PC Building

[Duron999Mhz]

Hello everyone, This isn't exactly a pc building topic but one I thought some folks may find of interest, so I thought I would share my experience. 

I recently acquired a used RX Vega 64 Limited Edition. The model I have looks to have been made by XFX in August of 2017. I had some suspicions this GPU may have been used as part of a mining rig but it wasn't mentioned in the advertisement. After patiently waiting for the card to arrive, I was delighted when it did a few days ago. I quickly unwrapped the package and admired at the beautiful GPU that I held in my hands. I must say, the AMD reference cards from my RX 6800 to the RX 5700XT are built quite solid and feel like a premium product. After recently cleaning, repasting, and repadding my previously enjoyed RX 5700XT reference card and a Radeon Pro W5500, it gave me the experience and confidence to attempt to clean up this seemingly neglected Vega GPU. 

After inspecting the Vega, I noticed tell tale signs of some rust freckles on the I/O plate and that the GPU core tension screws had already been previously been touched as the warranty stickers were both pushed into the screws. I had thoughts this card may have had some third party liquid cooling installed before being sold off but I held my breath as I cautiously put the Vega into my spare PC, plugged into the power cables as well as the HDMI cable, and pressed the power button on my pc. To my relief, the pc booted up without a fuss and the Vega's "RADEON" and "R" logo were both illuminated. I also took notice of the GPU "Tach meter" that is, I think, a very neat feature. There was no signal on my monitor so my heart sank as the card as sold as a working example. Not being deterred, I grabbed a display port cable and and tried one of the display ports and viola, my monitor came to life with a picture. I got into windows, downloaded the latest Vega drivers and checked GPU Z to see if the original bios was still intact and sure enough, it was. 

I got about exploring the options in Wattman only to find out that the manual tuning options with the latest Vega drivers (23.12.1) seemingly crash the driver when set to a manual tuning mode. A quick search online revealed others were having a similar issue so rolled back to the previous gen driver, 23.8.2, and have had no issues adjusting any of the Wattmann manual tuning options. I do hope AMD fixes this glitch but I'm not getting my hopes up as Vega support has been discontinued recently. 

I then ran some tests on Heaven benchmark with a custom MSI afterburner overlay and watched the GPU temp, Hot spot, HBM temp, clock speeds, and gpu wattage while also looking over HWInfo64 to see some of the finer temperature and power measurements. Everything seemed to be running normal and I took note that this card likes to run warm but this seems to be an item of note with the Vega reviews back in the day. After running some more Heaven benchmarks, I paused for the evening and would continue in the morning. 

The next morning, I decided to take apart the card as the rust spot on the I/O plate combined with the punched warranty stickers made me wonder what may be under the shroud. I cautiously got to work and removed the shroud screws and lifted the shroud off to see that the fan was quite filthy with dust but nothing out of the ordinary. I then removed the backplate and noticed that the PCB was in good condition with no visible signs of damage. The next step was to loosen the vapour chamber mount screws and the metal shroud frame screws followed by the three fan screws. I removed all three and then was able to remove the I/O plate to complete the disassembly. 

What I found underneath confirmed my initial suspicions that this Vega was neglected at some point and not taken care of properly. The expected dust build up aside, the back of the I/O plate had rust freckles present. This may or may not explain why the HDMI port was not working although I could see no corrosion around or inside the HDMI port. Two modules on the PCB, just forward of the blower fan, had a significant amount of corrosion present on the top and sides. The one thing that shocked me was the lack of thermal paste on the heatsink and on the Vega chipset as you could see areas that were not even covered by the thermal paste. 

With the initial diagnosis completed, the real fun began as I meticulously cleaned the accumulated dust off the blower fan. Next was cleaning the heatsink and removing the old thermal pads and replacing them with newly cut ones. I used Thermalright extreme odyssey 1.0mm pads for this job after measuring some of the pads. This measurement wasn't easy as the original pads simply disintegrated when touched. After that was done, the remaining heatsink was wiped down and cleaned. Now the careful work began with removing the rust from the affected components on the PCB. I used a little bit of steel wool to very carefully polish the top and sides as best as I could. After a few minutes, the components were looking much improved. This cleaning continued with the back of the I/O plate and it too looks much better. The next part was cleaning the dust and old thermal pad compound from the VRMs and components that surround the Vega chip. With a clean PCB, I moved next to removing the old thermal paste from the chip and the vapour chamber, being careful not to rub the sides of the chip where the interlace connect is located, and applied a thin layer of fresh Noctua NT-H1 paste on the chip and the vapour chamber contact plate.

Now, the fun part, reassembly ! I carefully put all the pieces of this Vega puzzle back together again, making sure the fans and LED light cables were plugged in properly. I added a few small plastic washers to the main GPU tension screws as this seemed to work nicely with the RX 5700XT as well as some smaller plastic washers under the fan screws to make sure it wouldn't vibrate (as my 5700XT blower fan liked to do before I fixed that issue). The left over pieces of thermal pads were applied on the back of the PCB to give some contact with the metal backplate, and then the backplate was resecured. The Vega was again assembled and ready to tackle some new challenges.

The moment of truth came when I put the freshly cleaned Vega back in my spare PC, plugged in, only see the picture cut out of my back up PC's monitor. Some quick detective work determined that the HDMI port on the monitor was loose and not giving any signal. I moved over to my main 1440p monitor and plugged in the display cable to be very happily surprised that the desktop was shown and the Vega was still recognized. Since that moment, I have been typing out my experiences here as my Vega 64 is faithfully running through multiple Heaven benchmarks at 1440p high settings in order to break in the new thermal paste and thermal pads, while at stock settings. So far, the results look encouraging. The card, albet being close to the hot spot limit, does not appear to be thermal throttling and all the temperatures look good. Now the interesting part comes with some undervolting and over clocking to see what this card can do. I have had good success at taming the 5700XT reference design, with the help of previous posters online who shared their results, and that card runs quite well. I am hoping for similar success with this Vega 64.

I am very pleased with the look of the card and, for a 2017 GPU, it is still holding up strong. My only wish is for AMD to fix that final Vega drivers as they are glitching when manual tuning is selected and are not reliable in that sense. 

I hope you all enjoyed this lengthy post about saving a forgotten Vega 64 card but I had enjoyed the experience and it will be a nice card for my AMD collection.

Edit: Sorry the pictures are out of order. I tried to re-order them properly but they were not saved. 

[BigAl01]

Thanks for sharing your experience with cleaning this older generation card.  I put one of these into my daughters computer and undervolted it a smidge to make it run cooler.  She has been happy with it for several years.  

This Vega 64 GPU was installed into 'Monolith', a name I came up with because the case was gigantic.

[Duron999Mhz]

Hey Al,

Thanks for pulling my topic out of the spam bin. 

I am wondering, would you share your undervolt settings for the P states ? Undervolting this card does not seem to be as straight forward as the 5700xt and the best success I've had so far in taming my Vega is to use the power saver preset that seems to limit the PPT to 165W and the core frequency between P4 and P5 at that power level. 

[BigAl01]

That was back in the 2018 era.  I don't remember where I got the information for my settings, but give this site a look.  There are a few other references on the Internet.

[Duron999Mhz]

Hey again, 

Thanks for the link, looks like it will get a good read that may explain some of the quirks that this GPU seems to exhibit. 

I did take my Vega apart again today as I noticed a slight PCB flex and I wanted to make sure everything was seated properly. I did take the oppourtunity to repaste the GPU, this time only putting paste on the GPU itself and it looked like I had put a little bit too much the first time. I also removed the washers from around the tension screws and fan base to eliminate the possibility of that causing any flex. For thermal paste, I decided to try a thin layer of Arctic MX4 to see how that would do. 

I am happy I did take the GPU apart again as after I was looking around the PCB for any indications of something causing the flex, I did notice some light corrosion around the sides of the VRMs that are around the GPU core. I very carefully cleaned the sides of the ones I could reach and began the reassembly process. I took it slow and made sure all screw holes were carefully aligned, including the vapour chamber, before tightening anything down. After reassembling the GPU again, the flex is still there but it is reduced some so I'll take that as a win. 

At present, I have been breaking in the recently reassembled Vega, and new thermal paste, on Heaven benchmark at 1440p high settings and the stock Vega 64 profile. The test has been running for close to an hour and so far, it looks like the temps are consistent and in a good range. Here are the results that I have noticed:

GPU Temp: 75 deg C
Hot Spot: 89 deg C
Core clock: ~ 1370Mhz (on average, it sits around here though it will boost up to 1450Mhz somewhat frequently)
HBM2 Temp: 80 Deg C
Memory clock: 945 Mhz
GPU Watts: 220 W
Fan Tach: ~ 2600 RPM

I know the card is hitting the power limit of 220 W though I still haven't quite figured out if the flucuations in core clock (from 1370 mhz to 1450 Mhz range) is tied to the GPU, hot spot, and / or HBM temperatures in someway. I did read somewhere that the temperature values for Vega are lower than what we see on RDNA so perhaps this is having an effect on it. That is only speculation though at this time until I can read up more on the clock behaviour. 

I will say though, I do like the blower fan design. I currently have the Vega in a mini tower, on an mATX board with a PCIe M.2 expansion card right below it. Thanks to the blower fan, it is forward and clear of the expansion card that would otherwise block a fan or two in a dual or triple fan configuration. The blower card also pushes the hot air out the back of the case compared to my RX 6800 that dumps it into the case. The Vega card, like my 5700 XT and RX 6800, is a dual slot card so it is perfect for a smaller build. Putting my 5700 XT and Vega 64 side by side, the dimensions are nearly, if not, identical, albeit without the "dent" that is present in the side of the 5700 XT, something about "Bending the rules" as part of the marketing for the RDNA cards I believe but still odd none the less. For those curious, I attached a picture showing the two cards side by side. 

For now, I got some reading to do to see if I can further tame this power hungry Vega. 

[BigAl01]

I think you will benefit from a slight undervolting of the P states.  Keep us posted on your progress.

[Duron999Mhz]

I had a wonderful afternoon benchmarking my Vega 64 after reading the posts from the link you shared, Al. I now have a better understanding of the profile settings when you change them. 

I was, however, not able to replicate the same clock settings as the OP as it soft crashed my driver right away during any benchmark load. I also was not able to go below 950 mV for the memory controller as it would downclock to 500 Mhz at 900 mV or for anything between 900 and 940 mV, it would stay put at 800 Mhz. I did, notice, with 950 mV, I was able to bring the HBM2 memory frequency up to 1020 Mhz. Anything above that frequency would have the HBM2 downclock to 800 Mhz. 

With my HBM2 overclocked and 40 minutes of testing applied, I then moved onto my P states. Again, I was not able to use 900 mV as it crashed the driver and caused some weird memory frequency fluctuations where the HBM2 would jump between 800 Mhz and 1020 Mhz frequently. I then fixed P4 to P7 at 950 mV and and down clocked to 1500 Mhz on P6 and 1550 Mhz on P7. I then started increasing both of the P6 and P7 clocks by 10 Mhz and tested for stability in heaven benchmark. After a couple hours of testing, I was able to find a sweet spot that seems to be holding up nicely. Here are the results over two hours of Heaven benchmark at 1440p high quality:

GPU Temp: 65 deg C
Hot spot: 75 deg C
Core clock: 1460 Mhz
HBM2 Temp: 71 deg C
Memory clock: 1020 Mhz
GPU wattage: 163 W
Fan tach: 2150 rpm 
EDIT: PL set to + 50%

Overall, I found some interesting results compared my stock test:
- A reduction of power used of nearly 35% 
- A 6% improvement in core clocks
- A nearly 8% improvement in memory clocks
- A 15% reduction in core temperature
- A nearly 19% reduction in hot spot temperature
- A 12.6% reduction in HBM2 temperature
- A nearly 21% reduction in fan RPM

Overall, a worthy endeavour as this undervolt and overclock profile beats the default power saver while operating at the close to the same PPT limit that the power saver preset applies. 

I haven't played around with the HBCC option yet. The one video on youtube by Iceberg Tech, doesn't show a meaningful gain for most games unless the 8GB of vram is met or exceeded. Has anyone else tried this feature out yet ?

Overall, I would say this was a very fun and relaxing afternoon with my Vega, maybe I will find a Radeon VII to play around with but used prices are near to what I paid used for my RX 6800 and that is not encouraging right now. 

[BigAl01]

Good documentation and test method.  You were successful and that RX Vega 64 should last you a while.  This should be an incentive for others to follow with their older AMD graphics cards.

[mengelag]

That's awesome! Almost a retro GPU as this point. 

[jaandoh]

Well I throughly read this post, and I wish I had the patience to do as you have done and restored your Vega 64.

I do so love my Vega 64 too, it has worked hard gaming without relief or fault for many years, 6 or 7 I think...

I do t even comfortable enough to clear the cooler fan lol (I'm such a scaredy cat Hahaha).. 

You have made me think about cleaning it, but guess I will just try that compressed air stuff... 

Thank you for a lovely detailed post too, although that tweaking is far above my head, I loved reading about it, but don't feel confident in myself tweaking stuff I don't understand, I just tend to run stuff at stock settings lol... 

I liked all the photos too, which helped give a visual understanding of all the conclusions... 

PS. 

Where does the rust speckles come from? Or how are they created? 

[Duron999Mhz]

Merry Christmas everyone ! I thought I would share an update regarding my Vega 64 adventure. 

I am still working on stability testing for my undervolt and overclock profile. I did lower the P6 and P7 clocks by 10 Mhz and I will be running some tests this afternoon along with adding in some pc games to check the gaming performance. 

I did re-try the AMD official 23.11.1 drivers again, with the use of DDU in safe mode to clean off the 23.8.2 drivers being previously installed. After a clean restart, I loaded up the Adreneline software and attempted to load my profile and noticed it wasn't loading the presets on the tuning page. I then manually enabled the presets to see that some of the settings had loaded with the exception of the power limit slider and the memory overclock. One glitch I did note was when the advanced voltage control was toggled, the memory frequency slider and values visually disappeared. I then attempted to load up Heaven benchmark and the driver soft crashed while the benchmark was loading and not actually running.

Somewhat annoyed, I performed a DDU uninstall of the drivers via safe mode and installed the 23.8.2 drivers again and performed a restart afterwards. I loaded up my profile, which correctly loaded the first time and showed all the values, and ran Heaven benchmark which loaded up and ran perfectly fine.

Something is definitely messed up with the last release of the Vega supported drivers from AMD with regards to the tuning page and loading custom profiles. It got me thinking about the third party R.ID drivers I had seen mentioned online and folks say have active support enabled for SAM and driver refinements that can enable some improvements when trying out the newer game titles. After some testing on 23.8.2, I was thinking about giving these third party drivers a try, has anyone had any experience with these drivers and older AMD GPUs ?

Jaandoh, your questions regarding the rust freckles, my best guess is the GPU was kept in a humid environment and perhaps not a sealed container after it was last used or was simply operated in a humid environment for a long time. The last thought may explain why I noticed the most corrosion on the pcb components that were ahead of the blower fan and on an area of the PCB that has very little to any airflow through it. The remainder of the corrosion was more spreadout on the pcb components around the GPU core and not as concentrated on any one spot. This may be explained by being in the air flow path from the blower fan, which always provides some air flow in this area. The previous time I did take the GPU apart to investigate the PCB flex, upon closer inspection, I did notice some corrosion around the base of a component just forward of the HDMI port. I wonder if this is the reason for the HDMI failing to display a signal but sadly I do not have the equipment nor expertise to test out this component. Regardless, the other three display ports are working just fine. 

You had also wondered about the dust accumulation on the blower fan. I did try some computer duster and after giving the fan a good blow through (pardon the pun), I noticed it didn't do anything significant. After cleaning all the fins with a Q tip to loosen the dust, the computer duster did help blow out the loosened dust deposits. 

With regards to the blower fan, I actually do like it. From a performance perspective, it achieves the goal of getting alot of airflow through the card quickly and at very high RPM. I noticed the max RPM to be around 4700 RPM. Although being quite audible, it cooled the GPU down right away compared to my RX 6800 where the fans will max out at around 2900 RPM. The cooling solution on the RX6800 is about twice the size of the Vega and the 5700 XT but I do like the hot air is pushed directly out the back of the card. It made me think about the thermal limits for the display port cable having hot air being pushed out it for a longer period of time, something I may have to look into. With that being said, I did notice an overall lower operating temperature in my computer case when using the Vega or the 5700 XT compared to my RX 6800 (where the exhaust air of the GPU is pushes against the motherboard and the side of the case). It got me thinking about any future GPU upgrades and looking for a GPU that has an open back plate and exhaust air direction towards the rear of the case. 

Something I did notice while using the Vega and the 5700 XT with regards to their blower fans. The fans are identical models for those wondering as I compared the pictures of the model number from my 5700 XT disassembly and my recent Vega 64 disassembly. The model number is "BFB1012SHA01" made by Delta electronics. It may be possible to swap fans if you having issues with one or the other but I have not tested this as it would involve a repasting of one or two GPUs to accomplish. My 5700 XT fan does have a low but audible hum at low RPM that my Vega does not but otherwise it works perfectly fine. 

I did notice the blower fan behaviour that AMD must have programmed in. The blower fans will pulse every 3 to 4 minutes by 200 to 300 RPM and slowly decrease back to a steady RPM. Perhaps this is done to ensure no hot air stagnation in critical areas ? Originally I thought this was only happening when you approach a temperature limit but it was happening with my undervolt profile where I am no where near any limits other than lower power limit. I also noticed, when I was finished testing and ramping the fan RPM up to max to desaturate the heatsink, that the blower fan will very slowly decrease RPM from max back down the fan curve versus the instant response I noticed with my RX 6800 fans. It does take a few minutes for the blower fan to come down to the minimal RPM but your GPU is definitely cool afterwards. 

On a similar note would anyone like to see pictures of my RX 5700XT disassembly and cleaning ? I would be happy to make a separate post for that if anyone is interested. 

[jaandoh]

Thank you, for your detailed description of the source /cause of the rust speckles... 

I don't know about the other members but I find a lot of the restoration threads inspiring and empowering,  maybe because I'm getting lazy, in my old age (58), I 'm always saying... One day when I get time, I'd like to do that to my computer. .. 

[Duron999Mhz]

Hey, I am glad you enjoyed it. It has been an interesting experience for me but I do find it fulfilling to bring back these old flag ship cards. My 5700XT is still going strong as a 1080P gpu and it shows you these previous generation gpus have some life left in them.

At the moment, I am waiting for my spare pc's monitor to return from RMA and then I can continue the testing process for the Vega 64.