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

The question is 3-5% or 30-50%. Performance is likely to increase, but is it enough to justify running all the benchmarks again?
The question is, what's the base for those comparisons. All those game-ready drivers (whether from Nvidia or AMD) are there for a reason …
 
The question is, what's the base for those comparisons. All those game-ready drivers (whether from Nvidia or AMD) are there for a reason …
The basis would be features designed to waste fewer resources being further enabled or improved. The worst optimized games likely seeing the largest benefit. A game in effect rendering everything backwards could waste a lot. The numbers presented in the whitepaper are nearly a year old at this point, involve mostly programmable techniques, and at launch I'd argue rather limited in implementation. Async compute for example was on consoles for years before devs really discovered viable optimization techniques. It stands to reason a feature such as primitive shaders, which AMD views as the pipeline of the future, isn't fully hashed out yet. Nvidia took years and a couple product generations to achieve a similar implementation.
 
I was talking about the base for comparisons wrt to driver improvements (that's why I wrote a second sentence in my post). If it's like you describe, hats off. If it's a comparison between a driver completely unaware of certain caming titles vs. drivers optimized for this title, then not so much.
 
Ordered two ASUS Strix Vega 64 OC 8GB. Why?

Because I'm insane.

Also, because it goes against my principles to buy a GPU which is a year and a half old already even if it is bloody fast (GF 1080). Also, I'm an ASUS guy, so this is how it will be. Delivery somewhere around december 13th to 18th. Is going to be a loooong week and a half until I have them... :???:
 
can't answer for him, but for gaming it should have very poor value. (for mining as well). If the rumors are true and the next Geforce will not be Volta - based, i'd be quite sad to be one step behind as well.
But if not, yeah paying more for early access could be somewhat justifiable

Might be attractive for some though if it's super good for folding @ home..
 
Software is always hard. Good thing is, you can fix it later. Why would anyone think, performance would not be going up? Just look at tried practices to make new driver releases appear as the next best thing to sliced bread. You just have to pick your cherries accordingly.

Depends what you are doing with the software. The overlay that is coming with Adrenalin is relatively straightforward I think. What AMD is trying to do with DBSR and primitive shaders I would rank more towards difficult. DBSR is a new way of doing things for them. Primitive shaders no one has done before. Take Nvidia's legendary DX11 performance driver. It took them 3 years to implement that scheme. AMD has to do something hard in somewhere between 1 and 2 years. They got their first Vega samples back in I think late October of 2016. Initially PS were touted as a developer exposed tool. They then changed their stance several months later to being "the driver will take care of it." That is a tall task on a tight timeline. As far as performance someone already touched on the difference I was trying to illuminate. Are we going to see a reasonable jump as drivers march onward or are we going to see something worthy of "poor volta?"

Now that I think more about this scenario, why the change in position on PS? AMD's whole angle with their offerings is on the open source side of the spectrum. They build it and give everyone full access and/or build the initial tools and release the source. The two possibilities I can think of is 1) Su took Raja to task about performance improving features that don't help sell cards now. This caused them to go the driver route. It's contrary to their typical way of doing things though. or 2) If something is actually broken in Vega there would be no way you could expose it to devs. They would have to build the work around in the driver.
 
Initially PS were touted as a developer exposed tool. They then changed their stance sever
I don't remember that.
Though no one would mind it being exposed anyway.
AMD's whole angle with their offerings is on the open source side of the spectrum. They build it and give everyone full access and/or build the initial tools and release the source.
A certain AT editor has already explained why its not exposed to devs.
 
Why not just buy a Titan V? :)
A: I don't have THAT much money... It costs roughly as much as I spent on my PC minus these new GPUs.

B: Even if I had ludicrous amounts of money, I hadn't actually even heard of the T V by the time I ordered the cards this morning. :p

Might be attractive for some though if it's super good for folding @ home..
It's most likely going to be a friggin' wrecking ball at folding. Current pascal cards are stomping anything AMD currently makes into the ground from what I understand and with volta's arch improvements and general upsizing of everything it is going to be simply fearsome I would imagine.
 
I was talking about the base for comparisons wrt to driver improvements (that's why I wrote a second sentence in my post). If it's like you describe, hats off. If it's a comparison between a driver completely unaware of certain caming titles vs. drivers optimized for this title, then not so much.
Definitely since launch, however it's a question of what has been optimized? DSBR and PS specifically could have a large impact at the flip of a switch and it's unclear if they exist even partially enabled. While I wouldn't expect much in optimal cases like Wolfenstein or any DX12/Vulkan title, I'd expect huge gains in titles where Nvidia appears disproportionately ahead. Where geometry rate and overdraw are problematic. While all should benefit, the cases where async can't limit the bottlenecks should be significant.

The compiler bug I alluded to involved surface format descriptors with seemingly random bit errors. The scope of which could prohibit certain early z and occlusion tests in the pipeline. The pipeline may not have realized certain stencils existed and acted accordingly. In the case of DSBR and overdraw that could remove a lot of work. The remaining question being how deep the bins go: binning within single, multiple, or many draws. It's difficult to say without a detailed understanding of how they are processing bins. It could be the difference between forward and deferred renderers. It's a rather large, complex system to get working.

Now that I think more about this scenario, why the change in position on PS? AMD's whole angle with their offerings is on the open source side of the spectrum. They build it and give everyone full access and/or build the initial tools and release the source. The two possibilities I can think of is 1) Su took Raja to task about performance improving features that don't help sell cards now. This caused them to go the driver route. It's contrary to their typical way of doing things though. or 2) If something is actually broken in Vega there would be no way you could expose it to devs. They would have to build the work around in the driver.
My guess is it's another Mantle scenario where all IHVs are standardizing the implementation, if that's possible. A graphics pipeline of the future per IHV isn't ideal from a dev standpoint, but it could work as the console devs would make use of it. Cries of gimped when those titles hit Nvidia PC users though...

1) More performance doesn't hurt.
2) Software broken or incomplete seems more likely in this case. As I alluded to above, it's possible they are standardizing the approach for future extensions which takes time. Not unlike the Mantle to Vulkan process.
 
Definitely since launch, however it's a question of what has been optimized?
So, Crimson Relive Edition launched at 16.12 I believe, and now the new one probably being 17.12. Good comparison. Bound to be much faster for many of this years titles. Does it warrant all the commotion?
 
So, Crimson Relive Edition launched at 16.12 I believe, and now the new one probably being 17.12. Good comparison. Bound to be much faster for many of this years titles. Does it warrant all the commotion?
Many of the titles should have already been optimized in the course of their release. What I'm saying is that DSBR and primitive shaders should provide a nice boost on top of those WHEN there is room for improvement. Skilled devs with highly efficient rendering pipelines just won't be leaving much performance on the table. The best examples would be unoptimizied masterpieces like PUBG. As well as DX11 titles that likely don't use async culling. Tiling in Doom with its 50+ render targets and async culling would already be close to what those features provide universally.

As for the commotion I'm not expecting much, but I could be wrong there. It's prime holiday shopping season and Vegas are still hard to come by at reasonable prices. Even if the features worked it wouldn't matter all that much. The news on Adrenaline sounds nice for current owners though. Bit odd with GV100 showing up on the likely release date though.
 
So, Crimson Relive Edition launched at 16.12 I believe, and now the new one probably being 17.12. Good comparison. Bound to be much faster for many of this years titles. Does it warrant all the commotion?

Ryan from PCPer had a chance to look at Adrenalin, he is saying the release is only about UI improvements and several ease of use features.

 
As for the commotion I'm not expecting much, but I could be wrong there. It's prime holiday shopping season and Vegas are still hard to come by at reasonable prices. Even if the features worked it wouldn't matter all that much. The news on Adrenaline sounds nice for current owners though. Bit odd with GV100 showing up on the likely release date though.

Not really that odd, some of us said awhile ago we thought it would be December or if pushed November, albeit not the 'Volta' Titan probably all expected.
Fingers crossed we get a more prosumer Titan launching early next year, probably a GPU still designed around supporting Universities and labs.
 
Last edited:
Ryan from PCPer had a chance to look at Adrenalin, he is saying the release is only about UI improvements and several ease of use features.
Well shit.

Ok, there went that theory then. Not that improvements and features is necessarily a bad thing. In particular, AMD needs to add hotkey support back in which has been MIA for what, five fricken years now?
 
At least for this last half of year, the Crimson Radeon Settings are a joke of a software. An update to that department is much needed

Pretty UI, some of the buttons don't work, some work only in certain conditions, some work but the UI doesn't show they do.

Also when you move/resize the window, theres a huge chance it just crashes.
 
Also when you move/resize the window, theres a huge chance it just crashes.
Not seen that issue myself, but some versions have crashed for me when closing the radeon settings window, and the auto update feature has been intermittently borked and nonfunctional. It's not exactly great quality software... :p

Also, the UI design is pretty crap. Reaching wattman takes way too many clicks for example and wattman's design is pretty bad as well, spread out too much vertically requiring scrolling up and down to change a setting, then go back up to the top to click "apply" and so on. Feh.
 
Are people playing Games, or do you Play the hole day with wattman? A good UI is nice, but i want a drivere where i dont have to go to the UI. They want better Performance!
 
Are people playing Games, or do you Play the hole day with wattman? A good UI is nice, but i want a drivere where i dont have to go to the UI. They want better Performance!

Beyond your rethoric, I'm glad you have better experience with the UI side. I'm a user who hasn't and of course I want that fixed as well.
 
Back
Top