By now some of us have played with either the Radeon RX Vega 56 or 64; so far it's been exciting to see the driver developments and improvements in game performance (and in compute abilities). If you have either RX Vega, let's see some of those numbers, and if you know someone who does, refer them to our Red Team community!
I use Rivatuner Statistics Server (guru3d.com/files-details/rtss-rivatuner-statistics-server-download.html), combined with MSI's Afterburner (msi.com/page/afterburner). Installation is simple: Download and install the latest Rivatuner from the link above, then install MSI's Afterburner. I don't use MSI's built in installation option for Rivatuner as it tends to conflict with Windows 10 Creators edition. That's also the reason why I install Rivatuner first, followed by Afterburner.
I try to limit docked apps as much as possible, and I keep my Windows start-up apps to a bare minimum. So I leave "Start with Windows" off only to launch Rivatuner when I need to benchmark games. All the options are fine on default settings; if you need to change your OSD color, you can select it here as well.
Important note: I use a very minimal setup for Afterburner - I do not use it at all for Overclocking or adjusting any hardware characteristics on any AMD/Radeon GPU; Use WattMan for that, always.
Step 1: On the monitoring tab, select the options you want to display
Step 2: Select "Show in On Screen Display".
You can assign hotkey's to toggle/show/hide the OSD.
That's it - Now get to posting your numbers and let the tweaking begin!
So this was a victory for both the Empire (EA's Battlefront), and a buttery smooth performing RX Vega 64 (Sapphire RX Vega 64 AIO on a custom cooling loop).
Battlefront is set up at Ultra settings for everything and on the RX Vega in WattMan, I'm using the Balanced mode, pulling 403 Watts at the wall (whole PC - Oscar Mike )
Of course, some more tuning can be done to bring this down even more without affecting performance much, if at all. Using the following custom settings dropped an average of 20W at the wall:
I posted a Youtube video of this here: RX Vega 64 - Benchmark and tuning series: Star Wars Battlefront - YouTube Other videos will include live tuning options showing the change in wattage draw at the wall.
Right... there we go. Undervolted, overclocked - really temperature sensitive this Vega... So for now: 1675Mhz@1120mv and 1000Mhz memory.
Update: 1725MHz@1150mv / 1100MHz HBM clock
Unfortunately, I don't have the luxury of using WattMan as it just breaks the overclocks since it's an older version (you lucky RX owners you... ) but WattTool and Afterburner do the trick just fine.
I honestly think it can squeeze out a lot more but I am overclocking this card on air, and had to put headphones on during testing I am absolutely sure I can push this even further but I'm not comfortable doing so without water-cooling.
This is ON AIR... with relatively unstable/old drivers! Just for reference, for those who haven't followed my thread regarding performance, out-of-box score for the FE on this test is on average 24500. Do the math: 28% increase in performance... how is that not simply amazing?
I only have a 1080p screen for now, a very good freesync one but 1080p nonetheless so I'm not sure in-game stats are relevant since out-of-box it churns out 200fps in maxed-out Doom if uncapped. So, yes, just let me know what tests to take
Update: to test stability - BMW27 blender times: 1:46.61 (for reference, 1080Ti FTW3 - ~1:59)
Power draw is definitely a concern to many so I ran another test (see video link at the bottom) which shows a live watt meter at the wall while making changes to Radeon Settings and running a few benchmarks. I was able to log all the queries and create a couple of handy dandy spread sheets for an easier overview.
I used Tom Clancy's The Division by Ubisoft for this portion of the series as the benchmark is relatively short to others. Keep in mind, the focus here was on power draw, not necessarily performance.
I ran the benchmarks using the Balanced and Turbo presets, followed by a custom profile based on the Turbo preset. The Balanced mode preset had a power draw of 381 Watts averaged at the wall.
The second run will be on the Turbo preset, with an average power draw of 418 Watts. There are a few spikes in balanced mode, but nothing out of the ordinary – overall the modes seem to use power accordingly. The Turbo preset consumed about 40 Watts more at the wall, but with an 10% gain in performance (only 4 frames per second). In most cases I can’t justify such a small gain versus the extra power consumed, so let’s look at setting a custom profile.
The third and final run is a modified Turbo mode, with power settings set to +50%, core voltage on p-state 6 set to 1050, and p-state 7 set to 1100 mV.
This yielded overall better performance of (only) 7 frames per second – or 18% performance gain- depending on how you want to look at it, but the important thing to note is that the overall power consumption dropped by an average of 60 Watts at the wall! A slight performance gain coupled with a lower load IS a gain to me.
Of course there’s room for more tweaking if I wanted a higher performance gain, I could play around with the HMB2 frequency and to some extend I should technically also see a slight bump in performance from adjusting the core frequency. I’ll leave specific game tuning options for a different article.
While tabulating the numbers, I compared the balanced preset against the custom mode and noticed not only a better performance gain, but also a marginally lower load at the wall, interesting!
Here's the video to see it in action live: RX Vega 64 - Benchmark and tuning series: Tom Clancy's The Division - YouTube
I don't have a socket watt meter, but my PSU (Corsair HX1200i) does give all sorts of metrics including power usage so I can definitely plot these out for Vega FE to complete the spectrum. I apologise if it looks like I'm butting in, this is an RX thread, but really the RX is a derivate of the FE so I hope someone will find the metrics useful in some way. Also, I can only test Stock vs OC essentially, the WattMan version I have is older, not as reliable and less features.
I'm just curious to see if there's a difference, the Liquid Cooled FE seems to be the most power-hungry out of the box. Bet I can top it! =)
I believe the liquid cooled versions of the cards come with a different UEFI installed that allows them to push to even higher frequencies/power limits. The air cooled versions are basically capped off, but some users on overclock.net have had success flashing air cooled cards with the liquid cooled UEFI.
Probably true, but forced BIOS flashing is a matter for the mad dudes at oc.net, we're looking to push the individual models as AMD ships them. Besides, Vega responds very well to low temperatures, so it would clock itself higher (as you can see on jayz2cents videos on youtube) and therefore draw even more power from the depths of hell... Hmm, maybe that's why DOOM runs so well on these cards =)
So it would be safe to say that if cooled enough, it's almost guaranteed to have higher stable clocks and therefore draw more power as shipped. No power-capping involved
True only if thermal throttling is what is limiting your clockspeeds. The air cooled versions of Vega do not allow a setting higher than 1200mv in a powerstate, while the liquid cooled versions do. Liquid cooling an air-cooled Vega does allow it to run at it's max clock speeds far more stably and consistently, but won't allow for access to the full voltage range enjoyed by those that purchased the AIO liquid cooled model. For that, a UEFI swap will be necessary .
... I don't think we're talking about the same thing here, I have the aircooled FE and just set 1250mv @ 1727Mhz and worked just fine just to test your 1200mv limit statement, I didn't take any screenshots because it heats up like crazy and throttles very quickly. No way I'm putting that voltage on air for more than 2 minutes, barely get by with 1150 for higher clocks without cooking marshmallows 2 feet behind the exhaust The real limitation (besides thermal) on Vega FE is the power limit, not voltage limit.