RX Vega Owners Thread

The clocks are just weird. It never really holds a single clock, moving around the whole time ie: 1742,1743,1756,1748... Not quite what I am used to, set the clock and that is where it ran.
 
Even without the OC, my clocks vary widely. Usually around 1300 but +-50 or so sporadically.

I also noticed some strange behavior in GPU usage, at least in GTA V. With GPU at 100%, I could get around 60 fps. But GPU will not stay at 100%, and can dip down to around 70% for 30 seconds or so at a time, and then jump to anywhere from 70 - 100% for a short period again. At 70%, I would only get 40 fps.

The odd part is that I don't think it's a CPU limitation, as on this same system, same game, with RX 480 CF both GPUs would mostly stay at 100% and I could get close to 70 fps solid. So I don't understand the low usage.
 
JustReason said:
The clocks are just weird. It never really holds a single clock, moving around the whole time ie: 1742,1743,1756,1748... Not quite what I am used to, set the clock and that is where it ran.
Click to expand...


That's the way the clocks work with Vega. It tends to jump all over the place as that's how it's designed. Something like Nvidia and their clocks where it will bounce as well.

As AMDmatt from another forum stated "Clock fluctutation is a feature of ACG = Advanced Clock Generator, which is enabled for DPM states 5/6/7. It will naturally fluctuate when there are current spikes. When current spikes occur voltage droops, to avoid instability during the droop, frequency must be decreased to match. This is why you see the fluctuation in the frequency."
 
cageymaru said:
Interesting read about how adding a water block to a Vega card can boost performance. Of course this is coming from a water block manufacturer, but looking at the numbers above in this thread it might be plausible. ;)
https://www.ekwb.com/blog/can-water-block-really-boost-gpu-performance/

Performance-PCS has those EKWB in stock.
http://www.performance-pcs.com/catalogsearch/result/?q=radeon+vega

I wonder if the 10% off on Labor Day works on them? Spend over $50 get 10% off: "LABOR17-10"
http://mailchi.mp/performance-pcs/happy-labor-day-from-ppcs-coupons-new-products-2440317
Click to expand...

Code works, I have an itchy finger!
 
Ok so after getting up this morning decided to throw a couple more runs of Firestrike extreme at my Vega and got lower scores. First I noticed the first test would not go into state7 clocks. Finally figured out you MUST set voltage to Manual whether you change anything or not, otherwise it stays very conservative. Ok so after that I noticed the clocks never reached the same levels as last night with stock volts, regardless where you have your max clock set. LOWER VOLTAGES GETS HIGHER CLOCKS. Have mine set for 1777Mhz. At stock volts (1200mV) the max clock I would see was about 1742. @1180mV max clock was 1768. So far that works out quite well. Then I ran Kombuster KMark and reached 1804, max still set at 1777. Still got a lot of learnin to go apparently.
 
JustReason said:
Ok so after getting up this morning decided to throw a couple more runs of Firestrike extreme at my Vega and got lower scores. First I noticed the first test would not go into state7 clocks. Finally figured out you MUST set voltage to Manual whether you change anything or not, otherwise it stays very conservative. Ok so after that I noticed the clocks never reached the same levels as last night with stock volts, regardless where you have your max clock set. LOWER VOLTAGES GETS HIGHER CLOCKS. Have mine set for 1777Mhz. At stock volts (1200mV) the max clock I would see was about 1742. @1180mV max clock was 1768. So far that works out quite well. Then I ran Kombuster KMark and reached 1804, max still set at 1777. Still got a lot of learnin to go apparently.
Click to expand...

That goes for many owners. It may be lesser of a performance card than a 1080Ti, but I'm having fun tweaking and finding the sweet spot for my card. I guess you could say Vega is great for tweakers unlike the 1080 is already maxed out from the start and not much tweaking can be done.
 
Definitely seems like everyone is hitting substantially higher clocks with undervolting. Seems like during Vega reviews clocks we're not sticking. I'm interested in seeing benchmarks vs 1080/ti when over clocked. Seems like people are picking up 20%.
 
upload_2017-9-3_0-4-27.png

so this is my first try tonight and I decided to leave the core clock stock and ONLY change the voltage (all this after seeing some reviews of flashing the air cards with LC/WC bios'). Clock never was as high during the tests as previously yet managed a better benchmark run... go figure. Gonna try and see how low but black screening sucks with my PC being in my air-conditioned cabinet ( too muchwork getting to the reset button, lol).

EDIT: Wait I see what happened, maybe. I accidently set the memory to 1100 not 1000 like I had before.
 
Last edited:
JustReason said:
View attachment 35293
so this is my first try tonight and I decided to leave the core clock stock and ONLY change the voltage (all this after seeing some reviews of flashing the air cards with LC/WC bios'). Clock never was as high during the tests as previously yet managed a better benchmark run... go figure. Gonna try and see how low but black screening sucks with my PC being in my air-conditioned cabinet ( too muchwork getting to the reset button, lol).

EDIT: Wait I see what happened, maybe. I accidently set the memory to 1100 not 1000 like I had before.
Click to expand...

You are getting closer to my score at least!
 
JustReason said:
View attachment 35293
so this is my first try tonight and I decided to leave the core clock stock and ONLY change the voltage (all this after seeing some reviews of flashing the air cards with LC/WC bios'). Clock never was as high during the tests as previously yet managed a better benchmark run... go figure. Gonna try and see how low but black screening sucks with my PC being in my air-conditioned cabinet ( too muchwork getting to the reset button, lol).

EDIT: Wait I see what happened, maybe. I accidently set the memory to 1100 not 1000 like I had before.
Click to expand...
its nearly DOUBLE my firestrike score lol..........Not bad for a Vega:)
 
So I've been messing around with undervolting a little bit to see what is going on

Here is default turbo:
vegasuperpositionturbo.png


And then we have custom with 1050Mhz memory, increased fan speed (3400rpm) lower temp max and target (80 and 72) and plus 15%

superpositiondefaultvoltage.png




Lastly I decided to keep the above settings, but I bumped the voltage to 1050 for state 6 and state 7 from 1020
superposition10.6k.png




Quite the drastic improvement from bumping down voltage. A 9% improvement in performance vs turbo just from lowering voltage.I don't have a kill-a-watt to measure power differences, but ill consider ordering one if you guys are interested.
 
wx9100.jpg


Not gaming, and it's not even going in a very powerful box. Intel i7-x990 with 24GB of RAM. Used for professional CAE software development and validation, running Linux.

But hey, it's fun having the top of the line Vega card to play around with. Here's hoping the Corsair 850W PSU can handle it.
 
jjmcwill2003 said:
View attachment 35710

Not gaming, and it's not even going in a very powerful box. Intel i7-x990 with 24GB of RAM. Used for professional CAE software development and validation, running Linux.

But hey, it's fun having the top of the line Vega card to play around with. Here's hoping the Corsair 850W PSU can handle it.
Click to expand...

whats the price on that?
 
jjmcwill2003 said:
View attachment 35710

Not gaming, and it's not even going in a very powerful box. Intel i7-x990 with 24GB of RAM. Used for professional CAE software development and validation, running Linux.

But hey, it's fun having the top of the line Vega card to play around with. Here's hoping the Corsair 850W PSU can handle it.
Click to expand...
you need that in a threadripper setup! For Real!:D Nice card Anyway. (these exotic cards are interesting cause really know one has them)
 
primetime said:
you need that in a threadripper setup! For Real!:D Nice card Anyway. (these exotic cards are interesting cause really know one has them)
Click to expand...

I'm trying to get one, because compiling on that i7-x990 is takes WAY too long.
I have access to a Lenovo Thinkstation P900 with dual Xeon E5-2695v3's(28 cores, 56 threads), but it's only certified with NVidia cards, so the wx9100 can't go in it, so where's the fun in that? :)
 
as far as power wattage is concerned... With 50% slider for power I still only pull 540w from the wall, total system. My 1800x can pull 240w on it's own and likely during my benches is only pulling 90-120w, then all other components are prob up to 35w so my Vega64@1770/1180mV - 1100/( volts unchanged and unknown) is pulling in the vicinity of about 400W max, likely closer to 350w actual considering psu efficiency and error. Most meters shown in HWiNFO seem accurate and put my Vega between 300-350w during benches, obviously lower during frame limited gaming.
 
sabrewolf732 said:
So I've been messing around with undervolting a little bit to see what is going on





Quite the drastic improvement from bumping down voltage. A 9% improvement in performance vs turbo just from lowering voltage.I don't have a kill-a-watt to measure power differences, but ill consider ordering one if you guys are interested.
Click to expand...


Then why is AMD pumping so much voltage in these cards? Why can't they bin and apply the necessary voltage instead of just going balls to the wall voltage on all cards??
 
Rvenger said:
Then why is AMD pumping so much voltage in these cards? Why can't they bin and apply the necessary voltage instead of just going balls to the wall voltage on all cards??
Click to expand...
did the same with the APUs. I am curious about this as well.
 
Rvenger said:
Then why is AMD pumping so much voltage in these cards? Why can't they bin and apply the necessary voltage instead of just going balls to the wall voltage on all cards??
Click to expand...

I'm thinking that yields are terrible so they're just using the voltage necessary to hit desired clocks on all cards, IDK. If amd reduced voltage it seems like they'd benefit on two fronts, one being performance second being power consumption, hard to say being that im not an rtg engineer.
 
Got mine up and running, had to tear apart the Nano rig, set it on top of the workstation vice integrated into it. This is with a 600w Corsair Power Supply which so far is doing great:
The highest wattage from the wall as read on a Kilowatt meter was 457w when doing the benchmark in Dirt Rally at 4K Ultra preset and it average was in the high 70's. Other benchmarks it was less than that.

RxVega64LC.jpg

It has to be the most beautiful looking GPU I've ever own or seen, the packaging was huge, a lot of foam rubber - this thing could be drop out of an airplane and it would probably survive. :ROFLMAO:

Packaging.jpg
​
 
sabrewolf732 said:
ontariotl and JustReason what clocks were you guys at to break into 11k in superposition?
Click to expand...
In Balance power mode, I got 10670 something 447w from the wall

Power Save 99xx something 385w from the wall.

Balanced mode clock speed maintain around 1600mhz, Power Save it was around a consistent 1500mhz.

Undervolted GPU from 1200mv to 1150mv and it crashed, had HBM2 up to 1000mhz.
 
sabrewolf732 said:
ontariotl and JustReason what clocks were you guys at to break into 11k in superposition?
Click to expand...

noko said:
In Balance power mode, I got 10670 something 447w from the wall

Power Save 99xx something 385w from the wall.

Balanced mode clock speed maintain around 1600mhz, Power Save it was around a consistent 1500mhz.

Undervolted GPU from 1200mv to 1150mv and it crashed, had HBM2 up to 1000mhz.
Click to expand...

He has his blocked so his temps are incredible. But anyway my clocks for that are 1772core @ 1170mV with memory 1100@ (says 950mV but that cant be right so I haven't touched it, think it is around 1.356mV if HWiNFO64 is accurate). Another thing DO NOT LEAVE VOLTAGE SLIDERS ON AUTO!!! It will kill benchmark runs. Even if you don't touch voltage, although knocking of 10-20mV is a good idea, setting it to manual will make sure you hit STATE7, which for me wouldn't in test one of firestrike.
 
sabrewolf732 said:
ontariotl and JustReason what clocks were you guys at to break into 11k in superposition?
Click to expand...

I'm using the high power AIO bios and I have mine set to 1732 as the highest P state. I can't go any higher. And I have the Ram set to 1105.

JustReason - Thanks for the voltage tip. I didn't think about turning off auto. I'll see if it makes any difference.
 
You must log in or register to reply here.
Back
Top