Soldato
Just a brief bit of info for you all I've only really had this morning to get to grips with it and the driver is a "beta" driver so whilst I wouldn't expect 1080ti gains in coming driver releases I would at least expect a more stable performance.
Stock testing and AMD's preset profiles
So here is vega running on the 3 preset profiles, which basically alludes to undervolted or overvolting the GPU via changing the power limit.
Power saving profile
http://www.3dmark.com/spy/2222494
Balanced Profile
http://www.3dmark.com/spy/2222470
Turbo profile
http://www.3dmark.com/spy/2222518
After burner read out
added a line to show the slight difference in temps between the 3 settings.
These results combined with looking at the behavior on afterburner you can see that there is some jitter on the core speed fluctuating up and down this type of behavior may have no impact on gaming performance but obviously, in synthetic benchmarks that apply 100% load for a set period of time, it does matter.
Overclocking
1. Core speed
You can slide it all the way up to 1980 but it'll throttle back the memory and won't perform properly
It would appear that the margin for improvement is low, The liquid cooled Vega is just a mere 47 MHz ahead of the standard black Vega 64 this suggests that even with better cooling the cards are pretty close to their limit. My testing so far has confirmed this.
Even an OC of 5% returned a degradation in benchmark scores (mainly due to the thermal limitations)
2. Memory speed
The most gains seem to be from memory boosting with 1100 being just out of reach at stock clocks I managed to get 1090 all the way through a bench yield a pretty good gain over the previously shared stock score.
A few runs at 1090 memory and an attempt to improve core speed
The low scores are where I've tried to under volt the cards which unsurprisingly whilst overclocking caused issues.
You can see from the comparison that the best score came from 1630 core and 1090 memory this was at + 50% power limit
The issue is, of course, about heat most of you are not going to want to run the fan at 4000 rpm like I was to do the testing.
Also, I would add that whilst it was in a TR system it pulled a whopping 535w peak from the socket in the wall during the most GPU intensive part of the time spy bench. compared to around 450w when in the turbo profile.
I ran VEGA at 1630 core 1090 mem on 0% power +25% power and +50% power limit the 50% power being the highest score.
You can see that at +50% power there is next to no jitter just a perfectly flat line on the core speed when the GPU is under load.
I would like to add that to run the card at 1090 I had to have the fan on max which I'm sure most of you will be unprepared to do.
Stock testing and AMD's preset profiles
So here is vega running on the 3 preset profiles, which basically alludes to undervolted or overvolting the GPU via changing the power limit.
Power saving profile
http://www.3dmark.com/spy/2222494
Balanced Profile
http://www.3dmark.com/spy/2222470
Turbo profile
http://www.3dmark.com/spy/2222518
After burner read out
added a line to show the slight difference in temps between the 3 settings.
These results combined with looking at the behavior on afterburner you can see that there is some jitter on the core speed fluctuating up and down this type of behavior may have no impact on gaming performance but obviously, in synthetic benchmarks that apply 100% load for a set period of time, it does matter.
Overclocking
1. Core speed
You can slide it all the way up to 1980 but it'll throttle back the memory and won't perform properly
It would appear that the margin for improvement is low, The liquid cooled Vega is just a mere 47 MHz ahead of the standard black Vega 64 this suggests that even with better cooling the cards are pretty close to their limit. My testing so far has confirmed this.
Even an OC of 5% returned a degradation in benchmark scores (mainly due to the thermal limitations)
2. Memory speed
The most gains seem to be from memory boosting with 1100 being just out of reach at stock clocks I managed to get 1090 all the way through a bench yield a pretty good gain over the previously shared stock score.
A few runs at 1090 memory and an attempt to improve core speed
The low scores are where I've tried to under volt the cards which unsurprisingly whilst overclocking caused issues.
You can see from the comparison that the best score came from 1630 core and 1090 memory this was at + 50% power limit
The issue is, of course, about heat most of you are not going to want to run the fan at 4000 rpm like I was to do the testing.
Also, I would add that whilst it was in a TR system it pulled a whopping 535w peak from the socket in the wall during the most GPU intensive part of the time spy bench. compared to around 450w when in the turbo profile.
I ran VEGA at 1630 core 1090 mem on 0% power +25% power and +50% power limit the 50% power being the highest score.
You can see that at +50% power there is next to no jitter just a perfectly flat line on the core speed when the GPU is under load.
I would like to add that to run the card at 1090 I had to have the fan on max which I'm sure most of you will be unprepared to do.
