AMD's next-gen RDNA 2 rumor: 40-50% faster than GeForce RTX 2080 Ti

[Vayra86]

What ? How is it "ancient" ?

Its just GCN with different letters and new VRAM.

 

[Vya Domus]

Its just GCN

No, it's not. How many times will we go over this ?

 

[Vayra86]

No, it's not. How many times will we go over this ?

Okay, its not, but the difference is irrelevant to the market it caters to, maybe that explains it better. Technical differences need to translate to customer advantages, Navi does not do this. It just manages to keep up. Nvidia on the other hand translates new architecture into new features. Navi 2 is hopefully doing similar.

 

[Vya Domus]

The market it catered to was mid range, where the most money AMD could have made. That market does not care if it's GCN or not or even about much in the way of features, they just care about perf/price.

 

[Vayra86]

The market it catered to was mid range, where the most money AMD could have made. That market does not care if it's GCN or not or even about much in the way of features, they just care about perf/price.

The market definitely does care about features, that seems to be a returning mistake in considerations between the two camps. More importantly, the feature set in the mid range actually. There are segments and one feeds the other, Navi 1 is in no way feeding anything but revenue. It does not make up a stack, because Navi 2 will make 1 obsolete with a bigger feature set. Its basically an odd one out, and it is based on updated, old technology.

But you know all this and in a basic sense I don't think we see different things here. We just appraise them differently.

 

[Super XP]

The market definitely does care about features, that seems to be a returning mistake in considerations between the two camps. More importantly, the feature set in the mid range actually. There are segments and one feeds the other, Navi 1 is in no way feeding anything but revenue. It does not make up a stack, because Navi 2 will make 1 obsolete with a bigger feature set. Its basically an odd one out, and it is based on updated, old technology.

But you know all this and in a basic sense I don't think we see different things here. We just appraise them differently.

I really didn't want to get into the history as to why VEGA and why NAVI etc., short and sweet.
AMD put most of its resources into ZEN back in 2012 when Jim Keller was hired. So they dabbled just enough in the GPU department just to stay afloat and compete on a price/performance level. That's basically all they could do, without losing the server market fully to Nvidia.
As soon as ZEN was released in 2017, a lot more resources went into the Radeon Technology Group to develop RDNA 1. After ZEN's success, especially ZEN+ & 2's success, AMD moved several key ZEN engineers to assist the Radeon Technology Group and to help finish off the new GPU design known as RDNA2.

AMDs bread and butter is in CPUs. AMDs strategy actually worked out quite well. Now we will soon find out if this strategy worked to pull its GPU department out of the gutter and into Nvidias fat face.

AMD Processor Designs Leading up to the Superior ZEN Micro-Architecture

Loop Finance

trybe.one

 

[EarthDog]

I really didn't want to get into the history as to why VEGA and why NAVI etc., short and sweet.
AMD put most of its resources into ZEN back in 2012 when Jim Keller was hired. So they dabbled just enough in the GPU department just to stay afloat and compete on a price/performance level. That's basically all they could do, without losing the server market fully to Nvidia.
As soon as ZEN was released in 2017, a lot more resources went into the Radeon Technology Group to develop RDNA 1. After ZEN's success, especially ZEN+ & 2's success, AMD moved several key ZEN engineers to assist the Radeon Technology Group and to help finish off the new GPU design known as RDNA2.

AMDs bread and butter is in CPUs. AMDs strategy actually worked out quite well. Now we will soon find out if this strategy worked to pull its GPU department out of the gutter and into Nvidias fat face.

AMD Processor Designs Leading up to the Superior ZEN Micro-Architecture

Loop Finance

trybe.one

Interesting take. For all intents and purposes RTG and the CPU group are separate, right? Meaning they could focus, concurrently, on CPUs and GPUs at the same time as they are two different things too. Different people, etc. What Zen engineers were moved over for GPU?

 

[Super XP]

Interesting take. For all intents and purposes RTG and the CPU group are separate, right? Meaning they could focus, concurrently, on CPUs and GPUs at the same time as they are two different things too. Different people, etc. What Zen engineers were moved over for GPU?

With regards to other companies, that may be so, but I was referring to an article I read way back, and an interview from Dr. Lisa Su or one of the companies executives, in which they explained what I posted. If I come by this information again, I will be happy to share it. I have great memory, I just need to make it a habit of linking sources.

 

[Vayra86]

I really didn't want to get into the history as to why VEGA and why NAVI etc., short and sweet.
AMD put most of its resources into ZEN back in 2012 when Jim Keller was hired. So they dabbled just enough in the GPU department just to stay afloat and compete on a price/performance level. That's basically all they could do, without losing the server market fully to Nvidia.
As soon as ZEN was released in 2017, a lot more resources went into the Radeon Technology Group to develop RDNA 1. After ZEN's success, especially ZEN+ & 2's success, AMD moved several key ZEN engineers to assist the Radeon Technology Group and to help finish off the new GPU design known as RDNA2.

AMDs bread and butter is in CPUs. AMDs strategy actually worked out quite well. Now we will soon find out if this strategy worked to pull its GPU department out of the gutter and into Nvidias fat face.

AMD Processor Designs Leading up to the Superior ZEN Micro-Architecture

Loop Finance

trybe.one

But I don't question that and I do agree that their overall strategy was good when it comes to Zen. I don't agree that it was a wise move regarding GPU - the moment they announced their focus on midrange with Polaris was the moment they fell behind so far that they still struggle to keep up - even with said midrange. Navi only underlines this so far, being clocked out of its efficiency curve to be competitive at its price point. They really did play a dangerous game the past years, giving Nvidia every reason and chance to come out on top.

As far as focus is concerned for AMD then yes, even their GPU strategy was wise because they simply didn't have resources for more and their focus for performance was targeted around maximum use of their technology - in consoles, with Intel (KabyLake with their IGP), towards dGPU and professional markets. But that does not affect me as a customer. That is an important distinction and one reason for the disconnect I often see in these discussions.

 

[Assimilator]

Regardless of whether RDNA is a rebrand of GCN or not, it can't be denied that Navi made massive efficiency strides against NVIDIA. Recall that the previous best AMD GPUs, Fury and Vega, were such power hogs that they had to use HBM, and even then weren't power-competitive with NVIDIA - but Navi uses GDDR and competes with Turing in perf/W. It's a pretty remarkable improvement.

The big question is whether AMD is able to continue this trend with RDNA2, especially considering the inclusion of RT hardware.

 

[EarthDog]

Regardless of whether RDNA is a rebrand of GCN or not, it can't be denied that Navi made massive efficiency strides against NVIDIA. Recall that the previous best AMD GPUs, Fury and Vega, were such power hogs that they had to use HBM, and even then weren't power-competitive with NVIDIA - but Navi uses GDDR and competes with Turing in perf/W. It's a pretty remarkable improvement.

The big question is whether AMD is able to continue this trend with RDNA2, especially considering the inclusion of RT hardware.

55/5600XT, sure. 5700 XT, was punching above its weight class and a bit out of its sweetspot. It loses out to the 2070S by 12% (1440) while that card uses 5% less power. It isn't a big difference and better agreed, but 5700XT wasn't a mark of efficiency.

I'm wondering if we'll see more of the same with bNavi.

 

[Super XP]

But I don't question that and I do agree that their overall strategy was good when it comes to Zen. I don't agree that it was a wise move regarding GPU - the moment they announced their focus on midrange with Polaris was the moment they fell behind so far that they still struggle to keep up - even with said midrange. Navi only underlines this so far, being clocked out of its efficiency curve to be competitive at its price point. They really did play a dangerous game the past years, giving Nvidia every reason and chance to come out on top.

As far as focus is concerned for AMD then yes, even their GPU strategy was wise because they simply didn't have resources for more and their focus for performance was targeted around maximum use of their technology - in consoles, with Intel (KabyLake with their IGP), towards dGPU and professional markets. But that does not affect me as a customer. That is an important distinction and one reason for the disconnect I often see in these discussions.

Agreed.
Sometimes when you are shedding $$$$ because of the Bulldozer debacle you have no choice but to GUT the GPU department and pour all resources into the companies Bread and Butter, the CPU.

Great point though,

 

[RedelZaVedno]

Never sit on a GPU hype train, if you don't want to derail

 

[Super XP]

Never sit on a GPU hype train, if you don't want to derail

RDNA2 is the real deal. Finally!!!!!!

 

[EarthDog]

RDNA2 is the real deal. Finally!!!!!!

^^

Spurs dug in that horse..........

 

[moproblems99]

^^

Spurs dug in that horse..........

Conductor you mean?

 

[kapone32]

When I look at AMD GPUs I only compare them to themselves. Was the 7950 noticeably faster than the 6850 yes. Was Polaris faster than Tahiti? Not really but at noticeably less power. Was Vega noticeably faster than Polaris yes. Is RDNA faster than Vega. Yes and it uses less power at 2/3 the power draw and 1/2 the processors. If the Big Navi has the refinements of a year's more knowledge of the node and the same or similar amount of processors as Vega 64 it should be noticeably faster than the 5700XT. This could be why Nvidia is talking about a 12 pin connector for their upcoming GPUs. What I don't like is how people can be so quick to focus on the high end crown that they forget that there is a huge and I mean huge price difference between the 2080TI (Propoganda's darling) vs the 5700XT which costs less than 1/2 in some cases. If AMD does produce a Navi card specced similar to the Vega 64 though I would venture to say it should be a very decent card.

 

[EarthDog]

it should be noticeably faster than the 5700XT.

Of course. But the 5700XT is also nearly 50% slower than the 2080Ti. Price to performance doesn't scale and never has... especially with true flagships and high-performance.

But only comparing anything to itself/previous generations feels myopic to me. Akin to a horse running a race with its blinders on so they can't see the other horses next to it and get freaked out (with ponies, that is good, not so with products). To get a good idea of the market and a product's placement, you have to consider the competition as part of the metric. Then you filter the data by your own personal needs. Some couldn't care less about power, for example, others do (and yet others only care when the competitor uses more, LOL).

 

[moproblems99]

Of course. But the 5700XT is also nearly 50% slower than the 2080Ti. Price to performance doesn't scale and never has... especially with true flagships and high-performance.

But only comparing anything to itself/previous generations feels myopic to me. Akin to a horse running a race with its blinders on so they can't see the other horses next to it and get freaked out (with ponies, that is good, not so with products). To get a good idea of the market and a product's placement, you have to consider the competition as part of the metric. Then you filter the data by your own personal needs. Some couldn't care less about power, for example, others do (and yet others only care when the competitor uses more, LOL).

In a competition, the competitor very much matters. AMD doesn't need to win this round to start crawling out if the gutter. A good start is to make it look like their GPU group is actually trying. Within 10-20% of NV top card is all they need this round and that isn't going to be easy. They are going to have to double Navi 1.

 

[FinneousPJ]

Would be great if true. However I'm expecting 20-30% better than 2080 To which is already HUGE.

 

[EarthDog]

In a competition, the competitor very much matters. AMD doesn't need to win this round to start crawling out if the gutter. A good start is to make it look like their GPU group is actually trying. Within 10-20% of NV top card is all they need this round and that isn't going to be easy. They are going to have to double Navi 1.

Thats what im saying too more or less.

5700xt is 45% behind a 2080ti. RDNA2 flagship has to overcome that deficit with a process tweak and new architecture while NV has a big process shrink and new architecture. Even if Ampre's non titan flagship is 30% faster than a 2080ti (same margin as 1080ti to 2080ti) that's 75% increase to match it. I dont recall ever seeing a 75% increase from flagship to flagship, do you? Id be happy if it was within 10% of Ampre honestly... power be damned. I dont recall a 65% increase either.........

Then... these rumors of ampre using an arseload of power (300W nominal, 400w oc) on a new arch and process shrink..... id have to imagine its going to be well over 30%.... approaching 50% if that all pans out.

Where does that leave big navi? Needing to be almost 100% faster? Yikes..

 

[moproblems99]

Thats what im saying too more or less.

5700xt is 45% behind a 2080ti. RDNA2 flagship has to overcome that deficit with a process tweak and new architecture while NV has a big process shrink and new architecture. Even if Ampre's non titan flagship is 30% faster than a 2080ti (same margin as 1080ti to 2080ti) that's 75% increase to match it. I dont recall ever seeing a 75% increase from flagship to flagship, do you? Id be happy if it was within 10% of Ampre honestly... power be damned.

I know what you meant, just extending. My memory isn't good enough (nor do I care enough) to remember past yesterday

There isn't any reason they can't do it. It is just a massive task and will be super impressive if they even come close.

 

[Dante Uchiha]

AMD could already hit/tie with the 2080 ti using an RDNA1 GPU w/ 64CU and better memory, has an article in techspot that shows well that Turing and RDNA1 are architectures of the similar level. AMD's problem is software, not hardware, they need to focus on that point...I'm not just talking about fixing bugs, I'm talking about working on optimizing more games to maintain a more uniform superiority/performance.

 

[Fluffmeister]

Yeah it's interesting, one of the common myths that was peddled was AMD would pull ahead when they got both major consoles using their hardware, thus all games would run great on AMD hardware, and slowly and surely Nvidia would suffer till eventually they died. The world would then be a better place and gamers and children alike would play with gumdrop smiles.

 

[dragontamer5788]

Its just GCN with different letters and new VRAM.

Navi is actually a huge change in architecture.

Navi is Wave32 native. While GCN was Wave64 native. A huge amount of lower-level code doesn't work anymore. The assembly language is grossly different, especially with regards to cross-lane operations. I'd say RDNA / Navi is the biggest leap in 10 years for AMD... strictly from an assembly-language perspective.

https://developer.amd.com/wp-content/resources/RDNA_Shader_ISA.pdf

https://developer.amd.com/wp-content/resources/Vega_Shader_ISA_28July2017.pdf

---------

NVidia's stuff is all PTX-compatible, but that doesn't change the huge leaps they made from Maxwell through Turing. Similarly, AMD has kept large parts of GCN compatible (ie: there's still a Multiply, Dot-product, and XOR instruction). But the major changes to architecture (Wave32, independent load/stores, etc. etc.) make Navi a hugely different beast.

Different enough that Navi isn't supported by AMD's ROCm platform. Because ROCm makes a lot of Wave64 assumptions. Navi literally broke a bunch of code that AMD has been trying for months to fix.

---------

AMD's weaker software team has shown the advantage NVidia has. Something like Navi (with huge leaps) comes out, but the software isn't really ready to take off (at least in the compute sector. DirectX / Vulkan seemed ready). NVidia can be more agile since they have a leaner, quicker moving, more mature software team. This also hit them when Vega came out, and their software team was unable to get primitive shaders implemented. Maybe it was the hardware team's fault. Or maybe it was the software team's fault. But a more agile software process would have figured out that primitive shaders weren't ready long before the release.

Overall, AMD has hardware that's as strong, or maybe even stronger, than NVidia's. But their weak software is holding them back.