News AMD Financial Analyst Day 2020: A Lot of Details to Digest

[moinmoin]

You may be aware AnandTech posted a lot of articles about the AMD Financial Analyst Day 2020. A lot of details were discussion which I think deserve a dedicated thread. I'll keep updating the OP while going through them. Done

AMD Financial Analyst Day 2020 Round-Up: Laying A Path For Bigger & Better Things

www.anandtech.com

AMD Clarifies Comments on 7nm / 7nm+ for Future Products: EUV Not Specified

www.anandtech.com

AMD no longer references 7nm+, so the connection to TSMC's N7+ is no longer a given.

AMD Unveils CDNA GPU Architecture: A Dedicated GPU Architecture for Data Centers

www.anandtech.com

In place of the current Vega for computing in the server market CDNA will come as counterpart to RDNA for consumers.

AMD Moves From Infinity Fabric to Infinity Architecture: Connecting Everything to Everything

www.anandtech.com

To cover the increasing scope of Infinity Fabric it's being renamed to Infinity Architecture.

AMD Discusses ‘X3D’ Die Stacking and Packaging for Future Products: Hybrid 2.5D and 3D

www.anandtech.com

AMD teased X3D packaging as a successor to the current chiplets MCM packaging in an unspecified future.

AMD Shipped 260 Million Zen Cores by 2020

www.anandtech.com

AMD shipped 260 million Zen cores by 2020. Ian Cutress pixel counted the graph and came up with these numbers: 2017-2018: ~30m cores, 2018-2019: 80m cores (~110m total), 2019-2020: 150m cores (~260m total)

Note of caution: This is cores, not chips. So with up to 64 cores per chip this number looks more impressive than it would otherwise. For comparison, alone through sales of PS4 an Xbox One there are far more than 1.2 billion Jaguar cores in the wild. So we can expect the numbers of Zen cores to vastly increase further once the new Zen 2 based consoles launch later this year.

AMD's 2020-2022 Client GPU Roadmap: RDNA 3 & Navi 3X On the Horizon With More Perf & Efficiency

www.anandtech.com

AMD's RDNA 2 Gets A Codename: “Navi 2X” Comes This Year With 50% Improved Perf-Per-Watt

www.anandtech.com

RDNA 2 is called Navi 2X, further improves perf per watt and includes hardware ray tracing.

Updated AMD Ryzen and EPYC CPU Roadmaps March 2020: Milan, Genoa, and Vermeer

www.anandtech.com

Zen 3 based Milan coming in ‘late 2020’, Zen 4 based is coming out by 2022.So the cadence is more to the 15 months than 12 months side. Zen 3 will hit the consumer market ‘later this year’.

 

[soresu]

In place of the current Vega for computing in the server market CDNA will come as counterpart to RDNA for consumers.

There will definitely still be professional/workstation SKU's for RDNA due to the lack of rasterization logic in CDNA - too many hardcore graphics applications in that space to ignore the potential revenue.

It also seems likely that many pro applications will end up making use of the RT acceleration hardware in RDNA2 forwards, somehow I doubt that will be in CDNA.

 

[moinmoin]

There will definitely still be professional/workstation SKU's for RDNA due to the lack of rasterization logic in CDNA - too many hardcore graphics applications in that space to ignore the potential revenue.

It also seems likely that many pro applications will end up making use of the RT acceleration hardware in RDNA2 forwards, somehow I doubt that will be in CDNA.

I counted the professional/workstation market as part of the consumer one, as opposite to the headless server market. Graphics computing market and headless computing market may be more apt names for how the TAM is split for "G"PUs.

 

[moinmoin]

Some more stuff from the AMD Financial Analyst Day 2020:

• Impact by Covid-19 within guidance so doesn't need update, supply chain is at near normal.
• Growth for the year 2019 ~16%, expected 28-30% for 2020, planned to be ~20% in the long term.
• Gross leverage ratio (essentially ratio between debts and free cash flow/liquid assets) is at 0.5x (down from 1.9x in 2018 and 10x in 2016).
• Gross margin ~45% in 2020, up from %43 in 2019, planned to reach >50%.
• Data center becomes 30% of revenue, up from 15%, approximately, now.

Financial statistics, models and summary:

Reddit - Dive into anything

www.reddit.com

Good summary of AMD's status quo in the server market and what's to come:

AMD Is Determined To Get Its Rightful Datacenter Share

There is an equally virtuous and vicious cycle that propels all computing: Innovation requires competition to propel it, and competition requires

www.nextplatform.com

CDNA - a pure ML play leveraging AMD's expertise in semi-custom designs.
https://www.reddit.com/r/AMD_Stock/comments/fegkqa/_/fjo8rq4

 

[DrMrLordX]

Have to say I'm disappointed that they've pushed Genoa back to 2022. That's probably the date for general availability, so ODM shipments will likely begin in late 2021, but still . . . that's a lengthened cadence, and over the long term, that will be bad for the company.

 

[moinmoin]

Have to say I'm disappointed that they've pushed Genoa back to 2022. That's probably the date for general availability, so ODM shipments will likely begin in late 2021, but still . . . that's a lengthened cadence, and over the long term, that will be bad for the company.

Is it a lengthened cadence? It still looks within the 15 months cadence. As Ian Cutress put it:

Zen 4 based Genoa has already been announced as the CPU to power the El Capitan supercomputer, and in this roadmap AMD has put it as coming out by 2022. We asked AMD for clarification, and they stated that in this sort of graph, we should interpret it as the full stack of Genoa should be formally launched by the end of 2022. Given AMD’s recent 12-15 month cadence with the generations of EPYC, and the expected launch of Milan late this year, we would expect to see Genoa in early 2022.

An exactly 12 months yearly cadence would certainly be nicer (cleaner) to have psychologically but aside consumer facing holiday business I fail to see other down sides to it.

 

[jpiniero]

Have to say I'm disappointed that they've pushed Genoa back to 2022. That's probably the date for general availability, so ODM shipments will likely begin in late 2021, but still . . . that's a lengthened cadence, and over the long term, that will be bad for the company.

It's only really a problem if Intel delivers Granite Rapids in 2022.

 

[DrMrLordX]

Is it a lengthened cadence?

Basically yes. Zen2 consumer (Matisse) was an aberration, and there's no sign that Rome was affected (it really wasn't). Rome was pretty-much on-track, and they'll mostly be on-track up through Zen3, though if Milan lags too much it'll be a bad sign. Early 2022 for Genoa is ~6 months off from where they should be, and that's a long-enough period of time to permit a counterattack.

Please consider that buying decisions for major installations of servers has a lot to do with trust and psychology. There are also some critical milestones for their competitor that hinge on 2022 and the success of a certain 7nm EUV node.

It's only really a problem if Intel delivers Granite Rapids in 2022.

That's one of AMD's potential problems. AMD needed to face off against Granite Rapids with Zen5, not Zen4.

 

[yeshua]

So, what's there to digest except the already known roadmaps and vague promises? A new compute GPU architecture? Well, it's great except ... we know nothing about it other than its name. Also, given AMD's track record of overpromising and underdelivering I'd take their promises, e.g. RDNA 2.0 is 50% more power efficient than RDNA 1.0, with a lot of grains of salt.

Another issue is that I am not a fan of how they name their consumer GPU architectures: RNDA, RDNA 2.0 and RDNA 3.0. These are big numbers which usually represent truly massive changes. On the other hand earlier AMD has been talking a lot about how RDNA 2.0 is just a refinement of RDNA (RTRX + VRS). Has something suddenly changed?

Last but not least I expected some hard figures in terms of Zen 3 improvements in regard to Zen 2 (IPC and maximum operating frequency). Nothing was revealed which is a complete bummer. In short the presentation was quite hollow in terms of raw good data. Hopefully investors were satisfied.

 

[jpiniero]

That's one of AMD's potential problems. AMD needed to face off against Granite Rapids with Zen5, not Zen4.

We don't know how realistic it is for Intel to be able to deliver Granite Rapids in 2022.

 

[Markfw]

So, what's there to digest except the already known roadmaps and vague promises? A new compute GPU architecture? Well, it's great except ... we know nothing about it other than its name. Also, given AMD's track record of overpromising and underdelivering I'd take their promises, e.g. RDNA 2.0 is 50% more power efficient than RDNA 1.0, with a lot of grains of salt.

Another issue is that I am not a fan of how they name their consumer GPU architectures: RNDA, RDNA 2.0 and RDNA 3.0. These are big numbers which usually represent truly massive changes. On the other hand earlier AMD has been talking a lot about how RDNA 2.0 is just a refinement of RDNA (RTRX + VRS). Has something suddenly changed?

Last but not least I expected some hard figures in terms of Zen 3 improvements in regard to Zen 2 (IPC and maximum operating frequency). Nothing was revealed which is a complete bummer. In short the presentation was quite hollow in terms of raw good data. Hopefully investors were satisfied.

overpromising and underdelivering ? You are totally wrong there. At least with Zen. They have under promised and over delivered. If you think I have it wrong, give me one example in the last 2 years.

Intel is the one that has been bad as of late. You got it backwards.

 

[tamz_msc]

Last but not least I expected some hard figures in terms of Zen 3 improvements in regard to Zen 2 (IPC and maximum operating frequency). Nothing was revealed which is a complete bummer. In short the presentation was quite hollow in terms of raw good data. Hopefully investors were satisfied.

You won't get hard performance numbers from an event like this. You'll need to wait till an architecture day or some other event closer to release.

 

[exquisitechar]

So, what's there to digest except the already known roadmaps and vague promises? A new compute GPU architecture? Well, it's great except ... we know nothing about it other than its name. Also, given AMD's track record of overpromising and underdelivering I'd take their promises, e.g. RDNA 2.0 is 50% more power efficient than RDNA 1.0, with a lot of grains of salt.

Another issue is that I am not a fan of how they name their consumer GPU architectures: RNDA, RDNA 2.0 and RDNA 3.0. These are big numbers which usually represent truly massive changes. On the other hand earlier AMD has been talking a lot about how RDNA 2.0 is just a refinement of RDNA (RTRX + VRS). Has something suddenly changed?

+50% perf/watt is a massive change to me. And I don’t see why it has to be a massive change in the first place. It’s just a nice and simple naming scheme like they have with Zen, tells you what’s the successor/predecessor of what at a glance.

The people responsible for over promising and under delivering for the graphics group are long gone, Zen people have turned that place around. You can expect RDNA2 to bring a similar perf/watt uplift as RDNA did over Vega, which AMD delivered on.

 

[yeshua]

overpromising and underdelivering ? You are totally wrong there. At least with Zen. They have under promised and over delivered. If you think I have it wrong, give me one example in the last 2 years.

Intel is the one that has been bad as of late. You got it backwards.

I was strictly implying their GPU division. Too many launches over the past ten years have left people with a sour taste. Zen has been exceptional so far - there's no doubt about that. They basically reinvented themselves and made Intel sweat.

 

[Olikan]

I was strictly implying their GPU division. Too many launches over the past ten years have left people with a sour taste. Zen has been exceptional so far - there's no doubt about that. They basically reinvented themselves and made Intel sweat.

While you do have a point, since Raja and Hook left, Amd didn't overhyped/lied about Radeon 7 and Navi...

 

[yeshua]

While you do have a point, since Raja and Hook left, Amd didn't overhyped/lied about Radeon 7 and Navi...

Yeah, but some AMD fans expected 5700XT to be a 2800Ti killer ;-) Didn't quite pan out.

 

[Markfw]

Yeah, but some AMD fans expected 5700XT to be a 2800Ti killer ;-) Didn't quite pan out.

So a couple of people have high expectations, and you blame that on AMD ?

 

[Hitman928]

So a couple of people have high expectations, and you blame that on AMD ?

Yeah, if we're going to have fanboy expectations as the standard here, then Nvidia massively overhyped and underdelivered with RTX and has done the same with pricing for generations. Intel has been overhyping and underdelivering for generations of CPUs as well. It's a pretty weird standard to hold though.

 

[eek2121]

You may be aware AnandTech posted a lot of articles about the AMD Financial Analyst Day 2020. A lot of details were discussion which I think deserve a dedicated thread. I'll keep updating the OP while going through them. Done

AMD Financial Analyst Day 2020 Round-Up: Laying A Path For Bigger & Better Things

www.anandtech.com

AMD Clarifies Comments on 7nm / 7nm+ for Future Products: EUV Not Specified

www.anandtech.com

AMD no longer references 7nm+, so the connection to TSMC's N7+ is no longer a given.

AMD Unveils CDNA GPU Architecture: A Dedicated GPU Architecture for Data Centers

www.anandtech.com

In place of the current Vega for computing in the server market CDNA will come as counterpart to RDNA for consumers.

AMD Moves From Infinity Fabric to Infinity Architecture: Connecting Everything to Everything

www.anandtech.com

To cover the increasing scope of Infinity Fabric it's being renamed to Infinity Architecture.

AMD Discusses ‘X3D’ Die Stacking and Packaging for Future Products: Hybrid 2.5D and 3D

www.anandtech.com

AMD teased X3D packaging as a successor to the current chiplets MCM packaging in an unspecified future.

AMD Shipped 260 Million Zen Cores by 2020

www.anandtech.com

AMD shipped 260 million Zen cores by 2020. Ian Cutress pixel counted the graph and came up with these numbers: 2017-2018: ~30m cores, 2018-2019: 80m cores (~110m total), 2019-2020: 150m cores (~260m total)

Note of caution: This is cores, not chips. So with up to 64 cores per chip this number looks more impressive than it would otherwise. For comparison, alone through sales of PS4 an Xbox One there are far more than 1.2 billion Jaguar cores in the wild. So we can expect the numbers of Zen cores to vastly increase further once the new Zen 2 based consoles launch later this year.

AMD's 2020-2022 Client GPU Roadmap: RDNA 3 & Navi 3X On the Horizon With More Perf & Efficiency

www.anandtech.com

AMD's RDNA 2 Gets A Codename: “Navi 2X” Comes This Year With 50% Improved Perf-Per-Watt

www.anandtech.com

RDNA 2 is called Navi 2X, further improves perf per watt and includes hardware ray tracing.

Updated AMD Ryzen and EPYC CPU Roadmaps March 2020: Milan, Genoa, and Vermeer

www.anandtech.com

Zen 3 based Milan coming in ‘late 2020’, Zen 4 based is coming out by 2022.So the cadence is more to the 15 months than 12 months side. Zen 3 will hit the consumer market ‘later this year’.

I just wanted to add here that AMD has not stated desktop Ryzen chips are shipping at the end of 2020. Desktop Ryzen chips are shipping “later this year”. Only EPYC is shipping EOY, which makes sense considering the timing for Rome.

 

[RetroZombie]

Yeah, but some AMD fans expected 5700XT to be a 2800Ti killer ;-) Didn't quite pan out.

I'm a low expectations guy, so it was an huge surprise it could keep up with the RTX 2080 (non super).

 

[RetroZombie]

e.g. RDNA 2.0 is 50% more power efficient than RDNA 1.0, with a lot of grains of salt.

Why so, it's well known that almost all their cards are pushed to the limit. When they stop doing it, let's assume it will be with RDNA2 that's some pretty easy to achieve 50%.

You already have recently one example of what it could be one efficient card from them the RX5600XT (that ended up changed last minute), and in the past the fury nano is also a good example of efficient.

Their apus are also great examples of efficiency from them, much more performant and efficient than nvidia.

 

[Tabalan]

Why so, it's well known that almost all their cards are pushed to the limit. When they stop doing it, let's assume it will be with RDNA2 that's some pretty easy to achieve 50%.

You already have recently one example of what it could be one efficient card from them the RX5600XT (that ended up changed last minute), and it the past the fury nano is also a good example of efficient.

Their apus are also great examples of efficiency from them, much more performant and efficient than nvidia.

But to do so you have to create bigger (more expensive) die and clock it lower. It's higher cost for AMD, which wants to improve their profit margin.

 

[RetroZombie]

The anandtech article have some amd slide that say's:
The Next Five Years
OUR TECNOLOGY INVESTMENTS

However didn't see more than two years being disclosed did i miss something?
Or they are adding the previous three years?

 

[RetroZombie]

But to do so you have to create bigger (more expensive) die and clock it lower. It's higher cost for AMD, which wants to improve their profit margin.

I don't disagree, but they also seem to have some voltage boost problem, it is ramping up to higher values than the ones needed.

They might have fixed that judging by renoir, which can get high clocks and still be power efficient.

 

[maddie]

I don't disagree, but they also seem to have some voltage boost problem, it is ramping up to higher values than the ones needed.

They might have fixed that judging by renoir, which can get high clocks and still be power efficient.

I've felt for a long time in looking at Nvidia's voltage frequency curves vs AMD ones, that most of Nvidia's superiority in perf/watt came from the ability to run at lower voltages. Whenever I did the rough calcs , I got around 75% of the improvement over AMD due to this alone.

Not saying this is easy to do, but with the CPU people there for some time, these techniques appear to begin showing up in the GPU circuitry.