AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

How do you install new drivers? Do you use DDU to uninstall old driver first?
I've tried using the "clean install" option in the installer integrated into adrenalin itself, now that it has become available. This time I didn't feel like having my PC reboot an extra time so I didn't, it hasn't made much difference in the past; traditionally I've just mangled the new driver on top of the old one and it has worked out fine. :p

I feel a driver should always be able to adequately manage the task of installing itself properly; that's the least anyone can ask of it. Nvidia drivers seem to do a good job of cleaning up the old driver first before installing the new. AMD might want to take note, but then again, their driver uninstalls have historically been really shitty (basically leaving all of it in place...)

I'm going to try out DDU even though I don't like or trust such programs. Plus they beg for donations, which I feel is annoying. I shouldn't have to use a third-party driver uninstaller in the first place. Driver should handle being installed; simple as that!

Can you check task manager for processes using the GPU?
Task manager says 0% GPU used, but I've found this indicator to be unreliable. It never seems to register anything at all. GPU-Z also says zero GPU load - probably because PC is idle. ;) Still, clocks and volts are maxed, with power saving profile selected in Wattman. Idle power draw is huge, more than 5x normal level.
 
Task manager says 0% GPU used, but I've found this indicator to be unreliable. It never seems to register anything at all. GPU-Z also says zero GPU load - probably because PC is idle. ;) Still, clocks and volts are maxed, with power saving profile selected in Wattman. Idle power draw is huge, more than 5x normal level.

Yeah it shows 0 % for me when my Vegas are mining, increasing to 10-20% if i'm watching something like youtube
 
When mining, task manager only shows activity in Compute 2 and 3 I think.
What is this Compute 2 and 3 devilry you speak of? :p I see nothing like that in my task manager...

im having the exact same issue, it happens after selecting a wattman profile, takes idle from 6 watts to 60 watts.
I ran DDU and then reinstalled latest driver. Card now clocks down properly - at the moment at least; I haven't tried setting a wattman profile yet.
 
What is this Compute 2 and 3 devilry you speak of? :p I see nothing like that in my task manager...

Exactly. We don't care about any of this. All that's obvious is that the GPU is (heavily) used and task manager fails to display this
 
What I don't understand at primitive shaders. They said they have similar primitive shader Technics integrated in Wolfenstein, so why they can not activate it?

And does it mean that primitive shaders works over the normal shaders?
 
What I don't understand at primitive shaders. They said they have similar primitive shader Technics integrated in Wolfenstein, so why they can not activate it?

And does it mean that primitive shaders works over the normal shaders?
Primitive Shaders replace "normal shaders" (read: vertex & geometry shaders, not all). Wolfenstein doesn't use primitive shaders, but it does use GPU culling which is "similar in principle" to what primitive shaders enable you to do (not that it would be all primitive shaders are good for)
Also they still need to bring out the primitive shaders API in the public to enable devs to support them if they wish, as the plan of drivermagic fell through
 
Vega 20 has been "known" (in leaked slides) for quite a while.

It's a 7nm Vega with 64 CUs, 4*HBM2 channels and, most of all 1:2 FP64 rating. It's a chip for HPC and it'll probably come out at the same time as the first Navi cards.
AMD has been holding on to Hawaii for FP64 workloads for way too long now, and it's their only offering with 32GB RAM (which V100 will now achieve with 8-Hi stacks, and over 3x the FP64 throughput at similar TDP).

It's not for desktop gaming, but if AMD also duplicate the number of ROPs to 128 and/or managed to significantly increase the clocks, then they might release a Titan-esque version.
They could call it the Fury V or something, and sell it for ~$1500 or more (at 7nm it shouldn't have nearly the same production costs as a V100).
 
Vega20 has shown up in linux driver patch, six pci ids compared to nine of Vega10.

https://videocardz.com/75680/amd-vega-20-listed-in-linux-driver-patches

7nm or 12nm?
7nm
2Ghz reachable or not?
Maybe
And most importantly, will it be for desktop gaming or not.
No

It's 4096bit. Only thing new here are the pci id's, AMD told about everything else already at ces or wherever[/quote]

edit: 4096bit wasn't in the official CES notes I think, but it was leaked some other way earleir
 
Last edited:
I think it's too early to say there won't be a desktop gaming version. The already advertised HPC and Deep Learning components mean it's definitely not a gaming-centered product (like e.g. Polaris 10 or Vega M), but neither is nvidia's V100 and they're still selling the GPU in a pseudo-prosumer gaming form factor.


Regardless, here is one of the slides mentioned at CES regarding 7nm Vega:

2iwamYR.jpg


And this is a leaked slide from way back in late 2016 about the same GPU:

iVrN5m8.jpg


Notice how the high-speed xGMI interface and HPC capabilities were already mentioned back then.
 
I think it's too early to say there won't be a desktop gaming version. The already advertised HPC and Deep Learning components mean it's definitely not a gaming-centered product (like e.g. Polaris 10 or Vega M), but neither is nvidia's V100 and they're still selling the GPU in a pseudo-prosumer gaming form factor.
I'm pretty sure they said at CES quite straight that only "consumer Vegas" this year will be mobile ones
 
That CES presentation slipped my mind.
So 7nm confirmed, very likely to be >2Ghz unless AMD screw something up. But even then, it'd not be that ahead of 1080Ti in gaming and 4 stacks of HBM will cost a pretty penny too. Maybe they can disable a stack but keep the ROPs.
 
I'm pretty sure they said at CES quite straight that only "consumer Vegas" this year will be mobile ones

Prosumer != Consumer.

I'm not claiming it's likely to happen either. I'm only saying it's not impossible for AMD to release e.g. a lower binned part with only 3 HBM stacks and make use of a higher ROP count and clocks allowed by 7nm to release as a Prosumer card.
It would probably be just a tad better than high-end Navi in games (like the Titan V is to the Titan Xp), but the extra ML hardware would still be there.
 
I'm pretty sure they said at CES quite straight that only "consumer Vegas" this year will be mobile ones
If they push mGPU or whatever Navi's scalability means with multiple chips, is there a difference between mobile and consumer? Theoretically that design would only use mobile chips and scale out the design.
 
Back
Top