Vega owners club

[delshay]

Here's the latest superposition benchmark for my Vega 56 Nano. After the capacitors upgrade card has improved when compared to post #485.

I'm still learning new things about this card & this is what I know sofar. This card runs hotter than any other nano card. This is because changing the capacitors on the other side of the card to 820uf 3 mohms, the resistance to the core is lower, so the MOSFETS are running hotter. This is normal if you fit very low ESR capacitors on the output.

I just remembered after seeing a posting on YOUTUBE that the HBM memory sits slightly lower than the core. This explains why my HBM is running 2c hotter than the core. But here's the kicker. Those users including my card that have the molded fill-in on the interposer are actually at a disadvantage. The reason being i'm not able to sand down half of the coldplate the lower core contact as the mold will prevent the heatsink from sitting lower. Those user that have the unmolded version can sand down half of the heatsink (core side) to get lower temperature on the HBM.

Anyway here's my latest benchmark score with max fan speed

 

[delshay]

This is my second Vega 56 Nano which will be used as a backup card. This card suffers from coil whine, so it's going to be interesting if upgraded capacitors can fix this. Unlike my first Vega Nano in this thread which has Panasonic/AVX capcitors this card is being modded with it's first of two modifications with all-out Kemet capacitors.

This is probably the first GFX card to use Kemet 3 mohms Caps. These were released in the year 2021, but the full stack from the PDF Docs has not yet reach the market. The focus on this modification in not just on the ESR, but the "leakage current" of all new capacitors fitted, not forgetting what can be done about the coil whine.

Before any modification this card clocks higher than the other card in both core & HBM, but overall the other card is faster because it's able to hold a steady clock speed. It's going to be interesting what changes (if any) what those Kemet capacitors can do.

Please ignore the soldering of those capacitors in photo as this was done without taking the card apart & not breaking the seal. Soldering will change & will be cleaned up once the high-Side capacitors arrive in the next few days.

I will install this card tomorrow to check performance & coil whine.

 

[delshay]

Latest update 2nd Vega Nano Hardware Modding.

There's mostly bad news in this posting, but there is some good news.

First this card does not suffer from coil whine, it was the fan when rotating at high speed.

Secondly when taking the card apart for the first time there's is a dead spider inside the card along with spider web. Original thermal thermal paste can also in the same photo. ..see first photo.

Thirdly I took a gamble with the five of the capacitors thinking I could squeeze them under the cooling plate, I was wrong, so I needed to trim part of the Capacitors Housing. This is where you can see the exposed copper on top of some capacitors. NOTE: This opens the door for experimenting with extra flat braided wire for lower ESL. ..see second photo.

Fourthly here come the most devastating news. When cleaning off the old thermal paste I notice something strange about the core. I've seen this strange thing before on reviews samples & end users CPU/GPU. I took a closer look under magnification & notice a scratch. With a pair of super pointed tweezers I move it across the scratch & to my worse fear this is in-fact a fracture in the die. This means at some point in the future this card will have a nasty death by blowing out the MOSFETS. There's no getting around this. Heat cycle will kill it at some point. Sorry no photo at this time.

The only good news to come out modding this card is it no longer trips the Radeon Driver at extreme maximum temperature. I can run the card at maximum temperature for long period of time & it just throttles. This is about running the MOSFETS at destructive levels of high temperature but the card refuse to trip. So the question is, is this supposes to be normal interactive behavior of Radeon software/hardware. If so then my first Vega card is set-up incorrectly, but it also applies to my R9 Nano.

Please remember what this card modding is all about, it's main focus is on, "Leakage Current" of capacitors & I will need to confirm if this has changed the card behavior.


EDIT: Added very early benchmark

 

[Muck Muster]

A couple years ago, I modded the reference cooler on a Vega64. I bought this huge Raijintek Morpheus 2 cooler and ran into a similar problem of the uneven HBM/GPU die. I couldn't keep the GPU/HBM cool at even temps. At times it was 15c higher than the GPU. I was getting tired of this Vega card till read where someone had used the original mud that came with the cooler. This guy said to paste it 'moderately heavy'. Yahoo. This did the trick. Apparently Raijintek knew the problems associated with the Vega series but failed to disclose this anomaly. The GPU/HBM off-space was so fine a difference in height, it was hard to find a milled copper shim that thin (I was getting my copper from an auto parts dealer who told me to go find a jeweler who deals with hobby metals). In 2019, Raijintek stopped offering the Morpheus 2 cooler for the Vega series. Shame.
But anyway, the other heat problem I encountered was the chokes would heat up past 90c. I applied aluminum fins but got less than a 10c difference and would fall off with attaching thermo pads. Then I used copper fins with thermo glue and that made a 20c difference. 75/80c was acceptable. I was absolutely enamored by my undertaking. I felt like James Brown! And I got carried away. While venturing to file the top of the heatsinks under the mounting bracket arms, I busted the choke top clean-off. Bummer.
I was able to re-glue the choke together but to make a long story short, I could only run it at stock settings. And it was never cool again.
I politely coerce a friend to sell me a lack luster Gigabyte 2080 S which soon replaced it by an EVGA 3080ti.
I've been too disappointed to do anything more. To be honest, I I think the Vega's resolution was far better at rendering in that it was beautifully smooth, concise and detail at higher clocks. Tuning up the EVGA 3080 ti, (not to be kidding it's fast) had a slight mushy quality. I had some minute screen tear. Couldn't adjust it on a Freesync monitor

 

[userdmn]

Hi, Im a new radeon vega card owner, these are my results

This "was" a msi vega 56 air boost card, but I cut the internal plate "heatsink" for better external airflow plus a new 120mm fan (from CM hyper 212 led).
-VBIOS msi 231, no OC applied, just 50% extra gpu power limit.

Im happy with this card

 

[Muck Muster]

Hi, Im a new radeon vega card owner, these are my results

This "was" a msi vega 56 air boost card, but I cut the internal plate "heatsink" for better external airflow plus a new 120mm fan (from CM hyper 212 led).
-VBIOS msi 231, no OC applied, just 50% extra gpu power limit.

Im happy with this card

View attachment 290059View attachment 290060

"These are the voyages of the U.S.S. Enterprise.. . to boldly go where no man has gone before."

 

[userdmn]

@Muck Muster Ye looks like a Ship lol but the thing is this still heating up xd, this card still needs some upgrades, the big improve is that i remove the blower fan, oh god that thing.

 

[Muck Muster]

@Muck Muster Ye looks like a Ship lol but the thing is this still heating up xd, this card still needs some upgrades, the big improve is that i remove the blower fan, oh god that thing.

Heat. That's the problem I had. All the vram, MOSFETs, voltage regulator modules (vrm) and chokes need to be cooled. At first, I bought a bunch of small aluminum heatsinks and tack them on with double face thermo tape. They fell off. So I used thermo glue.
The thermo glue worked O.K. but the aluminum heatsinks didn't do the job. So I bought copper fin heatsinks. I had to shave off the tops of the copper heatsinks on the chokes closest to the gpu (the gpu's cooler bracket was in the way). I use a heat gun and acetone to eat off the thermo glue. But I busted off the top of one of the chokes. Bummer. I was able to glue the top back on but it didn't work as well. I'd have to un-solder and weld in a new choke. That would be time consuming. Ugh.
All of the aluminum heatsinks I replaced with copper heatsinks did well under the infa-red thermo gun except for the one I re-glued choke (which was almost 20c higher than the rest). At best, I was able to run 'zero' base clock speeds and mem. Not so good.
So I called a friend who sold me his old Gigabyte Nvidia RTX 2080 S (Wow. Wasn't a good graphics card at all.) The Vega64 had much better visual (eye to screen) quality resolving details, speed, texel and pixel rate. I was blown away by that. Everyone who says 2080 was better card than Vega hasn't played with both.
I think it has something to do with Vega's monstrous 2048 bit bus width. It packs a lot in the picture. The Nvidia 2080's 256 bit bus width is way too small. That's where I think the Vega series really shines.
So I fished around for 6 months for the best graphics card deal available. At the time I was hoping for an AMD RX 7900 XT which retailed $1000, had a 380 bit bus. But I just happen to find an EVGA 3080ti 384 bit bus for $600 and couldn't pass on it. Gave the 2080 S to friend who put it in his VU BF3 server. Truth be told, servers don't need a 2080 to run.

These are the heatsinks I bought for the Vega64

Alphacool GPU RAM Copper Heatsinks 10x10mm - 10pcs (17426) - Newegg.com

Buy Alphacool GPU RAM Copper Heatsinks 10x10mm - 10pcs (17426) with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

Alphacool GPU RAM Copper Heatsinks 6.5x6.5mm - 10pcs (17425) - Newegg.com

Buy Alphacool GPU RAM Copper Heatsinks 6.5x6.5mm - 10pcs (17425) with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

Newegg always runs out of these small copper heatsinks so you may have to find an electronic parts E-store or the manufacturer's website. The thermal glue I used was Gennel 10 gram Thermal Conductive Glue (I don't recommend this brand. It's just O.K.).

Long live Vega.

 

[userdmn]

Thanks for the advice, I was really thinking about it and the best thing I could do is buy a waterblock for this card, even so I plan to adapt 2 fans with better cfm and put a good thermal paste.

 

[LabRat 891]

Hi, Im a new radeon vega card owner, these are my results
This "was" a msi vega 56 air boost card, but I cut the internal plate "heatsink" for better external airflow plus a new 120mm fan (from CM hyper 212 led).
-VBIOS msi 231, no OC applied, just 50% extra gpu power limit.

Im happy with this card

Nice!

Mine:

It's a Radeon Instinct MI25 re-flashed into a (single mDP output) Radeon PRO WX 9100
16GB HBM2 w/ a full-fat Vega 10 GPU
My post about it in ghetto mods.
Sadly, I'm having some problems getting OverdriveNtool to work.
Though, even 'stock' it got me a Legendary score on 3dm

 

[userdmn]

Nice!

I found out about the mi25 and its bios mods a bit late, it would have been an option before a vega 56.

How good is the airflow?
Did you get good temperatures?

 

[Muck Muster]

Waterblocking. May be a better idea. Modifying the air cooling is b!tch. A lot more work than it looks. It's gratifying to see 36c/40c temps playing CoD ww2, Battlefield ww2 or Fortnite . Ramping up two120mm was too much fun. It's like eating too much ice cream. You want more. When you're ramping the bios and changing the cooling properties, you're not doing it for sake of longevity and redeeming the value at the end of it's life, lol. You're doing it because it's fun. Such is off-roading a brand new Jeep. Haw, haw. It's too much fun.

Nice!

Mine:

It's a Radeon Instinct MI25 re-flashed into a (single mDP output) Radeon PRO WX 9100
16GB HBM2 w/ a full-fat Vega 10 GPU
My post about it in ghetto mods.
Sadly, I'm having some problems getting OverdriveNtool to work.
Though, even 'stock' it got me a Legendary score on 3dm

Man, Lab-Rat, you're a man on a mission. This looks like fun. Once upon a time, I visited Dan Gurney's American Racing in Santa Ana, CA (1992) when they just finished winning the IMSA championships with the Toyota Eagle Mk III. Dan wasn't there but one of the cars's mechanics was. He told me the most important tool they use was duct-tape. Haw, haw!
Jeff Gordon who raced in NASCAR said something similar. "Duct-tape. You gotta do what you gotta do." Winning (or succeeding in the goal) is what matters no matter how janky it looks. I wish I could show you under the hood pictures. Lol.
Here is an article on the race car.

1992 Eagle Mark III

1992 Eagle Mark III The Toyota Eagle Mark III is one of the true legends of American motor sport, achieving a record-breaking performance in IMSA racing in the early 1990s. As well as being an all-conquering machine, it was also the last of its kind: the US sports series subsequently moved away...

www.supercars.net

 

[userdmn]

about the fans, I'm seeing these from ALI of video cards that include a frame, shell, housing, etc. they are usual 88mm or 95mm, dual or triple.

 

[kapone32]

My 2 cents on the conversation of Air vs Water cooling. The most mitigating factor for Water is the fact that the Memory and GPU are basically on the same die. That means that you will see that Vega cards on water do not have a huge variance between the GPU and Hotspot temps on water. When I had a block(s) on my Vega 64 crossfire array the temps never went above 60C under any circumstance including Games that gave you 90% more performance. As a result both of those cards are still rock solid and run at 1650+ MHZ. Funnily enough what got me into water cooling was the fact that as of the Vega series cards Sapphire has suffered a bug with their fan profile and that zero fan policy did not help the situation in anyway.

 

[LabRat 891]

BTW, that's not "duct / duck" tape, it's Aluminum tape (commonly used in ducting . Yeah... it's confusing).
The idea was to re-create the airguide that was on the card, but also hold the server fan in place (emulating the cooling the card would get in a GPGPUrack.)
(Alu tape is easier to remove vs. duct tape too)

It does work, and with the WX9100's lower power limit, it's cool and quiet. However, trying to get more performance through powerplaytable modding, I am seeing limits.
I have absolutely no idea how the MI25 was stable in its intended use, its PPTs are VERY aggressive with voltage; manually-copying the Pstates into a VegaFE PPT causes near-immediate thermal-limit-hit and crash, even w/ max fan. (I've been playing with various voltage and clock settings, and the best I can get so far is on a stout undervolt.)

I might have to find my NZXT Kraken G12 (assuming it has compatible mounting pitch It doesn't. The mod isn't difficult, though.) and mount my old H110i (from my pre-ryzen 3770k build) on the card. I simply do not have the case space for modding on a highTDP heatsink. (the h110 didn't have a ton of hours on it before I upgraded to Ryzen. It's been in storage since, and I'm hoping it hasn't degraded from age)
This is becoming an adventure...

My 2 cents on the conversation of Air vs Water cooling. The most mitigating factor for Water is the fact that the Memory and GPU are basically on the same die. That means that you will see that Vega cards on water do not have a huge variance between the GPU and Hotspot temps on water. When I had a block(s) on my Vega 64 crossfire array the temps never went above 60C under any circumstance including Games that gave you 90% more performance. As a result both of those cards are still rock solid and run at 1650+ MHZ. Funnily enough what got me into water cooling was the fact that as of the Vega series cards Sapphire has suffered a bug with their fan profile and that zero fan policy did not help the situation in anyway.

I'm not looking forward to it, but the bolded sections make the effort make sense. Seeing much lower Temp Targets on the LC FE cards makes a lot of sense suddenly.

 

[DR4G00N]

@LabRat 891 What's the CFM rating on your fan? I can't remember where I saw it but IIRC it requires at least 40 CFM through the heatsink.

I used a pair of sanyo 60mm 5600RPM fans totaling 52 CFM on mine prior to watercooling it and worked alright but even with those fans the amount of air getting through the heatsink was not exceptional.

 

[userdmn]

great improvement, today I did the "heatsink lapping" and the temperatures dropped a lot, in farcry6 (at Medium settings) I had temperatures
of gpu 72, hmb2 73, and now I was amazed because I played about 30 minutes at gpu 52, hbm2 54, we are talking about a reduction of 20c!

Before and after

 

[delshay]

great improvement, today I did the "heatsink lapping" and the temperatures dropped a lot, in farcry6 (at Medium settings) I had temperatures
of gpu 72, hmb2 73, and now I was amazed because I played about 30 minutes at gpu 52, hbm2 54, we are talking about a reduction of 20c!

Before and after
View attachment 290270View attachment 290271

That's something on my to-do list. I just checked my vapour chamber is not flat. it's pretty bad in-fact.

what did you use to lap it?. ...glass surface is best but I don't have any flat glass small enough to fit between the screw holes that hold the x bracket. Talking of x bracket, I swapped mine for the old type which has more mounting force.

EDIT: Also what is your "hotspot" temperature? Here it's normally about 6 - 8c higher than the memory.

 

[userdmn]

@delshay
You need to cut 4 pads, sanding (adding a bit of water) 15m per one, 1hr total, then i use a polish compound, 5 minutes.

I also didn't know what to use to polish and ended up using this lol

Pic

B4-A909-AF-470-E-4-AB7-B3-CB-91-B8714-A43-AB hosted at ImgBB

Image B4-A909-AF-470-E-4-AB7-B3-CB-91-B8714-A43-AB hosted in ImgBB

ibb.co

 

[delshay]

@delshay
You need to cut 4 pads, sanding (adding a bit of water) 15m per one, 1hr total, then i use a polish compound, 5 minutes.

I also didn't know what to use to polish and ended up using this lol

Pic

B4-A909-AF-470-E-4-AB7-B3-CB-91-B8714-A43-AB hosted at ImgBB

Image B4-A909-AF-470-E-4-AB7-B3-CB-91-B8714-A43-AB hosted in ImgBB

ibb.co

What's delta temperature between memory & "hotspot" under load?

EDIT: I'm looking at the glasses in my kitchen, not sure if I can find anything that can help. ...EDIT 2: Just found a glass with flat bottom, ..testing.

 

[userdmn]

I am waiting for a new thermal paste, at the moment I cannot benchmark and confirm temperatures because the one I am using only lasts a few hours, it is from an expired batch.

 

[LabRat 891]

@LabRat 891 What's the CFM rating on your fan? I can't remember where I saw it but IIRC it requires at least 40 CFM through the heatsink.

I used a pair of sanyo 60mm 5600RPM fans totaling 52 CFM on mine prior to watercooling it and worked alright but even with those fans the amount of air getting through the heatsink was not exceptional.

Based off the Amp rating compared to other 120x38mm fans I am familiar with, it should be north of 150CFM. I don't have the spec @ hand, but based off experience, it is DEFINITELY well over a 100CFM fan (in open air).
(I chose a 120x38mm fan, as they develop more static pressure, and used the non-porous Al tape to create a pressurized airguide)
Again, only speaking from experience, it feels like more than 40cfm coming out the exhaust of the card. (As compared to 35-55cfm 80-92mm fans I've played with)

 

[delshay]

Here's my Vega Nano Settings. ...With this settings card is immune to crashing AMD driver regardless of how high voltage is set. If voltage is set too high, core will just hit max temperature 85C faster & throttle. The picture below will show card is running at minimum voltage, but the question is, can Vega cards run below 800mv, because this is what required here.

NOTE: Increasing core voltage will increase clock speed even with something as small as 10mv. It can be increase all the way up-to 1.2v, but it will never crash the driver.


Core 1405MHz 864mv ....(max top speed 1425MHz with just 864mv)
HBM 900MHz ....max top speed 956MHz)

 

[LabRat 891]

Here's my Vega Nano Settings. This settings card is immune to crashing AMD driver regardless of how high voltage is set. If voltage is set too high, core will just hit max temperature 85C faster & throttle. The picture below will show card is running at minimum voltage, but the question is, can Vega cards run below 800mv, because this is what required here.

NOTE: Increasing core voltage will increase clock speed even with something as small as 10mv.


Core 1405MHz 864mv
HBM 900MHz (max top speed 956MHz)

Holy crap, assuming that' truly stable and performance-increasing those are some insanely good V/Mhz ratios.
Looking @ the MI25's stock PPT and the ES Vega 64's, (along with my problems getting anywhere near 1ghz core @ 1050mv) it looks like the MI25s might use ES- or early- revision silicon.
Simply loading an 80c temp-target VegaFE PPT causes performance regression or crashing.

MI25's extracted PPT:

Vega64 ES' extracted PPT:

WX 9100 conversion, stock PPT:

VegaFE (80c) PPT:

My 'WX 9100' card really doesn't seem to want to get out of p2-p3, or is clearly less-performing/unstable when using any unmodified 'Vega' power play table. I've spent literal days trying to 'figure this out' and only managed to 'change something': Now my card wants to always (deeply) undervolt itself (according to GPU-z, which had been reliable), regardless of PPT used.

 

[delshay]

Holy crap, assuming that' truly stable and performance-increasing those are some insanely good V/Mhz ratios.

It's 100% stable regardless how high voltage is set. This is how the card is tested.

Fan is manually stopped by my hand until card reaches 85C. Fan is still not released until card hits 87C. At theses temperature the card will throttle hard between 400 - 500MHz core. This is how it is tested for raw stability. The fan is then released by my hand to cool the core, but by this time the core has hit 88C. Not a single crash, just hard throttling.

END USERS SHOULD NEVER COPY THIS TEST METHOD