies on a single 6-pin (well, 8-pin with the two extra grounding pins...) connector; so depending on the PSU in use this could obviously have an influence on how power is drawn from the PSU. To some extent, anyway.

Thanks again for the input, and sorry again for the verbose++ messages. I'll report back here if/when I finally receive the MOSFETs and have put them on.

[elstaci]

Thanks for the update. I know a User here that can probably help you with repairing your card or maybe give his professional input about your analyst.

colesdav‌ is the User I am thinking about.

[theolu]

Thanks. Do they get notified when you tag them like that, or shall I contact them directly? Don't want to double-tap unnecessarily

[elstaci]

No, if colesdav‌ is interested he would comment. When you tag someone by "@" and then his forum name, that User is notified that someone mentioned his name in a thread.

I wouldn't contact him directly unless he suggests it. It is better to do it through the Thread you are active on.

EDIT: I mentioned that particular User because he uses a Reflow Station to repair his GPU cards from previous posts.

[colesdav]

Hi,

I am pretty busy right now but I quickly attempt to read this post. 

So.

I have the following information.

Motherboard = ASUS P7H55D-M EVO (https://www.asus.com/Motherboards/P7H55DM_EVO/specifications/ )

PSU              = Gigabyte Odin 585

CPU              = ?
GPU             = ASUS RX570 4GB Expedition (EX-RX570-O4G | Graphics Cards | ASUS Global )

RAM             = ?

OS                = Windows ?

HDD             = ?

Key points I take from your post, with some comments from me.

(A).If you use a GTX460 1GB (dual 6-pin power connectors) you do not get any problem.

(B).There were some performance issues and power consumption seemed to be limited to up to 129W peak.
      - This could be Wattman Power Limit Settings

(C). Performance Issues: This was with using settings like 'Chill' and/or the AMD framerate target setting, and without.

      - Radeon Chill kills keyboard only input FPS, it is broken in terms of getting  good gaming performance at values of Chill Min that actually save power. It needs to be fixed. Turn it off.
      - FRTC should be turned off initially if you are investigating GPU performance. 

(D). You mention use of a GTX750Ti 1GB 

(E). You "removed the cooler and redid the still-stock TIM, which was thoroughly dry and judging by the ease with which the heatsink detached, no longer cohesive".
- That is a pity, because if you have bought the card on Ebay you would have been able to rtest the GPU and see if it is working or not.
- If it is not working then return the GPU within first 14 day of purchase, and get a refund.
- That  is what I would have done.
- Honestly I do not think messing about with third hand GPU when AMD are practically giving those RX 570 4GB GPU's away if you include 2 free game codes at the moment and you get a Warranty.  Here they are on Newegg:

 

So first of all before going further. Would you not simply consider giving up, selling this third hand RX570 4GB for parts on Ebay, and just purchasing a new AMD GPU? You will likely spend more $$$ trying to fix this broken GPU.

If not then please fill in the missing system information above, and then I will try to understand the rest of your post and reply later this weekend.

Thank you.

[theolu]

Motherboard = ASUS P7H55D-M EVO (https://www.asus.com/Motherboards/P7H55DM_EVO/specifications/)

PSU = Gigabyte Odin 585 (https://www.gigabyte.com/Power-Supply/GE-P585A-C2#sp)

CPU = Intel Xeon X3460 (i7-860) @ stock (https://ark.intel.com/products/42931/Intel-Xeon-Processor-X3460-8M-Cache-2-80-GHz-)

GPU = ASUS RX570 4GB Expedition (https://www.asus.com/Graphics-Cards/EX-RX570-O4G/)

RAM = 2x Patriot 4GB 1333EL (can't find specs/datasheet), 2x Corsair XMS3 2GB

(https://www.corsair.com/us/en/Categories/Products/Memory/XMS3-%E2%80%94-4GB-Dual-Channel-DDR3-Memory...) - both sets at 1333Mhz and 1.5V (manually set)

OS = Windows 10 Pro build 17763 (freshly re-installed, updated et al)

HDD = Boot: ADATA SU650 240GB (https://www.adata.com/rs/specification/503) (new) - secondary: Seagate (https://www.seagate.com/www-content/datasheets/pdfs/3-5-barracudaDS1900-7-1706US-en_US.pdf) (also new)

In the course of testing the card's functionality I've tried disconnecting the drives and reducing the system's RAM to one stick, and have run the motherboard at default settings also; no improvement.

A) Yup
B) Noted for future reference
C) Yeah, I alternated between using either or both, and using neither; results were the same in all cases
D) Currently my primary GPU, yeah
E) The TIM I only changed after using the card for the first day (during which it did work, albeit with the performance issues and rapidity with which it reached 79C even under modest loads -> thermal throttling the core clock below even the lower amount it ought to run at of 1168Mhz - the lowest I noticed it go was somewhere down to 1120. This happened while FRTC and Chill were both off, and if I recall I tested at least a +5% increase in power setting in WattMan afterward with no improvement.

I should note at this point, I live in South Africa; while there is a policy in South Africa of being able to return a non-working item bought used within a set timeframe, since I had 'modified' the card where it was 'working' and then it ceased working, I essentially broke it and thus can't claim it needs to be replaced/refunded.

At the time I committed to buying it, the cheapest one was able to find RX570s here for was still around $285. I see now that one unit from Gigabyte is selling at a now-normal price of $183 and another XFX model on sale for about the same, both with the games offer; unfortunately I've no interest in the games and wouldn't be able to resell them to help cover the cost of the card, though...

While I'd considered selling the card for parts and would like to, shipping cost for it out of South Africa would amount to about as much as one does new on its own.

I've, in the meantime, been able to get some information on testing the mosfets to find if/which one has shorted, and will be checking what the buck converter's cost would be to replace also. The MOSFETs themselves will be very cheap, less than $10 altogether; the buck converter I don't know the cost of just yet. It's the reflow station I'd need to borrow/buy, though depending on the cost of that I just might go for it since I've had situations where I needed/wanted one in the past.

Thanks for the reply so far, and don't worry about attending to it ASAP - I'm going to just shelve the card for now and order the MOSFETs + buck convert I think, and then look at finding a reflow station. There are many available new here for less than $100, though obviously I'll investigate to ensure I'm not buying more 'junk' - that, or possibly build my own.  We've a many-decades-old electronics parts and repairs specialty store here called HAMRADS which will probably be able to advise and possibly sell me a good, used unit at a decent price.

[theolu]

An update:

I've been researching diagnosing the issue and testing with my multimeter now that I'm home. Something is definitely shorting to ground, I just can't figure out what with just the multimeter and without disconnecting any capacitors/other components; there are three capacitors in particular on the PCI-E Aux side which look suspect on their solder joints (dark-gray patches), although even after a thorough inspection (with a microscope, up to 30x power) I cannot see that any of the ceramic capacitors are cracked or otherwise damaged.

I'm going to be looking for some way to give the board a 1V power source of around 2A so I can use some alcohol to look for the shorting component. Short of buying a bench power supply though I'm not sure where I'll get such a power source from, and I'm not electronics-inclined enough to make it myself just yet either. I think my uncle had shown me during my last visit a bench power supply he has, so if he does have one I'll use that next time I visit him.

I'm going to be glad if it's no more than a dead capacitor, since that'll be easier for me to replace and won't necessarily mean getting a reflow station...

[colesdav]

Hi, 

I can't respond to you in detail today, but I will attempt to give some help to you tomorrow.
I understand your reasons for attempting a fix. 
Is the GPU tripping out the PC PSU at boot or is the GPU simply not recognised by Windows? 
If it is the latter, then put a working GPU in primary slot and fit the problem GPU into secondary slot and boot up and see what Windows reports.
You may simply have a corrupted BIOS for example. 

If the GPU is preventing your PSU from powering up, then first thing I would do is clean the board completely with Isopropyl alcohol and without removing or modifying any components.

Then I would run a solder reflow on the GPU:
See here: 
 

[colesdav]

Here is link to information on creating a cheap solder reflow station for an  HD7970 OC 6GB GPU and repairing it. 
I was getting VGA Error at Boot and the PC would not boot up with the GPU installed as primary. 
https://community.amd.com/message/2816083?commentID=2816083#comment-2816083 
The HD7970 OC 6GB is still working fine. 
Bye. 

[colesdav]

Hi, 

RE; I'm going to be looking for some way to give the board a 1V power source of around 2A so I can use some alcohol to look for the shorting component. Short of buying a bench power supply though I'm not sure where I'll get such a power source from, 

You already have a power source, your PSU Power Supply. 
There are electronics available to connect onto a PC PSU and turn it into a simple bench power supply. 
I own such a board, although it is a basic one. 

I will try to get you order and supplier details and a picture of the one I have. 

Bye. 

[theolu]

"Is the GPU tripping out the PC PSU at boot or is the GPU simply not recognised by Windows?"

The GPU will, when stone-cold, now occasionally still start up and display a picture for up to about 20s before going 'dead' again. Upon restarting the computer it will simply not detect a GPU is attached to the PCI-E port at all.

This is reflected in Windows also where when using an iGPU it's as though no card is attached to the PCI-E port even if dual-GPU mode is selected in the BIOS/UEFI (iGPU + PEG)

As for using my PSU (or any PSU) to at least partially-power the board, I've considered doing this but was concerned about running it for too long like that and possibly causing whichever component it is which is shorting -> overheating to straight-up pop and potentially damage the board around said component?

I was also previously concerned about the GPU overheating if I had to have the heatsink off, but I've remembered now I've got some other heatsinks here I can put on which will at least delay that happening (if it would) long enough I can notice the heatsink is getting too hot. I need to get more alcohol first anyway, so while I'm in the next town over where I do that I'll check by a small electronics supply store there if they happen to have a 1V power supply of some sort just in case, which I could then current limit with something I put in-line perhaps, as necessary.

Thanks for the link to the thread where you outlined the reflow setup you made. I also came across a few other articles and videos for creating different sorts of temperature-controlled setups, all of which I'm going to be looking over more thoroughly when a headache which has been plaguing me the past few days has gone away. Between them I'll figure out what I can source locally to create something usable and affordable to get started, and also approach some people I know of who may have used stations they're interested in selling to compare buying those vs building my own.

[colesdav]

Hi

RE: The GPU will, when stone-cold, now occasionally still start up and display a picture for up to about 20s before going 'dead' again.
- This indicates to me you need to reflow the GPU. The setup I pointed you to should be enough. 
If that doesn't work then you will need to look at the VRM's on the GPU board.
Then if they are o.k. you could look at solder re-ball the contacts on the rear of the GPU die, there is a good Youtube video showing that. 

But. 

In principle you should start off with the simplest way to get the GPU running again. Which is solder reflow, using correct reflow compound that I gave information on in the link I pointed you to. 
Remember the more you spend on fixing this GPU that is money that could have been spent on a new one. 

As for your headache - I hope you haven't been breathing in fumes from the industrial alcohol you used to clean the GPU printed circuit board, or solder fumes or some other chemical. You need to be careful and work in a well ventilated area, preferably with a fume extractor. Take care. Hope you get better soon. 

Bye. 

[colesdav]

Hi, 

Here is the type of PSU to Bench coversion module I have.

Search for this
"24 Pins ATX Benchtop Board Computer PC Power Supply Breakout Adapter Module"

Note that there is no inherent current control in the add in board.
If you have a PSU with Corsair Link type feature and an active PC to drive the PSU you can use Corsair Link / Corsair CUE  to set a current limit.
If that is too complex then you should in theory be able to build a simple transistor based voltage regulator with current limit from any of the outputs supplied by the breakout board. 

I really have to go now, but hopefully that gives you a starting point to look for a PSU -> Bench Power Supply module. 

If you find a better one with built in voltage output regulator and current limit please let me know.

Bye.  

[colesdav]

The PSU _> Bench conversion PCB should about $5 to $10. 
If you are prepared to spend ~ $45 you should be able to get something like this: