** The AMD Navi Thread **

[D.P.]

Stop the trolling, please.

The cache size and hierarchy is totally different. The CU / SP setup is completely different. Physical design is clearly very different. What are you expecting a new architecture to do? To be some completely new paradigm which ditches current CU / SP designs completely?

Who cares what they call it? RDNA or GCN? It's clearly a large departure from Fiji / Polaris / Vega, and does indeed appear to be a dedicated gaming architecture.

Most interesting thing is that clearly their GPU product stack is going to completely diverge, with GCN staying for enterprise and 'pro' markets.

P.S. If people missed it, Su said both that further Vega / GCN products are in the pipeline for non-consumer products, and that the 'Next Gen' architecture referenced in slides for next year is 'RDNA' based.

The cache hierarchy was already changed in Vega AFAIK and the CU to SP change with double the shader engines I already commented on. That was the natural extension of GCN, just the same way as nvidia changes the ratio of compute cores and CUDA units between generation to maintain efficiency with scaling.


You are othe one that seems to be upset at the fact that this is an evolution of GCN and not a whole architecture. I clear;ly stated it is irrelevant, and only the price and performance is what matters.

 

[D.P.]

So it is a new architecture then . Make your mind up!

No, it is an evolution of GCN but liekly a bigger change than in the last few generations.

 

[D.P.]

No they never, its guys like you that misinterpreted what was said for your own purpose.

I Can not be bothered digging ram through posts but I clearly remember you stating that amd wouldn't release a high end GPU to complete with 1080Ti when in fact they did better and brought a GPU to market on level with a 2080.

This is revisionist nonsense. I merely quoted what AMD said.

And it can be easily argued that AMD still have not go a GPU to go against the 2080. The radeon 7 is just salvaging Instinct cards at a loss. AMD have not released a Navi card to firmly go against the 2080 and 2080ti, maybe next year

 

[Ross Thomson]

The architecture is kinda irrelavent. The performance and cost is what I care about.

 

[D.P.]

ITT DP on the troll, why am i not surprised.

pointing out facts I think you will find.

 

[Gerard]

pointing out facts I think you will find.

LOL, spinning "facts" more like which is what you usually do with your selective memory.

 

[D.P.]

The architecture is kinda irrelavent. The performance and cost is what I care about.

Exactly, yet some people get so worked when you point out such things. I would actually be much more worried if this was a whole new architecture, as AMD have a very mixed track record. Resolving some of the issues with GCN i a much more certain path to performance gain given AMD''s R&D budget.

 

[D.P.]

LOL, spinning "facts" more like which is what you usually do with your selective memory.

No, pointing out facts. You will have to start quoting anything you disagree with.

 

[Gerard]

No, pointing out facts. You will have to start quoting anything you disagree with.

If i was to quote all the bs you've spouted it would probably crash the forum ffs.

 

[D.P.]

If i was to quote all the bs you've spouted it would probably crash the forum ffs.

well then you will find it trivial to provide 1 single quote then

 

[Gerard]

well then you will find it trivial to provide 1 single quote then

Nah, i'll leave you to your spin and delusions, enjoy

 

[D.P.]

Nah, i'll leave you to your spin and delusions, enjoy

So I am right and you are just full of BS then. Thanks for the clarification.

 

[Gerard]

So I am right and you are just full of BS then. Thanks for the clarification.

Amazing, a post from you without a typo, congrats, shill boy

 

[CAT-THE-FIFTH]

Lisa Su said RDNA was different to GCN,and GCN would be kept for non-gaming cards. The rumours it can move past the 4 geometry engine limit,is enough to show its the biggest change AMD have had since moving from Terascale to GCN. I would imagine like with the HD7000 series drivers will need to be in big shape also.

 

[Martini1991]

Lisa Su said RDNA was different to GCN,and GCN would be kept for non-gaming cards.

Different doesn't mean much.
Polaris was different and new.

 

[D.P.]

https://www.anandtech.com/show/1441...ducts-rx-5700-series-in-july-25-improved-perf

In what the company is calling their Radeon DNA (RNDA) archtiecture – I should note that it’s not clear if this is a branding exercise to downplay the GCN family name or if it's a more heavily overhauled architecture

The details AMD have released indicate this is indeed a branding exercie, but the changes to GCN are more more than the last minor changes. If AMD release more details suggesting something significant has changed then we might be able to reconsider but form AMD's description this is the same RISC SIMD based GCN.

Anandtech have clarified the efficiency claims a little, so the 50% increase does of course reflect going form 14nm to 7nm, which is know to give around 50% in itself.

What you can take form this is the 5700 will be around Vega64 to V64 + 10% while drawing around 200w-220w. This is much more reasonable.



I think the biggest difference is liekly going to come on the games that got severely bottle-necked by the 4 shader engine limit in previous GCN architectures. The games where AMD really lagged behind could see a big jump in performance, while other will see small changes dependent on resolution and setting.

 

[CAT-THE-FIFTH]

Different doesn't mean much.
Polaris was different and new.

Polaris was GFX8,same as Fuji and Tonga. Also GCN has had a 4 geometry engine limit since the start. Rumours hint at 8 geometry engines and possibly less shaders per group. My main concern will be how drivers will shape up though. The HD7000 series did have its niggles when launched IIRC.

 

[D.P.]

Lisa Su said RDNA was different to GCN,and GCN would be kept for non-gaming cards. The rumours it can move past the 4 geometry engine limit,is enough to show its the biggest change AMD have had since moving from Terascale to GCN. I would imagine like with the HD7000 series drivers will need to be in big shape also.

AMD also said Vega had new compute units and was different to GCN. The limited details AMD have given for Navi don't at all suggest a significant change as when AMD/ATI moved from terrascale to GCN, but a bigger change than we have see form Fuji to Polaris to vega.


When Lisa Su said GCN is staying of non-gaming, she means the current Instinct cards are not being replaced and navi is only for gaming. In the future who know exactly what will happen but there is a strong chance that the next compute focused AMD GPU wioll be based on the next Navi architecture, but that is at least 1 year away.

 

[Ross Thomson]

Rumours hint at 8 geometry engines and possibly less shaders per group. My main concern will be how drivers will shape up though. The HD7000 series did have its niggles when launched IIRC.

This doesn't bother me as I won't be buying at launch (same for a new CPU). I'll be waiting for reviews and then choose a new card (may not even be navi)

 

[D.P.]

Polaris was GFX8,same as Fuji and Tonga. Also GCN has had a 4 geometry engine limit since the start. Rumours hint at 8 geometry engines and possibly less shaders per group. My main concern will be how drivers will shape up though. The HD7000 series did have its niggles when launched IIRC.

AMd had repeatedly said that GCN was not limited to 4 geometry engines or 4 shader engines, but it would require changes and they didn't think those changes would warrant the performance upgrade given the transistor costs. With 7nm, AMD can spend more transistors widening up the backend.