AMD rumoured to be looking at replacing GCN entirely by 2021

[Master Disaster]

Usual caveat, WCCF so take with a dump truck full of salt....

 

In a rumour that will surprise nobody it seems as though AMD (or RTG to be precise) are looking at having a GCN replacement on the market by 2021 at the latest.

Quote

AMD’s GCN architecture has proven to be one of the most successful macro architectures of its time debuting with the Southern Islands micro-architecture (1st Generation GCN) and lives on, currently, with the Vega micro-architecture (5th Generation GCN). Navi, the upcoming next generation uArch from AMD in 2019 will constitute the 6th Generation GCN and will be the last and final breath of this overarching GCN series.

 

We were told that work had already started on the brand new major-architecture before Raja Koduri even left, but since we are hearing chatter about it then it means that development is picking up. RTG’s new leadership will be leading the design and execution of the major architecture. The efforts on the engineering side will be spearheaded by David Wang and Mike Rayfield on the execution and business side of things.

AMD has hinted about its existence before in this slide and it's looking like Navi will be GCNs final iteration

Quote

AMD has already hinted in its slide deck about the existence of this major architecture. Remember  the “Next-Gen” micro-architecture that was listed in their roadmap? Well, its the one and the same. We don’t know a lot of details about it at this point apart from the fact that it will be based on the 7nm+ process and will succeed GCN sometime in 2020/2021. If all goes as planned, Navi will be the last iteration of GCN macro architecture that hits the market.

 

At this time, information is pretty hush hush and we don’t even have some tantalizing codenames to offer but one thing is for sure, it looks like AMD is planning to improve performance in leaps and bounds (just like GCN did in 2011) over its predecessor and put AMD firmly back in the top spot. You can think of this like the ‘Zen’ for the graphics side of things.

According to WCCF (with no actual evidence to back the statement up) the new architecture will be as big of a jump from GCN as GCN was from Terra

Quote

GCN was first introduced in 2011 to succeed the TeraScale series of micro-architectures and was one of the most disruptive innovations of its time putting the company firmly ahead of the competition (at the time) and putting immense pressure on NVIDIA. For the past year however, RTG has struggled to compete for the top spot in the graphics segment and all eyes are waiting for NAVI to arrive (probably in early 2019).

 

NAVI however, will be the last optimization and iteration of the GCN macro-architecture, the 6th generation to be specific. GCN has had a great run and marked a turning point in AMD’s philosophy when it moved from the VLIW (very long instruction word) to an RISC (reduced instruction set based computing) based approach. This meant that GCN required a vastly greater number of transistors on die then TeraScale (the predecessor to GCN) but also resulted in serious performance and GPGPU computational advantages.

 

It is known that this brand new architecture will result in a leap that is at-least as great as the TeraScale to GCN shift. Since the process will be the 7nm+ optimized node, we can be rest assured that yields will not be a problem (which makes sense considering AMD will want to focus only on the architecture and not bleeding edge node issues at that point). Since the company has other Zen 3 based optimizations planned for this time frame, it is now looking that the company is set to reaching its all-time high of market success on both CPU and GPU sides sometime in 2020/2021.

https://wccftech.com/amd-new-major-gpu-architecture-to-succeed-gcn-by-2020-2021/

 

Typical WCCF to just say things with little to no evidence however the chances of this being true are pretty high IMO.

 

Vega was a failure and GCN is getting old now, it does seem about time for a replacement.

 

Thoughts?

[Devin92]

too late. not even sure RTG can survive to 2021 without competing high end GPU nor adaptation in AI.

[Orcblood]

19 minutes ago, Devin92 said:

too late. not even sure RTG can survive to 2021 without competing high end GPU nor adaptation in AI.

I think you underestimate Navi. Navi will be 7nm, have infinity fabric, and will probably have more mature drivers while still supporting hbm2. That could hold them out till 2021. I also believe that hbm2 memory shortages will be improved (though probably not fixed) and that the 7nm process node will do wonders. Then again Nvidia will more than likely come out with 7nm too soooo really comes down to infinity fabric making GCN6 baller or not compared to GCN5 but I think it will.

 

Also gddr6 will probably help AMD out more on their low end for navi than Nvidia since AMD seems to be more memory starved.

[Devin92]

6 minutes ago, jman629 said:

I think you underestimate Navi. Navi will be 7nm, have infinity fabric, and will probably have more mature drivers while still supporting hbm2. That could hold them out till 2021. I also believe that hbm2 memory shortages will be improved (though probably not fixed) and that the 7nm process node will do wonders. Then again Nvidia will more than likely come out with 7nm too soooo really comes down to infinity fabric making GCN6 baller or not compared to GCN5 but I think it will.

 

Also gddr6 will probably help AMD out more on their low end for navi than Nvidia since AMD seems to be more memory starved.

we can always hope. 28nm to 14nm didnt help tat much, why does 14 to 7nm will be good? HBM1 didnt help tat much, why HBM2 will? The only new thing is infinity fabric, but then again, on the CPU side, its causing a lot of trouble (though nearly fixed by now), so theres no clear indication it wont cause trouble for GPU.

the reality now is NVidia is not threatened by any AMD product, and most likely wont for the next generation. I was once told, nvidia charge 4000 dollars for a product with the combined production cost of no more than 400 (R&D cost included). Why?, just because they have monopoly.

 

Let alone Nvidia's progress in AI and AMD have near 0 adaptation.

[cj09beira]

5 minutes ago, Devin92 said:

we can always hope. 28nm to 14nm didnt help tat much, why does 14 to 7nm will be good? HBM1 didnt help tat much, why HBM2 will? The only new thing is infinity fabric, but then again, on the CPU side, its causing a lot of trouble (though nearly fixed by now), so theres no clear indication it wont cause trouble for GPU.

the reality now is NVidia is not threatened by any AMD product, and most likely wont for the next generation. I was once told, nvidia charge 4000 dollars for a product with the combined production cost of no more than 400 (R&D cost included). Why?, just because they have monopoly.

 

Let alone Nvidia's progress in AI and AMD have near 0 adaptation.

people always seem to forget why "gcn is failing", thought it isnt at all, its just that they reached its max amount of cus and rops(and rbes) (64) with their high end products, the latter was reached with the 290x (its why even today the 290x is still really powerful, especially compared to cards of the time) this is because gcn as of right now is limited to 4 pipelines, 

navi is said to be "scalable" that can mean 2 things:

1 navi is a mcm

2 navi has more pipelines

in both scenarios amd is no longer limited to 64 rops (and rbe) which means better performance when it has 50+cus.

plus by that time amd will probably have enabled some of the vega features that arent working, which will help.

[Devin92]

7 minutes ago, cj09beira said:

people always seem to forget why "gcn is failing", thought it isnt at all, its just that they reached its max amount of cus and rops(and rbes) (64) with their high end products, the latter was reached with the 290x (its why even today the 290x is still really powerful, especially compared to cards of the time) this is because gcn as of right now is limited to 4 pipelines, 

navi is said to be "scalable" that can mean 2 things:

1 navi is a mcm

2 navi has more pipelines

in both scenarios amd is no longer limited to 64 rops (and rbe) which means better performance when it has 50+cus.

plus by that time amd will probably have enabled some of the vega features that arent working, which will help.

sure, i am not denying navi have potential. but just saying it's coming too late. if navi was in 2017, and pans out to be good, then hell yeah. rx290x was in 2013, 4 years later, they still havent break the limitation that you pointed out, that is taking too long. right now, with nvidia has got too far ahead in the AI industry, its really really hard for RTG to catch up and fight back market share. like now, for all the AI companies, the goto hardware is from NVidia, (except some companies that is making their own chip). NV has some competitions in that field, but not from AMD. all the AI libraries are written supporting NV's CUDA core, no one even mentions AMD's gpu. these foundation takes years to build, and can't be caught up instantly.

[Bananasplit_00]

The "like Zen but for GPUs" part has me laughing so you are saying it's going to be a competatively priced special cases product(video editing, rendering so on)  that's slightly behind what the opposition has on the market for games? 

[leadeater]

8 minutes ago, Devin92 said:

AI industry, its really really hard for RTG to catch up and fight back market share. like now, for all the AI companies, the goto hardware is from NVidia, (except some companies that is making their own chip).

Revenue share AI does not make up a large portion for Nvidia, it's a big talking point and focus area currently but it's not their core business and is only a fraction compared to game GPUs.

 

Nvidia is successful because gamers want to buy their cards and they are in all high end laptops.

[leadeater]

1 minute ago, Bananasplit_00 said:

The "like Zen but for GPUs" part has me laughing so you are saying it's going to be a competatively priced special cases product(video editing, rendering so on)  that's slightly behind what the opposition has on the market for games? 

That was Vega until nvidia released the driver update for Titan cards to make them not suck 

[Okjoek]

1 hour ago, Devin92 said:

too late. not even sure RTG can survive to 2021 without competing high end GPU nor adaptation in AI.

Coulda sworn RTG actually made more money with Vega than the stuff before, just that it wasn't with gaming.

[cj09beira]

28 minutes ago, Devin92 said:

sure, i am not denying navi have potential. but just saying it's coming too late. if navi was in 2017, and pans out to be good, then hell yeah. rx290x was in 2013, 4 years later, they still havent break the limitation that you pointed out, that is taking too long. right now, with nvidia has got too far ahead in the AI industry, its really really hard for RTG to catch up and fight back market share. like now, for all the AI companies, the goto hardware is from NVidia, (except some companies that is making their own chip). NV has some competitions in that field, but not from AMD. all the AI libraries are written supporting NV's CUDA core, no one even mentions AMD's gpu. these foundation takes years to build, and can't be caught up instantly.

its simple amd lacked the funds, now they are in a better place, and with navi coming probably q2/q3 2019 it isn't that much late, before that they should focus their efforts in implementing the missing vega features (driver side)

they now have rocm which is a good move, i blame the "consumers" because one should avoid investing into closed source apis made by one company that excludes the other, its a good way of getting yourself trapped to the company with no way out. (it pisses me off each time i pass by our cuda research center in my uni) said money should have been invested into a open api 

[cj09beira]

6 minutes ago, Okjoek said:

Coulda sworn RTG actually made more money with Vega than the stuff before, just that it wasn't with gaming.

not yet, polaris though seems to be a great success (and so would be vega if it was around the same size, as it avoids most of gcn's weakness)

[JediFragger]

As long as RTG can still compete in the lower/mid end of the market then they'll be fine, and within touching distance of the top-end

[Devin92]

32 minutes ago, leadeater said:

Revenue share AI does not make up a large portion for Nvidia, it's a big talking point and focus area currently but it's not their core business and is only a fraction compared to game GPUs.

 

Nvidia is successful because gamers want to buy their cards and they are in all high end laptops.

tat in my opinion is because AI is just taking off. its a new industry. just imagine if automatic cars get popularized in the next few years and AMD has no foundation in support of libraries to compete with NV even if their GPU has the similar performance.

[DanielMDA]

2 hours ago, Master Disaster said:

Usual caveat, WCCF so take with a dump truck full of salt....

 

In a rumour that will surprise nobody it seems as though AMD (or RTG to be precise) are looking at having a GCN replacement on the market by 2021 at the latest.

AMD has hinted about its existence before in this slide and it's looking like Navi will be GCNs final iteration

According to WCCF (with no actual evidence to back the statement up) the new architecture will be as big of a jump from GCN as GCN was from Terra

https://wccftech.com/amd-new-major-gpu-architecture-to-succeed-gcn-by-2020-2021/

 

Typical WCCF to just say things with little to no evidence however the chances of this being true are pretty high IMO.

 

Vega was a failure and GCN is getting old now, it does seem about time for a replacement.

 

Thoughts?

AS I SUSPECTED (I made a few posts about Navi beign the last GCN Iteration). Not surprised, GCN had ver clear limitations while having very good points. I hope the next architecture has better efficienvy, not having each TMU monitoring 4CUs and in general stop the mulit package design of GCN Graphics Cards in favor of a single package for better scalability. Cant wait to see what Navi and the nexta architecture bring to the table! (Hope the Efficiency jump comes in Navi, not later, dat Vega 64 TDP)

[mr moose]

I've been saying for a while now that GCN is a dead end architecture. AMD seems to spend years developing each iteration only to get shut down by Nvidia's next offering (in the case of vega it was nvidia's offering from the year before).

 

I really hope AMD can create something with really good long term outcomes.  Because I am tired of their being only one option on the market.

 

 

[Jito463]

2 hours ago, Master Disaster said:

Vega was a failure and GCN is getting old now

To call Vega a failure is bit too harsh.  Sure, it's overly power hungry as a gaming card, but it's still a great compute card (and as I understand it, with certain tweaks it's still a good gaming card).  Though I do agree that it's time to let GCN rest in peace.

5 minutes ago, mr moose said:

Because I am tired of their being only one option on the market.

There's not "only one option", unless you're chasing the absolute pinnacle of performance.  My R9 390 still performs great, even now.  Would I like to upgrade it?  Yes, though that's not likely to happen in the current market.  Do I need to upgrade it?  No.  AMD cards are still a viable option for gamers, it's only purists who demand the absolutely best benchmarks that would refuse to use them.  So far as I'm concerned, if it's playable, smooth and looks good, then the card is good for gaming.

[Coaxialgamer]

It's worth noting that GCN is used very loosely here. 

Vega and navi don't have much in common with gcn 1.0 (tahiti), just like the 2900xt don't have much in common qith cayman. 

They probably mean it as something thats not a RISC SIMD arch with 64 stream processors per CU. 

[mr moose]

Just now, Jito463 said:

To call Vega a failure is bit too harsh.  Sure, it's overly power hungry as a gaming card, but it's still a great compute card (and as I understand it, with certain tweaks it's still a good gaming card).  Though I do agree that it's time to let GCN rest in peace.

There's not "only one option", unless you're chasing the absolute pinnacle of performance.  My R9 390 still performs great, even now.  Would I like to upgrade it?  Yes, though that's not likely to happen in the current market.  Do I need to upgrade it?  No.  AMD cards are still a viable option for gamers, it's only purists who demand the absolutely best benchmarks that would refuse to use them.  So far as I'm concerned, if it's playable, smooth and looks good, then the card is good for gaming.

I had to  buy a GPU last week (Sons video card died).  I looked at benchmarks and compared local prices.  Yet again, as seems to be the story every time I look (over the last year), Nvidia was either cheaper or performed significantly better across the board, not just in the high end.  I really don't want it to sound like I am shitting on AMD but the reality is they did not have a card that performed adequately compared to cheaper Nvidia cards.  In the end I bought a 1050ti for $250. the closest performing AMD was the 560 which was $10 cheaper but only performing close to the 1050ti in a few titles and never beating it.  

[cj09beira]

17 minutes ago, mr moose said:

I've been saying for a while now that GCN is a dead end architecture. AMD seems to spend years developing each iteration only to get shut down by Nvidia's next offering (in the case of vega it was nvidia's offering from the year before).

 

I really hope AMD can create something with really good long term outcomes.  Because I am tired of their being only one option on the market.

 

 

polaris was great, vega is even better, amd's problem is that at 64cus they needed more rops/rbes, 

add more sensible voltages and power efficiency isn't bad either, plus tsmc's node is better than goflo's 

[cj09beira]

1 minute ago, mr moose said:

I had to  buy a GPU last week (Sons video card died).  I looked at benchmarks and compared local prices.  Yet again, as seems to be the story every time I look (over the last year), Nvidia was either cheaper or performed significantly better across the board, not just in the high end.  I really don't want it to sound like I am shitting on AMD but the reality is they did not have a card that performed adequately compared to cheaper Nvidia cards.  In the end I bought a 1050ti for $250. the closest performing AMD was the 560 which was $10 cheaper but only performing close to the 1050ti in a few titles and never beating it.  

well at mrsp a 570 is only a bit more expensive but much faster and the 560 should be even cheaper, the 570 at msrp is the price/perf king, right now is not the best of times to buy gpus though  .

there is a reason why amd cards are more expensive, they are better at mining,

[mr moose]

Just now, cj09beira said:

polaris was great, vega is even better, amd's problem is that at 64cus they needed more rops/rbes, 

add more sensible voltages and power efficiency isn't bad either, plus tsmc's node is better than goflo's 

 

And all that is great, but at the end of the day if they spend a year developing it and then their competition doesn't batter an eyelid and doesn't even bother releasing their next gen in a domestic product, then AMD are working hard for very little if any competitive edge. 

 

Only thing that matters in this market is performance and price.  I don't care how many universal joints or springs it has, if it is cheaper and performs better I'll buy it.

[mr moose]

Just now, cj09beira said:

well at mrsp a 570 is only a bit more expensive but much faster and the 560 should be even cheaper, the 570 at msrp is the price/perf king, right now is not the best of times to buy gpus though  .

there is a reason why amd cards are more expensive, they are better at mining,

 

I know, but MSRP means shit and I can only buy whats available.  Also it doiesn't matter why they are more expensive, the fact is they are.  I am not going to spend more for a card that doesn't perform where I need it to.  That'd be like buying an expensive truck because they can pull more weight when I need speed and a cheaper car offers that.

[cj09beira]

1 minute ago, mr moose said:

 

And all that is great, but at the end of the day if they spend a year developing it and then their competition doesn't batter an eyelid and doesn't even bother releasing their next gen in a domestic product, then AMD are working hard for very little if any competitive edge. 

 

Only thing that matters in this market is performance and price.  I don't care how many universal joints or springs it has, if it is cheaper and performs better I'll buy it.

thats your problem right there, you can't not care about how/why things are how they are, and at the same time shout that gcn isn't good.

[cj09beira]

Just now, mr moose said:

 

I know, but MSRP means shit and I can only buy whats available.  Also it doiesn't matter why they are more expensive, the fact is they are.  I am not going to spend more for a card that doesn't perform where I need it to.  That'd be like buying an expensive truck because they can pull more weight when I need speed and a cheaper car offers that.

sure but you missing the point, those prices are inflated meaning their position in the market changed, and its something amd can't do much about.