AMD has not announced all of Vega's architectural improvements

[WYP]

AMD's Scott Wasson has revealed that AMD has not announced all of the architectural improvements that are present in their upcoming Vega GPUs.



Read more on AMD's Vega GPU architecture.

What performance improvements do you think AMD have up their sleeves?

 

[AlienALX]

Hmm. AdoredTV said they were hiding something but couldn't quite put his finger on it lol. This was ages ago too !

 

[AngryGoldfish]

But will all this translate to a truly competitive and staggering product? AMD's Vega architecture looks to be vastly more intelligent than Pascal, but it still falls behind. If the final performance is around what we've seen so far, a 500mm chip is only slightly faster than a 300mm chip from Nvidia that's a year older.

 

[SuB]

But wait! There's more!!

 

[RobM]

I still think its gonna slap the 1080 down with room to spare

 

[Wraith]

I still think its gonna slap the 1080 down with room to spare

Nah, just more architecture stuff that won't be supported properly until the next, next, next gen cards come out and devs decide it's worth utilising.. looks good on paper but at the end of the day unless it's supported and fully functional it's just idle hardware.

 

[AlienALX]

But wait! There's more!!

They think it's all over.......

It isn't apparently

Nah, just more architecture stuff that won't be supported properly until the next, next, next gen cards come out and devs decide it's worth utilising.. looks good on paper but at the end of the day unless it's supported and fully functional it's just idle hardware.

You could well be right knowing AMD. However, AdoredTV seems to think they may be deliberately holding back performance. They may have done this for several reasons, but my guess is this.

Whenever AMD release a new GPU (ever since they changed to AMD from ATI oddly enough) they give us a rough idea of performance and we all go "FFS". Then the card comes out and quelle surprise, it's crap.

They may have deliberately derped the card so that when they release it it's actually a surprise. Like, it may actually be far better than we were assuming. Which would be the correct way to do it, IMO. That's what Nvidia do. Say nothing, show a few benchmarks and then defecate the bed it's actually as good as they promised and more.

Fury X was ultimate fail. Overclocker's dream they said. As fast as X Y Z they said. Yet in the cold light of day it was obvious they had been talking through their back end.

 

[AngryGoldfish]

They think it's all over.......

It isn't apparently



You could well be right knowing AMD. However, AdoredTV seems to think they may be deliberately holding back performance. They may have done this for several reasons, but my guess is this.

Whenever AMD release a new GPU (ever since they changed to AMD from ATI oddly enough) they give us a rough idea of performance and we all go "FFS". Then the card comes out and quelle surprise, it's crap.

They may have deliberately derped the card so that when they release it it's actually a surprise. Like, it may actually be far better than we were assuming. Which would be the correct way to do it, IMO. That's what Nvidia do. Say nothing, show a few benchmarks and then defecate the bed it's actually as good as they promised and more.

Fury X was ultimate fail. Overclocker's dream they said. As fast as X Y Z they said. Yet in the cold light of day it was obvious they had been talking through their back end.

I thought I remembered hearing AdoredTV say in his latest two videos that AMD have always put their best foot forward and are likely doing the same thing again. He seemed to think that what we saw is what we'll get. He initially might have said AMD are deliberately holding back performance, but then he analysed their past and has changed his opinion.

 

[WYP]

It makes sense for AMD to hold back information, as it prevents Nvidia from copying any a lot of their ideas.

Nvidia started using Tile based rasterizers with their first maxwell GPUs and never told anyone outside the company about it for years. Only now with Vega are AMD catching up with this specific feature.

It will be interesting to see what AMD does with Vega, though supporting extra DX12 features are a must for the sake of "future-proofing" the architecture. It was AMD's early adoption of Async compute that make AMD's R9 290X a much better GPU than the GTX 780/Ti today afterall.

AMD has already significantly improved their tessellation performance in recent years (R9 285 or newer) and MSAA performance with Polaris. It will be interesting to see what Vega focuses on improving.

A lot of Vega seems to be about making their architecture smarter, with their new rasterizer and other features allowing the GPU to waste less time on needless work/unseen pixels and getting more performance per clock/cycle of the GPU. Hopefully, this will lead to an insane increase in efficiency and performance, as AMD needs to beat the 1080 Ti and mot just the GTX 1080.

Early reports were saying that the samples shown at the Pre-Christmas Zen event and CES were only a few weeks old, which means that the early benchmarks/performance shown is pretty meaningless.

I am hopeful that AMD will have something good here, though mostly because I want Nvidia to actually work for their money for a change.

 

[AlienALX]

I thought I remembered hearing AdoredTV say in his latest two videos that AMD have always put their best foot forward and are likely doing the same thing again. He seemed to think that what we saw is what we'll get. He initially might have said AMD are deliberately holding back performance, but then he analysed their past and has changed his opinion.

haha I think if any one analysed their past they would come to the same conclusion really

 

[AngryGoldfish]

It makes sense for AMD to hold back information, as it prevents Nvidia from copying any a lot of their ideas.

Nvidia started using Tile based rasterizers with their first maxwell GPUs and never told anyone outside the company about it for years. Only now with Vega are AMD catching up with this specific feature.

It will be interesting to see what AMD does with Vega, though supporting extra DX12 features are a must for the sake of "future-proofing" the architecture. It was AMD's early adoption of Async compute that make AMD's R9 290X a much better GPU than the GTX 780/Ti today afterall.

AMD has already significantly improved their tessellation performance in recent years (R9 285 or newer) and MSAA performance with Polaris. It will be interesting to see what Vega focuses on improving.

A lot of Vega seems to be about making their architecture smarter, with their new rasterizer and other features allowing the GPU to waste less time on needless work/unseen pixels and getting more performance per clock/cycle of the GPU. Hopefully, this will lead to an insane increase in efficiency and performance, as AMD needs to beat the 1080 Ti and mot just the GTX 1080.

Early reports were saying that the samples shown at the Pre-Christmas Zen event and CES were only a few weeks old, which means that the early benchmarks/performance shown is pretty meaningless.

I am hopeful that AMD will have something good here, though mostly because I want Nvidia to actually work for their money for a change.

But it was mentioned that it takes so long to mass-produce a final design and release it to the market that any changes that can be made have to be made in the next few weeks. Beyond that and they'll go past their release deadline of June. I don't know how true that is, but it's what I heard. I personally find it hard to believe that what we saw at CES was anywhere near close to the final silicon. I had imagined that they might need 2-3 months to mass-produce, which leaves 2-3 months to improve the chip and the software. That's a very simple-minded and somewhat ignorant way of looking at it, but I'm no engineer.

 

[Shingara]

Well that only comes into effect if A, that was the best of the best at CES and B, that is the latest they have at CES. And from my understanding, these were the engineering samples that were at the new horizen event.

So what there showing and what they have maybe radically different things.

Edit, i also have to wonder with the way ryzen is designed if there isnt things that they havnt shown that is a major benefit for using these with ryzen.

 

[AngryGoldfish]

Well that only comes into effect if A, that was the best of the best at CES and B, that is the latest they have at CES. And from my understanding, these were the engineering samples that were at the new horizen event.

So what there showing and what they have maybe radically different things.

I'm hoping so. I was never expecting Titan XP-beating performance. I wanted slightly above overclocked 1080 performance in DX11 games, such as Dishonored 2, and significantly more (as you would expect) in DX12 games like Deus Ex, coming close to Titan XP—but only in DX12 games that favour AMD heavily such as Hitman and the aforementioned Deus Ex.

 

[Shingara]

well when it comes to dx11 and dx12 and these cards, dont we have to look how 'gameworks' has been integrated into some games and the effect it has not just on amd cards but also on old gen nvidia cards.

looking forwards and how amd has got into the console and its sort of forcing game devs in amd's technology path, and i hate using this term but its the future we look at. So performance on old games maybe meh but games that are developed to take full advantage of whats in vega that could be the key.

 

[WYP]

But it was mentioned that it takes so long to mass-produce a final design and release it to the market that any changes that can be made have to be made in the next few weeks. Beyond that and they'll go past their release deadline of June. I don't know how true that is, but it's what I heard. I personally find it hard to believe that what we saw at CES was anywhere near close to the final silicon. I had imagined that they might need 2-3 months to mass-produce, which leaves 2-3 months to improve the chip and the software. That's a very simple-minded and somewhat ignorant way of looking at it, but I'm no engineer.

Sorry, I could have been clearer with what I said. From what I hear those chips were pretty fresh, definitely not the first samples AMD has produced but certainly close to final. If the manufactured before the Zen event info was true it means they were made in mid-late November (approximately).

From seeing the boards in Linus' video we can see that AMD was at the phase where they were optimising internal power designs/voltages in order to see what kinds of clocks/ power efficiency that these samples can deliver.

(Please note that AMD was likely not running these early chips at their full potential, as nobody wants their PC to have any chance of crashing in front of the press)

AMD were not showing power draw in their tests or comparing it to a different GPU, so we can tell that clocks/ power draw/design are not final. We will likely see Vega GPUs coming up to GDC where AMD will deliver some more information, perhaps at a new capsaicin event.

We need to remember Polaris if we want to think about release timeframes, with AMD showing their first samples at GDC 2016 and releasing the GPUs in June. AMD have shown us working Vega samples much earlier, though it seems like they have shown them at an earlier development state.

As always it is hard to judge a GPUs performance based on early samples, though there is more reason to be optimistic about Vega than with previous AMD designs.

 

[AngryGoldfish]

well when it comes to dx11 and dx12 and these cards, dont we have to look how 'gameworks' has been integrated into some games and the effect it has not just on amd cards but also on old gen nvidia cards.

looking forwards and how amd has got into the console and its sort of forcing game devs in amd's technology path, and i hate using this term but its the future we look at. So performance on old games maybe meh but games that are developed to take full advantage of whats in vega that could be the key.

AMD powered previous consoles and it didn't appear to aid them very much. Maybe it did, but it's not evident to me.

Sorry, I could have been clearer with what I said. From what I hear those chips were pretty fresh, definitely not the first samples AMD has produced but certainly close to final. If the manufactured before the Zen event info was true it means they were made in mid-late November (approximately).

From seeing the boards in Linus' video we can see that AMD was at the phase where they were optimising internal power designs/voltages in order to see what kinds of clocks/ power efficiency that these samples can deliver.

(Please note that AMD was likely not running these early chips at their full potential, as nobody wants their PC to have any chance of crashing in front of the press)

AMD were not showing power draw in their tests or comparing it to a different GPU, so we can tell that clocks/ power draw/design are not final. We will likely see Vega GPUs coming up to GDC where AMD will deliver some more information, perhaps at a new capsaicin event.

We need to remember Polaris if we want to think about release timeframes, with AMD showing their first samples at GDC 2016 and releasing the GPUs in June. AMD have shown us working Vega samples much earlier, though it seems like they have shown them at an earlier development state.

As always it is hard to judge a GPUs performance based on early samples, though there is more reason to be optimistic about Vega than with previous AMD designs.

I understand what you mean, and I agree.

 

[Shingara]

Well the difference to see in most cases is if devs get a huge wad of cash to stamp gameworks on there games. Just have to look at the fun and games we all had with fallout 4.

 

[RobM]

What if, (and I am only theorising here) but lets imagine AMD have final samples almost ready and they chose to show early samples because they had good performance whilst keeping the real picture hidden.
Not likely but very possible although I would have though some leaks would be around by now.

 

[WYP]

One thing that needs to be considered is that AMD had a lot more influence than Nvidia when DX12 was designed, which is why AMD's Async Compute tech is now taking off.

The Original Xbox (or DirectXbox), was using an Nvidia GPU, which gave them a lot of influence in the early DirectX iterations. With DirectX 12 the ball has very much been in AMD's court, as DirectX 12 is designed for the Xbox One and PC.

The reason why AMD's implementation of Async Compute is adopted by pretty much every meaningful DirectX 12 game is because it also helps the consoles. This is one key area where AMD will clearly benefit from DirectX 12 moving forward (Not even Pascal has AMD's level of Async compute support).

AMD's 8-core APU in the Xbox has also forced DirectX 12 to become much more multi-threaded, which also has clear benefits for AMD on both the CPU and GPU (GPU driver overhead) sides.

 

[Shingara]

I think if anything they might be gimping the cards via drivers and not so much hardware, thats if there gimping them at all, but as for leaks, well apart from someone on here who knows a man in asia region i dont see many if any at all leaks when it comes to ryzen never mind vega.