The RX Vega 64 Owners Thread

gupsterg · 28 Aug 2017 at 01:23

What you see as 'memory voltage' in WattMan maybe VDDCI. That was what it was on Polaris, I know some OCN members that measured it via DMM. Polaris memory had dumb controller for VRAM with no data interface, so SW/VBIOS could not set voltage.

VDDCI on Fiji had a dumb control on PCB with no data interface, so SW/VBIOS couldn't change it's voltage. Fiji did have HBM voltage controlled by IR3567B so as it had data interface VBIOS mod allowed voltage change. No official or 3rd party app although allowed access.

As VEGA is seeming so close to Fiji, I reckon HBM voltage can be controlled via VBIOS/SW. Would AMD give access I don't think so. The numbers in Wattman just do not match known DMM readings and VBIOS info.

ubersonic · 28 Aug 2017 at 02:36

Figured I would put the card in one of my mining rigs today for a bit to see what it's performance was like, as expected it was poor (same power consumption as five RX560s and half the performance). But something interesting I noticed which I really wasn't expecting, is that when configured with a custom fan curve so it doesn't throttle under load, and placed in the same machine as a reference GTX1080 also configured with a custom fan curve to prevent throttling under load, the V64 is actually quieter. Obviously it's producing more heat and this can be felt at the back, but the better cooler means it's pumping it out quicker, and yet it's not as loud because while the whine of the GTX1080's fan drowns out it's airflow, the V64's airflow drowns out it's fan and is MUCH less annoying if you're in the same room as it.

TLDR: A Vega 64 reference sounds less annoying under load than a GTX1080 reference.

schneeb · 28 Aug 2017 at 05:43

Got a waterblock on my 64 now; HBM certainly seems to be quicker from just being cooler (someone on reddit was talking about some form of thermal throttling), it also still seems to get hotter than the core most of the time - which would make sense if it's at 1.35V!

LtMatt · 28 Aug 2017 at 07:19

TonyTurbo78 said:
Ok makes sense. One thing im not getting though. If HWinfo is displaying things vice versa, that the 1.356v that's being displayed in HWinfo is actually the memory voltage. My question as above is, how is there that much voltage going to memort when in Wattman it's only set to 1050mv?

Not sure on that one, I'll try to find out.

TonyTurbo78 · 28 Aug 2017 at 08:59

AMDMatt said:
Not sure on that one, I'll try to find out.

ok cool. Just trying to understand the voltage situation here as best as can be with the buggy software reporting. And regarding the default memory voltage, why is it set to 1050mv on my card where as the AIO versions and the 56 seems to default at 950mv? Shall I drop mine down from 1050mv to 950mv?

schneeb · 28 Aug 2017 at 09:20

Curious about HBM voltages too as that would explain the crazy heat it puts out; I got the GPU-Z beta and upping the HBM to 1100 definitely steals a chunk of power from the core clock.

Nemesis.ie · 28 Aug 2017 at 09:30

Are all those with the GPU-Z beta now seeing Samsung instead of Micron for the HBM2 maker?

LtMatt · 28 Aug 2017 at 09:34

TonyTurbo78 said:
ok cool. Just trying to understand the voltage situation here as best as can be with the buggy software reporting. And regarding the default memory voltage, why is it set to 1050mv on my card where as the AIO versions and the 56 seems to default at 950mv? Shall I drop mine down from 1050mv to 950mv?

I would leave HBM voltage at default settings and focus on core voltage for now.

schneeb · 28 Aug 2017 at 09:41

Nemesis.ie said:
Are all those with the GPU-Z beta now seeing Samsung instead of Micron for the HBM2 maker?

yep (no-one posted a 56 afaik though)

TonyTurbo78 · 28 Aug 2017 at 09:56

AMDMatt said:
I would leave HBM voltage at default settings and focus on core voltage for now.

ok will do.

And one last things, appreciate all your replies. If the HBM is plodding along at 1050mhz, then half way through the bench stress test it drops to 800mz and won't bounce back to 1050mhz, what causes that? It can't be throttling off temps because my core and memory temp max at full load are only 36c and 38c respectively. Is that a voltage thing, not enough juice?

Nemesis.ie · 28 Aug 2017 at 09:57

We might see some 56s later in the week if there is stock/proper prices today.

I don't see a "56 deals" thread from Gibbo yet though ...

Nemesis.ie · 28 Aug 2017 at 09:59

@TonyT - are you on GPU-Z beta? The latest version has a "Hot spot" function, it might be interesting to see what that is showing (we don't know where the sensor actually is yet) when you see the throttling.

LtMatt · 28 Aug 2017 at 10:01

TonyTurbo78 said:
ok will do.

And one last things, appreciate all your replies. If the HBM is plodding along at 1050mhz, then half way through the bench stress test it drops to 800mz and won't bounce back to 1050mhz, what causes that? It can't be throttling off temps because my core and memory temp max at full load are only 36c and 38c respectively. Is that a voltage thing, not enough juice?

Sounds like a bug i reported recently, wait for a driver update to fix it. I found adjusting HBM voltage was a sure fire way to trigger it.

Kaapstad · 28 Aug 2017 at 10:57

TonyTurbo78 said:
ok will do.

And one last things, appreciate all your replies. If the HBM is plodding along at 1050mhz, then half way through the bench stress test it drops to 800mz and won't bounce back to 1050mhz, what causes that? It can't be throttling off temps because my core and memory temp max at full load are only 36c and 38c respectively. Is that a voltage thing, not enough juice?

Does it do the same if you overclock it slightly less?

TonyTurbo78 · 28 Aug 2017 at 11:08

Nemesis.ie said:
@TonyT - are you on GPU-Z beta? The latest version has a "Hot spot" function, it might be interesting to see what that is showing (we don't know where the sensor actually is yet) when you see the throttling.

I'm on 2.2.0, don't see any hotspot function so I guess there might be a newer version.

AMDMatt said:
Sounds like a bug i reported recently, wait for a driver update to fix it. I found adjusting HBM voltage was a sure fire way to trigger it.

Ah ok, will just see how it goes for now then. Might just drop it to 1000mhz and wait for next driver. The good news is, I've benched the core at 1750mhz @ 1100mv and only hit 39c so I know the core is good for it. Memory will go to 1100mhz when the core clock isn't so aggressively overclocked.

Kaapstad said:
Does it do the same if you overclock it slightly less?

Kinda, it does and it doesn't, just trying to get my head around it really. I can bench for 30 minutes with the HBM at 1100mhz (when the core clock isn't overclocked through the roof) and it doesn't happen. But If I raise the core to 1700mhz and above, that's when the HBM drops to 800mhz at times and just will not revert back to it's speed, same if I drop the memory to 1050mhz with an aggressively high core overclock. To me, it feels like I either overclock one or the other to the limit, it doesn't seem to like it when I try being aggressive with both core and memory.

schneeb · 28 Aug 2017 at 11:23

My air freaks out (usually system crash) if you try to put over 1700 in state 7 too; did a quick 3dmark at 1697/1100 and its approaching 26k graphics score: https://www.3dmark.com/3dm/21826659

gupsterg · 28 Aug 2017 at 11:26

schneeb said:
(someone on reddit was talking about some form of thermal throttling)

HBM1 Jedec info in this post, IMO HBM2 will also have similar implementation.

schneeb · 28 Aug 2017 at 11:45

gupsterg said:
HBM1 Jedec info in this post, IMO HBM2 will also have similar implementation.

interesting, curious what this hot spot turns out to be; seems like it's VRM related with the crazy temps you can get with air coolers!

gtl_racer · 28 Aug 2017 at 11:50

Carmageddon Max Damage crashes on race start(runs ok with an old HD 7870). Can anyone confirm this? Thanks..

Acesbomb · 28 Aug 2017 at 12:15

Looks like it is possible to flash the AIO bios to a vega air 64. All is forgiven AMD.