Vega owners club

[springs113]

I haven't fully done anything with the cards as one is out for delivery, and the other right now is for my TR build which has been all over the place since I had to have 2 replacement boards sent out. Bent pins and non functional dimm slots. I did briefly game on it this morning however and I must say(not to exaggerate) but OMG I forgot how smooth of game play single card delivers. On top of that Vega delivers for every game that I have tried so far(MGS5, Tekken 7 and Doom) the game play is as smooth as butter. TR games just fine, no different than my current rig. Vega runs quite cool considering where I'm coming from(Hawaii cards) and although a tad loud it is expected. Both instances doesn't really matter to me anyways as I plan on going under water, just waiting on EK to send me my blocks. Aside from the motherboard issues, I would have done a lot more with this platform except now I'm case hunting again and I've got this weird issue with windows where if I turn off the login user screen it creates like a double user for some reason. Having 2 sign on but I'm the only user. SMH @ microsoft...it may be a user error but I have since temporarily fixed the dual log on but cant get rid of the required sign on.

 

[TheoneandonlyMrK]

I haven't fully done anything with the cards as one is out for delivery, and the other right now is for my TR build which has been all over the place since I had to have 2 replacement boards sent out. Bent pins and non functional dimm slots. I did briefly game on it this morning however and I must say(not to exaggerate) but OMG I forgot how smooth of game play single card delivers. On top of that Vega delivers for every game that I have tried so far(MGS5, Tekken 7 and Doom) the game play is as smooth as butter. TR games just fine, no different than my current rig. Vega runs quite cool considering where I'm coming from(Hawaii cards) and although a tad loud it is expected. Both instances doesn't really matter to me anyways as I plan on going under water, just waiting on EK to send me my blocks. Aside from the motherboard issues, I would have done a lot more with this platform except now I'm case hunting again and I've got this weird issue with windows where if I turn off the login user screen it creates like a double user for some reason. Having 2 sign on but I'm the only user. SMH @ microsoft...it may be a user error but I have since temporarily fixed the dual log on but cant get rid of the required sign on.

Are you going to be crossfiring two vegas , that would be interesting to see how well it works.

 

[springs113]

Are you going to be crossfiring two vegas , that would be interesting to see how well it works.

That was my initial plan, but honestly after seeing the power draw, i didn't think my psu would handle it (seasonic 1000w titanium). I even thought of returning the second card but after seeing the prices (no way). I'll surely test it out but gaming vega64 on a 1440p non freesync monitor is actually smooth, i can only wonder what those that game on a freesync panel are experiencing.

 

[TheoneandonlyMrK]

That was my initial plan, but honestly after seeing the power draw, i didn't think my psu would handle it (seasonic 1000w titanium). I even thought of returning the second card but after seeing the prices (no way). I'll surely test it out but gaming vega64 on a 1440p non freesync monitor is actually smooth, i can only wonder what those that game on a freesync panel are experiencing.

It's very good tbh .i have a freesync 1 samy.
With enhanced sync turning back up in that last driver im really loving single gpu gaming atm and i can play all games at 4k smoothly , even dx12 etc.
I'm still sacrificing the odd setting in some but 4k and crazy anti aliasing dont make sense to me anyway, pubg got quite a boost and its my main game atm , looks good in 4k too

 

[acperience7]

Just got my card yesterday (Sapphire (regular version). I wanted the Limited edition, but they never returned to stock. Would have been cool since they were the same price just after launch. Hoping to do some undervolting this weekend. It will be my first time attempting such a thing! I'll have some benches up too if all goes well.

 

[TheoneandonlyMrK]

Just got my card yesterday (Sapphire (regular version). I wanted the Limited edition, but they never returned to stock. Would have been cool since they were the same price just after launch. Hoping to do some undervolting this weekend. It will be my first time attempting such a thing! I'll have some benches up too if all goes well.

Welcome to the club it went well for me but im not sold on it completely since i dont have a clamp meter and to me im not entirely sure the setting actually changes this is an undervolted run ill show a balanced one soon as

I haveve ran hwinfo64, so the wattage of gpu core and memory can be monitored , might be unreliable but hwinfo is normally not bad as far as it goes. 0.975V on the core no power slider , i noticed max clock does drop a bit

the score disparity is explained by a slight memory overclock on the undervolted one but odd readings abound like low gpu watts etc.

i also ran three benches on the recently released drivers but the fact settings dont always save or change makes all this a bit meh.

ie it looks like it will run on thin air but i doubt it

 

[turbogear]

So I did undervolting tests on my water cooled Vega 64 setup.
In order to measure the power consumption, I have bought Brennenstuhl PM231E that measures the total power consumed by the computer at the wall.

This time the testing is done with the new AMD drivers 17.8.2.

I did not just undervolt the Vega, I also overclocked it slightly.
I decrease the Voltage Control parameter in steps and increased the Frequency (%) parameter until a stable condition was reached.
I was able to reduce the voltage for p6 and p7 states to 1075mV and increase the value of Frequency (%) parameter in Wattman to 2.5.

Note: Originally I wanted to set the p6 and p7 frequency manually but it seems this is still not working in Wattman on 17.8.2 drivers.
I then left the flip switch in Watmann on Frequency (%) in percentage and increased the value to 2.5 and this worked.

The peak power consumption that I measured at the wall (this is total PC power) is as follows:
1. Idle mode peak power consumption of the computer 125W.
2. Default mode with Wattman on Balanced profile, the peak power reading was around 430W while playing Witcher.
3. Peak value of power consumption with undervolting plus overclock was at 450W while playing Witcher.


Now the interesting part. The average GPU clock frequency in Witcher game was measured as follows:
1. Wattman at Balanced power mode
- Average GPU clock frequency achieved is 1524MHz
- Average GPU temperature is 40°C

2. Undervolting (1075mV) plus a slight frequency overclock (2.5%)
- Average GPU clock frequency achieved is 1620MHz
- Average GPU temperature is still 40°C

With around 20W of more peak power, I got almost 100MHz higher clock in Witcher game by undervolting the Vega.

I played Witcher with these setting for almost one hour without any crashing.
I also tested Battlefield 1 with the same settings, the average clock there was mostly hovering even higher than in Witcher and was at around 1638MHz.

The pictures below shows the Wattman setting and average clock measured while playing Witcher Wild Hunt.

Edited: I added the Time Spy scores for the overclocked Vega.

 

[xkm1948]

Please use FireStrike Ultra Stress Test to check fir stability. It has to be over 98% or higher for long term stability. One thing I learned from OC Fiji was the ability to undervolt and overclock would gradually decrease. Not by means of performance, but by means of stability.

 

[turbogear]

Please use FireStrike Ultra Stress Test to check fir stability. It has to be over 98% or higher for long term stability. One thing I learned from OC Fiji was the ability to undervolt and overclock would gradually decrease. Not by means of performance, but by means of stability.

Thanks for the hint. I will run Firestrike today in the night.
During my testing yesterday, I ran Unigine Heaven benchmark.

Edited in the evening:
I did the Fire Strike Ultra stress test as you suggested, and I got a result of 99.4% using the setting that I mentioned in previous post (undervolting to 1075mV and 2.5% frequency overclock).

The image of the result are attached below.

 

[xkm1948]

Thanks for the hint. I will run Firestrike today in the night.
During my testing yesterday, I ran Unigine Heaven benchmark.

Edited in the evening:
I did the Fire Strike Ultra stress test as you suggested, and I got a result of 99.4% using the setting that I mentioned in previous post (undervolting to 1075mV and 2.5% frequency overclock).

The image of the result are attached below.

Nice! You get a good piece of silicone!

 

[turbogear]

Nice! You get a good piece of silicone!

Yes, I think there is still more potential available there for tuning.

I did not own Fire Strike before. I bought 3DMARK yesterday from Steam.
In the next days, I will try to fine tune my Vega 64 further.

 

[acperience7]

Nice results so far turbogear. So far, my card crashes with any less than stock voltage (even with stock clocks). Still experimenting like everyone else though. Right now my main goal is to get the clocks more stable with cooler temps.

I have noticed that certain benches start at higher clocks than others for me. Specifically: Firestrike and Time Spy start out at 1500-1600, but for some reason (Unigine) Heaven and Valley start out at about 1430-1530. Anyone else seeing this?

 

[xkm1948]

Nice results so far turbogear. So far, my card crashes with any less than stock voltage (even with stock clocks). Still experimenting like everyone else though. Right now my main goal is to get the clocks more stable with cooler temps.

I have noticed that certain benches start at higher clocks than others for me. Specifically: Firestrike and Time Spy start out at 1500-1600, but for some reason (Unigine) Heaven and Valley start out at about 1430-1530. Anyone else seeing this?

FirsStrike Ultra/Time Spy can saturate the GCN CU array better than aging DX11 Unigine Valley/Heaven

The FireStrike Ultra /Time Spy stress test is very good at detecting instability.

 

[turbogear]

Nice results so far turbogear. So far, my card crashes with any less than stock voltage (even with stock clocks). Still experimenting like everyone else though. Right now my main goal is to get the clocks more stable with cooler temps.

I have noticed that certain benches start at higher clocks than others for me. Specifically: Firestrike and Time Spy start out at 1500-1600, but for some reason (Unigine) Heaven and Valley start out at about 1430-1530. Anyone else seeing this?

What I notived during my experimenting is that the average clock varies from one application to other.

I am using the following settings in Wattmann. I have 1075mV for core voltage and 2.5% overclock corresponding to a setting of P6 and P7 states to 1672MHz instead of default at 1630MHz.
The worst case average clock that I observed on my GPU with this overclock is 1620MHz which occurs in applications that stress the GPU to the limits, like Fire Strike Ultra stress test, Witcher Wild Hunt and Time Spy benchmark.
For other applications where the GPU may not be stressed to limits for example Battlefield 1, the average clock was in the range of 1638MHz.

I think 1075mV with 1672MHz is the sweet spot of my GPU with only 20W more peak power, I am getting 100MHz performace over the default mode.
Going beyound this does not help much with the performance in applications that stress the GPU. The clock always averages to around 1620MHz in such cases.

Further fine tuning the GPU lead to 4.5% overclock (i.e. 1702MHz in P6 &P7 state ) with a core voltage of 1090mV, but the peak power consumptions jumped up by 40W over the 1075mV & 2.5% overclock case.
In this case the GPU average clock in games and applications that put the stress on the GPU was still in the range of 1620MHz but the peak power draw was higher by about 40W.

The Time Spy benchmark for 4.5% overlock is attached below with GPU score being 7475.
The GPU score increases only by 52 points over the 2.5% overclock case.

I think I will stick to 1075mV with 2.5% (1672MHz) overclock that gives GPU score of 7423 but 40W less power use.

 

[turbogear]

Sorry guys for the double post, but I could not edit the above older post anymore for some reason.

I have done some Fire strike test, as a continuation for the results for my overclocked custom liquid cooled Vega 64 that I posted before:
https://www.techpowerup.com/forums/threads/vega-owners-club.236314/page-2#post-3716160

These Fire Strike results are better in comparison to those that Tweaktown found from their reference liquid cooled Vega review:
https://www.tweaktown.com/articles/8304/amd-radeon-rx-vega-64-56-more/index5.html


Did anybody of you try to overclocking the HBM on Vega 64 so far?

 

[springs113]

I haven't attempted any oc just yet but man I've been pushing this card immensely with the default profiles provided by amd...it is a hot one lol not as hot as my 290Xs but close enough. I originally wasn't going to keep my 2nd one but being that i got it at msrp, I'm keeping it.
Back to the heat:
Aftermarket cooling is a must for those not planning to go under water like i am. I looped heaven/valley for about an hour a piece and although 80c want bad (95c 290x) the card started to throttle. I forgot the numbers exactly but the core dropped roughly 100mhz or so.

@turbogear
I have read that Vega thrives with an increase in memory speed but recall where i saw it(may have been GamersNexus). I think also the concensus was that Hbm was clocked low because the increase in speed on the memory scaled real nicely especially with little to no increase in power draw.

 

[qubit]

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.

 

[Norton]

Sorry guys for the double post, but I could not edit the above older post anymore for some reason.

No worries- Not really a double post since there's a day between posts

Has the crossfire issue been resolved with Vega cards yet?

 

[springs113]

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 can't stress enough game play is incredibly smooth and i don't have a freesync panel. I can understand why AMD did that crappy tour .

Edit1:
@Norton, I'm not sure yet but i will give this a go soon enough. I wanted to do it today but i may have some running around to do. I'm gonna go out on a limb and say for now (not fixed). Vega definitely needs some driver work. My benchmark scores on Gow4 is low, lower than it should be.

Edit2:
I do have coil whine, it's annoying at that. Going to test that out of i get the chance to as well..since i only noticed those while runny valley/heaven.

 

[turbogear]

I haven't attempted any oc just yet but man I've been pushing this card immensely with the default profiles provided by amd...it is a hot one lol not as hot as my 290Xs but close enough. I originally wasn't going to keep my 2nd one but being that i got it at msrp, I'm keeping it.
Back to the heat:
Aftermarket cooling is a must for those not planning to go under water like i am. I looped heaven/valley for about an hour a piece and although 80c want bad (95c 290x) the card started to throttle. I forgot the numbers exactly but the core dropped roughly 100mhz or so.

@turbogear
I have read that Vega thrives with an increase in memory speed but recall where i saw it(may have been GamersNexus). I think also the concensus was that Hbm was clocked low because the increase in speed on the memory scaled real nicely especially with little to no increase in power draw.

Yes I read that also at semiaccurate in Vega forum that Vega is bandwidth starved and people were talking about some bandwidth bug in current state of the drivers.
I will do some overclocking of HBM for benchmarking, but I am not sure if I will use that setting afterwards.
HBM2 is for me too new and I don't know how safe it is to keep it overclocked without doing any damage.

I agree that the original cooler is underwhelming. It is not really suitable to operate the Vega with full potential.
Either one has to live with high amount of noise by increasing the RPM to above 3300 or have the GPU throttle all the time with default fan profile.
With my custom water loop, the clocks are much more stable.

I think partner cards that should start rolling out over the coming months would do much better job with cooling than original AMD blower cooler.

 

[springs113]

@turbogear how are your temps? I saw jayz2cents vid and granted he's using a test bench, his gpu under water was hovering around 45-48c after looping Heaven benchmark. That change reminds me of the drop my 290Xs saw when i put them under water.
I cant really test under my regular environment because i technically don't have my case as i was doing a whole new TR build. My blocks are here somewhere PPC did a weird thing to my pre orders and now i have to wait some more. Air testing on the other hand is definitely possible but when ones mind was set on watercooling, everything else fails in comparison.

Partner cards may be OK but from what GamersNexus said Vega (56, 64, FE) all uses the same pcb and the components on them are of high quality. They also noted that parts on AIB partners may not be as good. The one preview i saw of Asus' Vega 56, the fellow was pretty disappointed, it barely performed better than the reference.

 

[turbogear]

@turbogear how are your temps? I saw jayz2cents vid and granted he's using a test bench, his gpu under water was hovering around 45-48c after looping Heaven benchmark. That change reminds me of the drop my 290Xs saw when i put them under water.
I cant really test under my regular environment because i technically don't have my case as i was doing a whole new TR build. My blocks are here somewhere PPC did a weird thing to my pre orders and now i have to wait some more. Air testing on the other hand is definitely possible but when ones mind was set on watercooling, everything else fails in comparison.

Partner cards may be OK but from what GamersNexus said Vega (56, 64, FE) all uses the same pcb and the components on them are of high quality. They also noted that parts on AIB partners may not be as good. The one preview i saw of Asus' Vega 56, the fellow was pretty disappointed, it barely performed better than the reference.

The average temperature that I get on my Vega is 40°C under stress tests like Fire Strike and gaming like Witcher.
The peak temperature goes to 43°C for very short times.
These are observations at around 26°C room temperature.

I have three Radiator in a CM Storm Trooper case. 120mm rad at back, 240mm on top and 240mm on bottom.
All three rads push the air out of the case.
120mm+240mm radiator come after the Vega and output of these two goes to CPU.
The output of CPU goes to the 240mm radiator at the bottom before going into pump+reservior.

I agree that the reference card from AMD comes with high quality components.
That is why I buy most of the time their reference but their cooling solution is simply not enough.
If one is going to water cool then it is a good choice, otherwise one has to live with throttling or higher fan noise.

 

[Norton]

This is the Vega clubhouse, please stay on topic.

 

[turbogear]

Raja Koduri is back from Holidays and is active again on twitter regarding the Vega.
Here is the link: https://twitter.com/GFXChipTweeter/status/902651173894742016

He mentioned things concerning the Vega power, but I did not totally understand his point concerning the perf/watt vs perf/$.

Can somebody explain that?
Is he here referring to the binning process of Vega, when he talks about perf/$ vs perf/watt and says that they gave users options?




Here are some extracts from original twitter posting regarding Vega power:

Raja Koduri‏ @GFXChipTweeter 15 Std.vor 15 Stunden

Seems like a lot of questions around power of Vega. The initial round of reviews didn't showcase the perf/watt dynamic range of Vega well.

New articles in the last few days showing interesting range of options. Vega10 has the largest perf/watt dynamic range than any GPU I recall

There is a dilemma of whether the default setup be optimized for perf/$ or perf/watt. Ideally both..but..

Sales data tends to favor perf/$, but some opinions favor perf/watt. We tried to give user options to default based on their preference​

 

[springs113]

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.