AMD Vega Hardware Reviews

HWinfo have a "chip" power, which measure the whole card. It's not 100% accurate, but more than just gpu power. I use this with my Fury X and now Vega, and it's pretty good (I have a watt-meter to, and it pretty much confirms the HWinfo ....info.
 
My bad, I should've mentioned that I'm using Hwinfo for on screen display with RTSS since afterburner wasn't much of use and I was using hwinfo for CPU and memory monitoring before.

The chip power is what I was referring to, it's 165W at stock.
 
My bad, I should've mentioned that I'm using Hwinfo for on screen display with RTSS since afterburner wasn't much of use and I was using hwinfo for CPU and memory monitoring before.

The chip power is what I was referring to, it's 165W at stock.

Here is picture of 5min run of Firestrike extreme on loop with overclocked Vega 64 LC. Both HWInfo and Afterburner in screen for comparision (for me afterburner load shows exactly same as HWInfo's Chip Power does. Clocks do sail a bit more with hwinfo active for some reason, also games for me stutter with it a bit. Overall "oc'ed" 64 LC does take quite a bit more power as your 54 does, but it would have quite a bit more go still, if there wouldn't be that artificial 70C power temp limit on watercooled version :( Unfortunately custom bioses are no go for Vega's atm.
 

Attachments

  • Vega64-LC-HWinfo+AB.jpg
    Vega64-LC-HWinfo+AB.jpg
    1 MB · Views: 41
Last edited:
HWinfo have a "chip" power, which measure the whole card. It's not 100% accurate, but more than just gpu power. I use this with my Fury X and now Vega, and it's pretty good (I have a watt-meter to, and it pretty much confirms the HWinfo ....info.

Still worth noting that if the GPU card does not have the supporting IC or using a controller that allows for broader power sense then even HWInfo unfortunately will not help, and the Vega (amongst other modern AMD GPUs) do not have the complimentary IC to measure power demand/loss in same way as Nvidia.
It seems strange at 1st glance because they have great engineers and could do this, but one can see where AMD priorities lie when compared to Nvidia circuitry and more notably components used and both manufacturers have pros/cons.
 
Yeah I know, but to my surprise, it still pretty accurate (It's not perfect, but it's not like 50w off or something). My hardware vs hwinfo, I have only 5-15 watts differences, depending on the situation.
 
Yes I saw that however it only mentions the 1080ti, whereas for AMD it simply states AMD, thus affecting all GCN cards.
I actually had to turn Async off because it caused the game to freeze with a 1080 and no difference using latest drivers.
 
Yeah I know, but to my surprise, it still pretty accurate (It's not perfect, but it's not like 50w off or something). My hardware vs hwinfo, I have only 5-15 watts differences, depending on the situation.
How are you measuring the hardware outside of the software?
Unfortunately only accurate way is with a scope and a pretty fine tick/window, or a rather expensive clamp but not necessarily ideal.
For Vega to hit anywhere near the 1600 frequency its additional power use is well over 15 watts.
As an example look at those that talk about the power demand of Polaris 480 and then how it compared in reality when isolated and measured with a scope; serious difference.
 
Here is picture of 5min run of Firestrike extreme on loop with overclocked Vega 64 LC. Both HWInfo and Afterburner in screen for comparision (for me afterburner load shows exactly same as HWInfo's Chip Power does. Clocks do sail a bit more with hwinfo active for some reason, also games for me stutter with it a bit. Overall "oc'ed" 64 LC does take quite a bit more power as your 54 does, but it would have quite a bit more go still, if there wouldn't be that artificial 70C power temp limit on watercooled version :( Unfortunately custom bioses are no go for Vega's atm.

Stock voltage curve is too high, I've 1.2V for 1590Mhz clock on P7 at stock, I don't it'll ever reach it in scenarios with significant gpu load.

I read that disabling vrm monitoring or such on the card helps with the stutter issue or you can decrease the polling rate. AMD's components are more sensitive to voltage monitoring or such, afterburner used to have a warning about it.
 
How are you measuring the hardware outside of the software?
Unfortunately only accurate way is with a scope and a pretty fine tick/window, or a rather expensive clamp but not necessarily ideal.
For Vega to hit anywhere near the 1600 frequency its additional power use is well over 15 watts.
As an example look at those that talk about the power demand of Polaris 480 and then how it compared in reality when isolated and measured with a scope; serious difference.


A wattmeter at the wall. Then the difference between my system at idle and my system with the gpu loaded (and not the cpu)... It's not perfect but it's not waayyyy off. With the same stuff I know approximately how my cpu consume in multiples load scenarios, so, it's basic math in reproductible tests...
 
A wattmeter at the wall. Then the difference between my system at idle and my system with the gpu loaded (and not the cpu)... It's not perfect but it's not waayyyy off. With the same stuff I know approximately how my cpu consume in multiples load scenarios, so, it's basic math in reproductible tests...
Wattmeter/Killawatt/etc unfortunately is not enough, especially with Active Power Factor/Crest correction and a PC PSU, PSU efficiency, along with isolating the CPU and to a lesser extent peripheral powered components.
Like I said Polaris 480 was around 15W-20W power loss/not measured and that was the 1st release before they upped the clock frequency, Vega will see its own loss/not measured climb once it hits probably above 1400-1425MHz and definitely when up the power target control.
At 1425MHz Vega 64 is pulling 290W, where as 480 is pulling substantially less and reviews I am thinking of tested the Sapphire 480 model that probably cools better than the reference Vega (the heat will also contribute to power loss).

Now to be clear I am not being critical of AMD in this context, like I said both manufacturers have pros/cons with their GPU card engineering scope but it is something to be wary about when considering measuring the total power demand of a GPU; only 3 review sites really do this well and that is PCPerspective/Toms Hardware/Hardware.fr

Edit:
The Vega56 around similar envelope was pulling around 275W.
 
Last edited:
Wattmeter/Killawatt/etc unfortunately is not enough, especially with Active Power Factor/Crest correction and a PC PSU, PSU efficiency, along with isolating the CPU and to a lesser extent peripheral powered components.
What about one of them current clamp things?

It can't capture what passes through the PCIe slot, but I believe I read from PCPer's testing that reference Vegas didn't pull much from that source. One could conceivably go with whatever figure they measured up as a dummy template, and then add that to the clamp measurement's figures.
 
What about one of them current clamp things?

It can't capture what passes through the PCIe slot, but I believe I read from PCPer's testing that reference Vegas didn't pull much from that source. One could conceivably go with whatever figure they measured up as a dummy template, and then add that to the clamp measurement's figures.
Yeah a good calibrated clamp-multimeter can work but you need one of the reasonable models (meaning not the cheapest but not the most expensive) if looking to monitor this situation reasonably accurately, it still unfortunately is more generic because you do not control window/time interval-ticks for monitoring that is applicable to scopes but it is good enough unless looking to drill down into power behaviour and characteristics.
You would also need a riser you can measure for the 16x slot from the motherboard for either test setup equipment scenarios.
I was surprised that for once PCPer did not provide the detail breakdown of the power distribution like they usually do (no conspiracy they probably did not have time or felt too much info to present), in balanced mode the Vega 64 is pretty comfortable at around 25W with 75W the limits (although in reality most rely more on the 12V rail instead so would be 66W but still comfortably within limits and still within limits if pushed but obviously higher current demand).
 

Initial review:
99thValue.png

Today:
99thfps.png

http://techreport.com/news/32779/take-a-sneak-peek-at-our-geforce-gtx-1070-ti-results

RX 56 is now faster than 64 initially was. The speedup is close to 20%. So there, magic drivers really are a thing after all. :)

Not even remotely magic enough to threaten the 1080 Ti, however. Well, not yet, anyway.
 
Although that chart is fps per dollar.
We really need to see the performance per game charts that omit price factor to see what is changing and by how much.
 
Back
Top