New AMD Listings in Korean RRA Certification Point to Impending Graphics Cards Release - Big Navi?

[Tomorrow]

Do you, guys, realize that you are arguing over a non-existent product?
Rumors of "Big Navi" have been debunked some time ago, and "Arcturus" is a made-up name that someone came up with almost 2 years ago and everyone still uses it (even though it's not a thing).
Having a 225W RX5700XT is already a good-enough sign that consumer cards are near the upper TDP cap, so the most we could actually hope for is a 48CU Navi card, maybe on 7nm+ process.
Also, the majority of consumer GPU market segments are covered already, and as with Polaris, there's no need to rush for that 0.01% that'll give up their 2080Ti in favor of marginally faster AMD flagship.

Non-existant product that passed certification. Yeah seems really non-existant to me for AMD submit something that does not exist.
I would like to read those debunkings about big navi. Who said it and when. Even Lisa Su said you can expect a 4K capable card this year. And i doubt she meant running 4K 30 or 4K low.
That's as official as it gets. We will learn more next month during AMD's Financial briefing.

Codenames are irrelevant and often wrong. So what if Arcturus is not a name they will use? The product (MI100) still exists.

The only reason why we have a 225W RDNA right now is that it's the first generation. I have no doubt AMD will improve power efficiency with RDNA 2.0. Problably lower clocks and fitting a 80 CU card inside 300W limit is not a big deal. Plus IPC increase from architectural improvements.

 

[MDDB]

So if AMD is just catching Tu102 after all this time, what makes you think such a product will be remotely competitive with Am102? That's illogical.

Fortunately, some illogical things happen from time to time. Right, Ryzen? We can hope!

 

[Tomorrow]

Fortunately, some illogical things happen from time to time. Right, Ryzen? We can hope!

We can hope yes. In order to take the nessesary leap AMD would significantly impove both IPC and power efficiency of RDNA 2.0. My fear is that they will improve those but just enough to edge out TU102 and loose to GA/AM102 in a few months time again. The same happened with 1080Ti and 5700 when Nvidia launched Turing.

 

[ARF]

TSMC 12FF process node transistor density: 33.8 MTr/mm^2
TSMC 7nm+ EUV process node transistor density: 115.8 MTr/mn^2

7nm vs 10nm vs 14nm: Fabrication Process - Tech Centurion

Everything about 7nm, 10nm and 14nm Fabrication Process of TSMC, Intel, and Samsung. Transistor Density & Ranking of different technology nodes.

www.techcenturion.com

If you want to quibble over a few % off 4x scaling have at it, but I’m not interested.

Navi 10 is 41 MTr/mm^2.
Vega 20 is 39.97 MTr/mm^2.

So, how on Earth will N7+ EUV be 116 ? ?

Do you, guys, realize that you are arguing over a non-existent product?
Rumors of "Big Navi" have been debunked some time ago, and "Arcturus" is a made-up name that someone came up with almost 2 years ago and everyone still uses it (even though it's not a thing).
Having a 225W RX5700XT is already a good-enough sign that consumer cards are near the upper TDP cap, so the most we could actually hope for is a 48CU Navi card, maybe on 7nm+ process.
Also, the majority of consumer GPU market segments are covered already, and as with Polaris, there's no need to rush for that 0.01% that'll give up their 2080Ti in favor of marginally faster AMD flagship.

It's never nice to comment with false/fake opinion. It's trolling, yes?

that "smallest" chip in action

RX 5700 is 162 W.
RX 5700 XT is 204 W.

RX 5700 XT is pushed well beyond its comfort zone / factory overclock.

 

[ratirt]

We can hope yes. In order to take the nessesary leap AMD would significantly impove both IPC and power efficiency of RDNA 2.0. My fear is that they will improve those but just enough to edge out TU102 and loose to GA/AM102 in a few months time again. The same happened with 1080Ti and 5700 when Nvidia launched Turing.

True that. AMD said they will become competitive in the High-End GPUs. I think AMD would not have thrown these statements in the ether if they have nothing to show for. I'm sure they do realize that new NV graphics might be coming out this year. If they make any adjustments in their calculations for the new AMD High-End GPU time will tell.
I'm just gonna wait and watch what AMD will do and hope for the best.

 

[Platinum certified Husky]

AMD should hurry up with Big navi, so I can buy a cheaper RTX 2080 Super that doesn't have tons of driver issues
Come on AMD, I am counting on you

 

[ARF]

AMD should hurry up with Big navi, so I can buy a cheaper RTX 2080 Super that doesn't have tons of driver issues
Come on AMD, I am counting on you

AMD is rumoured to announce Big Navi at Financial Analyst Day, might detail new RDNA 2 microarcitecture on March 5.
Come March 5.

 

[Platinum certified Husky]

I hope RDNA 2 has less driver issues than current microarchitecture.
Well, one can only hope

 

[kings]

At this point it´s just getting boring. How long have we been waiting for big Navi? 10 months?

The only thing we know is AMD's very vague "promises" and AMD fans hyping it to the roof. It looks like a copy of what happened with Vega, months and months of hype and we all know how it ended!

 

[ShurikN]

I hope RDNA 2 has less driver issues than current microarchitecture.
Well, one can only hope

Why do you care, you're not gonna buy it anyway.

 

[Platinum certified Husky]

Why do you care, you're not gonna buy it anyway.

Because the more competitive AMD is the cheaper RTX 2080 is?
Because I want Nvidia to stop milking the market like Intel?
Because maybe RDNA 2 is so good that I might switch to AMD?

Why do you care, you're not gonna agree with me anyway

 

[Vya Domus]

TSMC 12FF process node transistor density: 33.8 MTr/mm^2

TSMC 7nm+ EUV process node transistor density: 115.8 MTr/mn^2

7nm vs 10nm vs 14nm: Fabrication Process - Tech Centurion

Everything about 7nm, 10nm and 14nm Fabrication Process of TSMC, Intel, and Samsung. Transistor Density & Ranking of different technology nodes.

www.techcenturion.com

If you want to quibble over a few % off 4x scaling have at it, but I’m not interested.

How do you come up with this stuff.

For one thing we were talking about 7nm which brings, by TSMC's, own admission 1.6x times the logic density compared to 10nm :

https://www.tsmc.com/english/dedicatedFoundry/technology/7nm.htm

This is confirmed by products such as Vega 20 which is roughly 70% the size of Vega 10 (that's from 14nm not 10nm so die area scaling is much worse than the transistor density suggests which is normal), the best figure I found for 7nm+ EUV is 20% reduction in area.

1.6 * 1.2 = 1.92, the SMs would still need to be twice as fast to reach the laughable 4x performance scaling estimation. You are not off by a few %, you're in a completely different realm.

Secondly, you should know by now that the transistor density figures these companies put out are just that, transistor density figures that are not representative of real products, they are also not equivalent with scaling factors for chips. If TSMC's 7nm 96.5 MTr/mm^2 transistor density was indicative of the scaling factor that can be achieved that would mean the chips would have to be 2.85x smaller, which is absolutely ridiculous, you should look for area related estimations not transistor density estimations. Use your brain for a moment and some of that objective engineering thing.

Do you, guys, realize that you are arguing over a non-existent product?
Rumors of "Big Navi" have been debunked some time ago, and "Arcturus" is a made-up name that someone came up with almost 2 years ago and everyone still uses it (even though it's not a thing).
Having a 225W RX5700XT is already a good-enough sign that consumer cards are near the upper TDP cap, so the most we could actually hope for is a 48CU Navi card, maybe on 7nm+ process.
Also, the majority of consumer GPU market segments are covered already, and as with Polaris, there's no need to rush for that 0.01% that'll give up their 2080Ti in favor of marginally faster AMD flagship.

I've said this a million times but no one seems to get it. Larger chips end up being more power efficient, RTX 2080 is also near the upper TDP cap yet there is also a RTX 2080ti which is almost 40% bigger.

 

[Tomorrow]

At this point it´s just getting boring. How long have we been waiting for big Navi? 10 months?

The only thing we know is AMD's very vague "promises" and AMD fans hyping it to the roof. It looks like a copy of what happened with Vega, months and months of hype and we all know how it ended!

This launch might be different. Im not sayin it is but several things are different this time: big navi has been on the rumor mill for a shorter time, there is less hype as people are very critical of current AMD drivers and everyone (i hope) knows Nvidia is about to drop their 7nm cards this year. Plus people remember Vega disaster.

Tho there is hope for optimism too. RDNA is a gaming first architecture and a much better one overall. Vega was compute first. Pluss the lying Raja Koduri is with Intel now where by the looks of it he continues to make Vega clones that don't really succeed in any workload and consume a ton of power in the process.

Also some marketing morons have been booted from AMD (Chris Hook for example) that made unnessesarily expensive launch events on aircraft carriers and came up with "Poor Volta" slogan.

 

[silentbogo]

Non-existant product that passed certification. Yeah seems really non-existant to me for AMD submit something that does not exist.

On GN there's a video. Those filings were made by one of AFOX retail partners in Russa, and neither AMD or AFOX (e.g. Foxconn) has anything to do with it.
And it's not a "certification", it's just a product name filing (basically an all-encompassing reservation of names). Sapphire did the same filing w/ EEC even earlier.
The only "solid" piece of evidence is one sentence(I actually do mean "one" as "singular") from Lisa Su Q&A, which is not really comforting.

 

[ratirt]

The only "solid" piece of evidence is one sentence(I actually do mean "one" as "singular") from Lisa Su Q&A, which is not really comforting.

Why not? This is not politics where head of a government would lie his ass off just to get votes. If it has been said it will come then it will. The only unknown is, how well will it do.

 

[Darmok N Jalad]

Do you, guys, realize that you are arguing over a non-existent product?
Rumors of "Big Navi" have been debunked some time ago, and "Arcturus" is a made-up name that someone came up with almost 2 years ago and everyone still uses it (even though it's not a thing).
Having a 225W RX5700XT is already a good-enough sign that consumer cards are near the upper TDP cap, so the most we could actually hope for is a 48CU Navi card, maybe on 7nm+ process.
Also, the majority of consumer GPU market segments are covered already, and as with Polaris, there's no need to rush for that 0.01% that'll give up their 2080Ti in favor of marginally faster AMD flagship.

Didn’t Lisa mention a bigger Navi was coming during the last AMD earnings call?

 

[Tomorrow]

On GN there's a video. Those filings were made by one of AFOX retail partners in Russa, and neither AMD or AFOX (e.g. Foxconn) has anything to do with it.
And it's not a "certification", it's just a product name filing (basically an all-encompassing reservation of names). Sapphire did the same filing w/ EEC even earlier.
The only "solid" piece of evidence is one sentence(I actually do mean "one" as "singular") from Lisa Su Q&A, which is not really comforting.

You're talking about a totally different thing here. Your are talking about the name filings from a retailer in Russia. Yes that was debunked.
This news you are commenting on is about a Korean certification of actual boards. Two in fact. Problably a XT and an non-XT variant. The editor makes a good case for this as a precursor for impending release.

 

[silentbogo]

Didn’t Lisa mention a bigger Navi was coming during the last AMD earnings call?

Just a Q&A/Interview post-CES on AMD's official YT channel. That's the only time Big Navi came out of anyone's mouth at AMD.
Earler - just vague references to Navi Refresh and "faster Navi". No specifics, and definitely no miraculous 80CU flagship-killer.

This news you are commenting on is about a Korean certification of actual boards.

That's a certification for something, not necessarily Big Navi. If I say, it's going to be a low-profile RX5300XT I'd be as right as you.

 

[ARF]

Just a Q&A/Interview post-CES on AMD's official YT channel. That's the only time Big Navi came out of anyone's mouth at AMD.
Earler - just vague references to Navi Refresh and "faster Navi". No specifics, and definitely no miraculous 80CU flagship-killer.

If RTX 2080 Ti 754 mm^2 is to be shrunk on a N7 process, its die would be roughly 377 mm^2.

Any chip larger than 377 mm^2 should be faster than RTX 2080 Ti.

We know Navi 21 is 505 mm^2.

 

[TheoneandonlyMrK]

If RTX 2080 Ti 754 mm^2 is to be shrunk on a N7 process, its die would be roughly 377 mm^2.

Any chip larger than 377 mm^2 should be faster than RTX 2080 Ti.

We know Navi 21 is 505 mm^2.

What? where are these Facts( )coming from ,please display some proof of such nonsense 505 mm^2 for navi 21.

I'm skipping the Nvidia one due to irrelevance to the thread but it's just as dubious.

 

[techguymaxc]

How do you come up with this stuff.

For one thing we were talking about 7nm which brings, by TSMC's, own admission 1.6x times the logic density compared to 10nm :

https://www.tsmc.com/english/dedicatedFoundry/technology/7nm.htm

This is confirmed by products such as Vega 20 which is roughly 70% the size of Vega 10 (that's from 14nm not 10nm so die area scaling is much worse than the transistor density suggests which is normal), the best figure I found for 7nm+ EUV is 20% reduction in area.

1.6 * 1.2 = 1.92, the SMs would still need to be twice as fast to reach the laughable 4x performance scaling estimation. You are not off by a few %, you're in a completely different realm.

Secondly, you should know by now that the transistor density figures these companies put out are just that, transistor density figures that are not representative of real products, they are also not equivalent with scaling factors for chips. If TSMC's 7nm 96.5 MTr/mm^2 transistor density was indicative of the scaling factor that can be achieved that would mean the chips would have to be 2.85x smaller, which is absolutely ridiculous, you should look for area related estimations not transistor density estimations. Use your brain for a moment and some of that objective engineering thing.



I've said this a million times but no one seems to get it. Larger chips end up being more power efficient, RTX 2080 is also near the upper TDP cap yet there is also a RTX 2080ti which is almost 40% bigger.

Why are you comparing 16ff/12ff "14nm" to 10nm and then to 7nm? Nvidia manufactures precisely zero Turing GPUs at either of these (sub-14nm) process nodes.
I cited transistor density comparing 12ff to 7nm+, which are the processes involved.

Or are you saying that Nvidia is going to use the old non-EUV 7nm process node at TSMC, instead of the new EUV 7nm process node which has been in production since Q2 last year?

If so, can you cite any source for this idea?

Also, the reason we use process node transistor density numbers (i.e. ideal scenario) to compare different products is precisely because ASIC design varies so much from generation to generation, let alone from IHV to IHV. Nvidia's GPUs are notoriously cache-heavy compared to their AMD counterparts, and as any engineer knows, SRAM is the very test case used when estimating transistor density on a given process node. Therefore, Nvidia stands to benefit more from a process shrink, than does AMD (in terms of total transistor density across any given ASIC).

Ampere is not yet available in the wild so we don't have any real die size measurements or even estimates to make. Thus, ideal density scaling is the only metric which we can use. Certainly you can admit that TSMC's 7nm+ is denser than 7nm, providing further density scaling which must be factored into any attempt at calculating an estimated transistor density for real silicon. So now we're over 2x scaling from process alone, approaching 3x. You don't think Nvidia can come up with the last ~33% from uarch and clock speed with a new generation design?

 

[ARF]

Like Ryzen, AMD’s Big Navi is “going to similarly disrupt 4K” gaming

AMD is looking to disrupt high-end PC graphics in the same way that its Ryzen processors have disrupted the high-end CPU market. Speaking at a recent AMD RX 5600 XT briefing, Radeon product management lead, Mithun Chandrasekhar, hinted at the red team’s plan for 4K domination once it’s got the mainstream market dealt with.

“With the Radeon 5000-series we are essentially covering 90-something-percent of the total PC gamers today,” says Chandrasekhar. “And so that’s the reason why no 4K right now, it’s because the vast majority of them are at 1440p and 1080p.

Like Ryzen, AMD’s Big Navi is “going to similarly disrupt 4K” gaming

"All of us need a thriving Radeon GPU ecosystem. So, are we going after 4K, and going to similarly disrupt 4K? Absolutely."

www.pcgamesn.com

AMD targets 90% of the market, but the real market penetration is much lower - something around 25-30% only.
The real GPU shipments depends on some factors including technological superiority/advantages, absence-presence of halo products, reputation of the company and others.

AMD has no halo GPU product.

What? where are these Facts( )coming from ,please display some proof of such nonsense 505 mm^2 for navi 21.

I'm skipping the Nvidia one due to irrelevance to the thread but it's just as dubious.

74 Google results for "AMD Navi 21 505 mm", and https://www.techpowerup.com/gpu-specs/amd-navi-21.g923 and https://semiaccurate.com/2019/11/15/a-new-high-end-gpu-just-taped-out/

 

[Vya Domus]

Thus, ideal density scaling is the only metric which we can use. Certainly you can admit that TSMC's 7nm+ is denser than 7nm, providing further density scaling which must be factored into any attempt at calculating an estimated transistor density for real silicon. So now we're over 2x scaling from process alone, approaching 3x. You don't think Nvidia can come up with the last ~33% from uarch and clock speed with a new generation design?

I guess you'll just continue to deny reality with these insane estimations, despite all evidence put forward and real life examples such as Vega 20 prove this sort of scaling is impossible. Hit me up when Nvidia launches this miracle 50 billion transistor TU102 successor on 7nm EUV and prove me wrong, till then you're literally making stuff up with this "approaching 3x scaling" nonsense.

If so, can you cite any source for this idea?

Not before you cite any source that says it will be on TSMC's 7nm EUV. If you do a brief check on what was supposedly leaked you'll find that there is actually no mentioning of TSMC but rather Samsung, you didn't even bother to check that did you.

 

[TheoneandonlyMrK]

Like Ryzen, AMD’s Big Navi is “going to similarly disrupt 4K” gaming

"All of us need a thriving Radeon GPU ecosystem. So, are we going after 4K, and going to similarly disrupt 4K? Absolutely."

www.pcgamesn.com

AMD targets 90% of the market, but the real market penetration is much lower - something around 25-30% only.
The real GPU shipments depends on some factors including technological superiority/advantages, absence-presence of halo products, reputation of the company and others.

AMD has no halo GPU product.






74 Google results for "AMD Navi 21 505 mm", and https://www.techpowerup.com/gpu-specs/amd-navi-21.g923 and https://semiaccurate.com/2019/11/15/a-new-high-end-gpu-just-taped-out/

74 google results point to one rumour started by semiaccurate, even Tpu's GPU database at this point does not make that a fact though it is more substantial,it's not proof of anything but a rumour ,possibly subject to change and certainly not an absolute bet on fact IMHO.

 

[ARF]

74 google results point to one rumour started by semiaccurate, even Tpu's GPU database at this point does not make that a fact though it is more substantial,it's not proof of anything but a rumour ,possibly subject to change and certainly not an absolute bet on fact IMHO.

What kind of change do you expect? A smaller than 505 mm^2 chip is meaningless because there are already 251 mm^2 and 331 mm^2 N7 GPUs.

505 is a small chip compared to the reticle size limit of well over 800 mm^2.

The rumour isn't in any way started by SemiAccurate.
There are other reports from Taiwan and China about the shipped Navi 21 board there.