Vega owners club

[TheoneandonlyMrK]

That makes no sense whatsoever lol.
Did a few runs and for those that own a Zenith Extreme + Vega...do some testing with AI Suite installed and not installed. I had a massive fps improvement by uninstalling that software. Major improvement, mins from 18 to 65 fps and average from 42 to 65 fps.

While I have not got one i had ai before, its terrible, totally, it used to shut my pc off often due to sensor glitches while using other software with it, and this doesn't surprise me either.
In general the software packaged with stuff is usually totally borked due to lack of support regardless of os and updates or will be soon.

 

[springs113]

While I have not got one i had ai before, its terrible, totally, it used to shut my pc off often due to sensor glitches while using other software with it, and this doesn't surprise me either.
In general the software packaged with stuff is usually totally borked due to lack of support regardless of os and updates or will be soon.

Yea but on my x99 platform it performs real well especially the fan xpert. I'll sum it up as the platform being new.

I ran Vega til it throttled like no other, i kept getting a warning from wattman about it too...cant wait for my blocks.

 

[TheoneandonlyMrK]

Yea but on my x99 platform it performs real well especially the fan xpert. I'll sum it up as the platform being new.

I ran Vega til it throttled like no other, i kept getting a warning from wattman about it too...cant wait for my blocks.

Might have been one app or setting doing the dead but mine worked well day 1 ,but it didn't support or get support for even win8/
Different boards of course too.

 

[springs113]

Might have been one app or setting doing the dead but mine worked well day 1 ,but it didn't support or get support for even win8/
Different boards of course too.

When I get a chance I will test my x99 board with Vega to see if AI Suite is a problem. I've got the Rampage 5 Edition 10.

 

[acperience7]

Ok, so Vega didn't look too great in reviews, but it's interesting to see what regular users think of it in practice. I'm especially interested in the amount of noise that the reference model makes.

I don't think it's that bad. My system is entirely air cooled. When play with the fan and bump the speed up I can hear it loud and clear, but in actual use I don't really hear it. I' sure I've just gotten used to tuning out my case at this point. I can take some decibel readings for you if you want, but it wouldn't be very scientific. I don't have a legit decibel meter, just a phone app. I run my card at 600rpm - 2600rpm.

 

[qubit]

I don't think it's that bad. My system is entirely air cooled. When play with the fan and bump the speed up I can hear it loud and clear, but in actual use I don't really hear it. I' sure I've just gotten used to tuning out my case at this point. I can take some decibel readings for you if you want, but it wouldn't be very scientific. I don't have a legit decibel meter, just a phone app. I run my card at 600rpm - 2600rpm.

It's ok, no need to take readings for me. If you want to share them for your own sake on here, then fair enough.

Sounds like it's not really noisy enough to bother you and that's great.

 

[terroralpha]

Has anyone tried an EK thermoshepe universal block on Vega? i can pick one up locally for $40, i'm thinking about trying that. i don't want to spend $150 (after shipping) on a full cover waterblock that this card doesn't even need.

EDIT: actually it might not work. the biggest mounting plate EK offers for the thermosphere is 61x61mm. vega's GPU mounting patter looks like it's 64x64mm.

 

[springs113]

Has anyone tried an EK thermoshepe universal block on Vega? i can pick one up locally for $40, i'm thinking about trying that. i don't want to spend $150 (after shipping) on a full cover waterblock that this card doesn't even need.

EDIT: actually it might not work. the biggest mounting plate EK offers for the thermosphere is 61x61mm. vega's GPU mounting patter looks like it's 64x64mm.

Vega needs better cooling, my house ambient temp is 20c, my current case is the lian li pc-011wgx with the side panel off and my card still throttles.

 

[turbogear]

Has anyone tried an EK thermoshepe universal block on Vega? i can pick one up locally for $40, i'm thinking about trying that. i don't want to spend $150 (after shipping) on a full cover waterblock that this card doesn't even need.

I am not too sure if it will even fit as on Vega the VRMs with the high standing inductors are very close to the ASIC.

Even if it fits, I am not too sure if that would be a good block for Vega. I think this block will not cover all the VRMs.
The VRMs on Vega should be cooled actively as it is after all a power hungry card.
As you can see from the installation manual of the EK full cover block VRMs are located next to ASIC and the full cover block covers the VRMs.
https://www.ekwb.com/shop/EK-IM/EK-IM-3830046994325.pdf

 

[TheoneandonlyMrK]

After water blocking my card it seams clear their is only one ref pcb even the founders shares it.
Mine had the packing ,was level though looked retouched but polished and the vrms etc and board are identical to all others I've seen online.

 

[xkm1948]

Now only if RTG opens up the BIOS mod. If one doesn't care about power draw and have beefy cooling he can push it to 2GHz maybe.

 

[springs113]

Vega on a water is a fine ass looking card.

 

[turbogear]

After water blocking my card it seams clear their is only one ref pcb even the founders shares it.
Mine had the packing ,was level though looked retouched but polished and the vrms etc and board are identical to all others I've seen online.

As you put your Vega under water did you see which packaging did your Vega have?
There are two different variants existing.
There is one with epoxy filling around the HBM and GPU ASIC and another without epoxy filling.
The one that does not have epoxy filling between HBM and GPU chip has a small height difference between the HBM and GPU ASIC.
In that case the GPU ASIC sits little bit higher than HBM.

My Vega is the one where epoxy is filled and has no height offset between the memory and GPU ASIC.

Here is a link that talks more about this:
https://videocardz.com/72173/there-are-at-least-three-variants-of-vega-10-gpu-packages

 

[londiste]

got an msi reference vega64.

overclocking, undervolting is very annoying ordeal. nothing really seems to work fine with vega yet, not even wattman itself. and when everything seems to work it is not stable. for example when i tried to reapply the exact same settings in wattman, i got different results every time.

idle power consumption is awesome. it quite literally seems to use no power at all. 32mhz clock at idle with 1-2 monitors is very-very cool.

3dmark graphics scores:

• timespy - 7062 (7407 oc)
• fire strike - 22150 (23439 oc)
• fire strike extreme - 10556 (11270 oc)

some benching against my current gtx1080 at 1440p:

vega64 oc is 1015 mhz hbm, +50% pwr limit, undervolt to something (wattman has 950mv, no idea what it really is), essentially ended up clocked at 1555-1560 mhz in games at somewhat reasonable power consumption (or rather reasonable noise, as this already needed 3400+ rpm from the fan).
1080oc is +150 mhz core /+550 mhz mem

everything except doom, witcher3 and bf4 are built-in benchmarks. multiple runs, within 1-2% of each other, usually at the lower end of that scale. witcher 3 result is basically from riding around in front of novigrad. battlefield4 is from campaign, the first driving section.
before anyone points out, i know doom result looks weird. it is the beginning of the game. ran across the map and had firefight with first set of monsters. i guess that is somehow not a good use case for vega. go figure.

 

[TheoneandonlyMrK]

As you put your Vega under water did you see which packaging did your Vega have?
There are two different variants existing.
There is one with epoxy filling around the HBM and GPU ASIC and another without epoxy filling.
The one that does not have epoxy filling between HBM and GPU chip has a small height difference between the HBM and GPU ASIC.
In that case the GPU ASIC sits little bit higher than HBM.

My Vega is the one where epoxy is filled and has no height offset between the memory and GPU ASIC.

Here is a link that talks more about this:
https://videocardz.com/72173/there-are-at-least-three-variants-of-vega-10-gpu-packages

Mine is the same as yours then.

 

[turbogear]

everything except doom, witcher3 and bf4 are built-in benchmarks. multiple runs, within 1-2% of each other, usually at the lower end of that scale. witcher 3 result is basically from riding around in front of novigrad. battlefield4 is from campaign, the first driving section.
before anyone points out, i know doom result looks weird. it is the beginning of the game. ran across the map and had firefight with first set of monsters. i guess that is somehow not a good use case for vega. go figure.

Thanks for the benchmark results.
What monitor size are you using? 1440p?

With my custom water cooled Vega 64 at 1075mV and 2.5% frequency overclock ( no HBM overclock), the average clock in gaming always hovers around 1620MHz.

In the Witcher 3 I get FPS in range between 65 and 84 depending on scene.
In Battlefield 1 multiplayer mode, the FPS stays in the range from 90 to 113.
For Rise of Tomb Raider with DX11, the FPS is in the range from 60 to 80 depending on diffrrent game scenes.

All the above games are played at Ultra settings on 1440p Monitor.

I read over at Hardocp that Vega does not perform very well with traditional Multi-Sample or Super Sample Anti-Aliasing (MSAA, SSAA).
Vega works better using shader based AA methods like SMAA, or FXAA or Temporal AA, or CMA.

Here is the link to original posting:
https://www.hardocp.com/article/2017/08/14/amd_radeon_rx_vega_64_video_card_review/17

Mine is the same as yours then.

Thanks. I am wondering on the cards where GPU ASIC is slightly heigher than HBM if original heatsink is also cooling HBM efficiently as it will not contact HBM directly but over thicker layer of thermal past.



Due to server problems my original post got lost.

Here I tried to write it again.

 

[terroralpha]

looks like my vega 64 can run perfectly fine at 1000mV core, but needs 1100mV for the memory to get to 970MHz. temp target is 80*C with 81*C being the max. fan speed target is 3250RPM. but i have my Vega in a 0.44 cubic ft mITX case, your temp results should be much better.

overall, power draw went down by 60W (from the wall through a bronze PSU) but performance is up 6.5% : https://www.3dmark.com/compare/fs/13542974/fs/13448004
why couldn't AMD just do this from the factory?

 

[xkm1948]

looks like my vega 64 can run perfectly fin at 1000mV core, but needs 1100mV for the memory to get to 970MHz. temp target is 80*C with 81*C being the max. fan speed target is 3250RPM. but i have my Vega in a 0.44 cubic ft mITX case, your temp results should be much better.

overall, power draw went down by 60W (from the wall through a bronze PSU) but performance is up 6.5% : https://www.3dmark.com/compare/fs/13542974/fs/13448004
why couldn't AMD just do this from the factory?

Not every piece of silicon on the wafer has the same quality. RTG using higher default core voltage so they can utilize as much of the wafer as possible.

Plus, stability is relatively speaking. For my FuryX i can undervolt to 1220mv from 1243mv. Can it play games? Sure. Can it pass FireStrike Ultra stability test? Not even once.

As for you i would highly recommend you do some stability test to find the downvolt sweet spot.

 

[terroralpha]

Not every piece of silicon on the wafer has the same quality. RTG using higher default core voltage so they can utilize as much of the wafer as possible.

Plus, stability is relatively speaking. For my FuryX i can undervolt to 1220mv from 1243mv. Can it play games? Sure. Can it pass FireStrike Ultra stability test? Not even once.

As for you i would highly recommend you do some stability test to find the downvolt sweet spot.

i'm too cheap to pay for the full copy of 3dmark. but i did put it though 1 hour of furmark. everything looked good. it stayed under 76*C at 80% fan speed. the only thing i care about is gaming performance. if it doesn't crash or throttle while gaming, i'm more than happy with that.

i know that not all vegas can do what my card can, but all vega cards can run at ~1050mV with the clock speeds slightly toned down. the affects on performance would be very small but the power draw would be cut significantly.

 

[xkm1948]

I believe the electronic engineers at RTG knows a bit more about stability and core voltage than we end users do. They left the default core voltage at current level for a reason.

 

[londiste]

What monitor size are you using? 1440p?

yup, 1440p. that is still the sweet spot for this performance level. as can be seen from the graph, we can get 60fps with ultra/maxed settings, sometimes more.

With my custom water cooled Vega 64 at 1075mV and 2.5% frequency overclock ( no HBM overclock), the average clock in gaming always hovers around 1620MHz.

i'm wondering if watercooling would really matter that much. i had cooler working to definitely keep temp under 80c with oc-d testing. mostly successfully. that seems to be low enough for no thermal throttling. tried 4900rpm for a couple tests but that did not affect the results and i did not want to be in the same room.

i was working on clocks with furmark which was somewhat surprisingly constantly getting better clocks than actual games. from my testing, clock speed was directly affected by core voltage. every time i lowered it, the resulting clock was reduced. at stock it was 1440mhz-ish, after +50% pwr limit , it stayed at 1597mhz. after some undervolting, it ended up at 1570mhz (on average 1555-1560 in games) that i was happy with (especially considering that 88w more power was needed for that +27mhz).

I read over at Hardocp that Vega does not perform very well with traditional Multi-Sample or Super Sample Anti-Aliasing (MSAA, SSAA).

the couple tests i tried with 4xmsaa (dx:md and r6:siege) do show that vega gets a bigger hit from it. not hugely so though.

 

[turbogear]

yup, 1440p. that is still the sweet spot for this performance level. as can be seen from the graph, we can get 60fps with ultra/maxed settings, sometimes more.

i'm wondering if watercooling would really matter that much. i had cooler working to definitely keep temp under 80c with oc-d testing. mostly successfully. that seems to be low enough for no thermal throttling. tried 4900rpm for a couple tests but that did not affect the results and i did not want to be in the same room.

i was working on clocks with furmark which was somewhat surprisingly constantly getting better clocks than actual games. from my testing, clock speed was directly affected by core voltage. every time i lowered it, the resulting clock was reduced. at stock it was 1440mhz-ish, after +50% pwr limit , it stayed at 1597mhz. after some undervolting, it ended up at 1570mhz (on average 1555-1560 in games) that i was happy with (especially considering that 88w more power was needed for that +27mhz).

the couple tests i tried with 4xmsaa (dx:md and r6:siege) do show that vega gets a bigger hit from it. not hugely so though.

One can also have overclocking on Vega without liquid cooling, but the problem I had was that the fan noise above 3300rpm was very annoying for my ears.

I made the same observation as you. If you undervolt for example to 995mV and just increase the powelimit by 50%, the frequency also goes down.
Mine was stable at 995mV with less power draw than default setting but frequency was lower.
That is why I increased voltage to 1075mV and made 2.5% frequency OC to get best performace with only around 20W more power use as compared to Turbo mode.

Going beyond this OC point brought much higher power draw with very little performance increase.

 

[xkm1948]

Furmark is a bad way to stress test. All current AMD and Nvidia drivers have built in protection for furmark, once detected it will throttle the GPU down.

Plus, real world gaming have both intensive scenes and not so intensive ones, making your GPU going from full load to light load and back to full load. This is what a good stress testing should mimic. The altering of GPU load can quickly detect OC instability.

I would highly recommend FireStrike Ultra, TimeSpy stress test. Or if you dont wanna pay, 20 cycles of valley benchmark should also work.

Brief benchmarking undervolt success doesn't translate to long term stability.

 

[HD64G]

I can even suggest Witcher 2 as a great way to check stability. When I passed 10-15 minutes of simple gameplay without CTD incidents, I was sure of the settings being rock stable. Dragon Age Inquisition also is a good way of this test imho.

 

[londiste]

Furmark is a bad way to stress test. All current AMD and Nvidia drivers have built in protection for furmark, once detected it will throttle the GPU down.

Plus, real world gaming have both intensive scenes and not so intensive ones, making your GPU going from full load to light load and back to full load. This is what a good stress testing should mimic. The altering of GPU load can quickly detect OC instability.

I would highly recommend FireStrike Ultra, TimeSpy stress test. Or if you dont wanna pay, 20 cycles of valley benchmark should also work.

Brief benchmarking undervolt success doesn't translate to long term stability.

in my case, i wasn't really stress testing. i was trying to find a fairly optimal conf for the card. for that purpose, furmark was spot on. games as well as 3dmark stress tests had too variable a load.
plus, as i said, in case of vega, furmark was actually running on higher clocks than games.

for good stress testing, metro:ll should be good. in my experience it heats cards up as much as they can. from benchmarking, aots should also be awesome but that will quite properly load up cpu as well.