According to him, it still downclocks to 1440MHz, though unlocking power limit achieves a solid 1600MHz, he even pushed it to a stable 1700MHz.A user on Reddit has the liquid-cooled FE SKU. They say they will do testing and game benchmarking later today:
https://www.reddit.com/user/ba1b0a
The alternative appears to be dropping the voltage from 1200mV to 1100mV to achieve a similar effect without all the effort. Lower the power draw to avoid having to increase the power limit in the first place. Vega should drop 20-30W just by cutting the memory pool in half on top of some undervolting gains. AIO or better cooling possibly doing the same with less leakage.GN review place the 400W Vega in between 1080s but the amount of effort needed to archive that is not something that we can call "normal".
I got the impression that lowering the vcore has little effect on the average clocks, despite lower power consumption and heat output. Looks like the clocks aren't exclusively related to power and temperature.Lower the power draw to avoid having to increase the power limit in the first place.
GN review place the 400W Vega in between 1080s but the amount of effort needed to archive that is not something that we can call "normal".
They previously determined Vega takes very little from PCIe, about 25W.Isn't it more like a 475W Vega? GN was not including PCIe power.
They previously determined Vega takes very little from PCIe, about 25W.
http://www.3dmark.com/compare/spy/2068702/spy/2065605#
Vega FE liquid @1700MHz tested in TimeSpy, slower than a 1080 by ~8%.
Vega FE air in Prey, slower than a 1080:
Isn't it more like a 475W Vega? GN was not including PCIe power.
Just looking at the GPU score, your 1080Ti there is exactly 40% faster.That's not that bad of a score. My very best score with a 1080 Ti is 37% higher: http://www.3dmark.com/spy/1620649
That's not that bad of a score. My very best score with a 1080 Ti is 37% higher: http://www.3dmark.com/spy/1620649
They probably don't expect for it to have enough volume to bother making another PCB just for the liquid variant.It's a detail but I don't get why the liquid cooled version is so long... I get that you need space for the blower one, but the liquid one ? Maybe the RX version will have a Fury size with another pcb ? I like short cards...
The right part of the pcb is nearly empty...
They probably don't expect for it to have enough volume to bother making another PCB just for the liquid variant.
Isn't that what happened with Kepler to Maxwell? Look at Linux driver development and those kind of swings are nothing. A few swings like that in a month aren't always uncommon, but they quickly become situational. If the gains weren't there why even delay Vega's release a month?I don't think theres enough black magic in the world to make the car 40% faster via drivers not just in a couple of months but in its entire life span.
They previously determined Vega takes very little from PCIe, about 25W.
http://www.3dmark.com/compare/spy/2068702/spy/2065605#
Vega FE liquid @1700MHz tested in TimeSpy, slower than a 1080 by ~8%.
Vega FE air in Prey, slower than a 1080:
Perhaps, but against GTX 1080's at reference clocks. Custom variants might be a different story with overclocking/similar cooling.@1600MHz and above Vega RX will have a better performance than a GTX1080 for sure in the vast majority of the cases but whithout a driver overhaul will still be far cry from a GTX 1080 Ti....