Friday, March 6th 2020

AMD RDNA2 Graphics Architecture Detailed, Offers +50% Perf-per-Watt over RDNA

With its 7 nm RDNA architecture that debuted in July 2019, AMD achieved a nearly 50% gain in performance/Watt over the previous "Vega" architecture. At its 2020 Financial Analyst Day event, AMD made a big disclosure: that its upcoming RDNA2 architecture will offer a similar 50% performance/Watt jump over RDNA. The new RDNA2 graphics architecture is expected to leverage 7 nm+ (7 nm EUV), which offers up to 18% transistor-density increase over 7 nm DUV, among other process-level improvements. AMD could tap into this to increase price-performance by serving up more compute units at existing price-points, running at higher clock speeds.

AMD has two key design goals with RDNA2 that helps it close the feature-set gap with NVIDIA: real-time ray-tracing, and variable-rate shading, both of which have been standardized by Microsoft under DirectX 12 DXR and VRS APIs. AMD announced that RDNA2 will feature dedicated ray-tracing hardware on die. On the software side, the hardware will leverage industry-standard DXR 1.1 API. The company is supplying RDNA2 to next-generation game console manufacturers such as Sony and Microsoft, so it's highly likely that AMD's approach to standardized ray-tracing will have more takers than NVIDIA's RTX ecosystem that tops up DXR feature-sets with its own RTX feature-set.
AMD GPU Architecture Roadmap RDNA2 RDNA3 AMD RDNA2 Efficiency Roadmap AMD RDNA2 Performance per Watt AMD RDNA2 Raytracing
Variable-rate shading is another key feature that has been missing on AMD GPUs. The feature allows a graphics application to apply different rates of shading detail to different areas of the 3D scene being rendered, to conserve system resources. NVIDIA and Intel already implement VRS tier-1 standardized by Microsoft, and NVIDIA "Turing" goes a step further in supporting even VRS tier-2. AMD didn't detail its VRS tier support.

AMD hopes to deploy RDNA2 on everything from desktop discrete client graphics, to professional graphics for creators, to mobile (notebook/tablet) graphics, and lastly cloud graphics (for cloud-based gaming platforms such as Stadia). Its biggest takers, however, will be the next-generation Xbox and PlayStation game consoles, who will also shepherd game developers toward standardized ray-tracing and VRS implementations.

AMD also briefly touched upon the next-generation RDNA3 graphics architecture without revealing any features. All we know about RDNA3 for now, is that it will leverage a process node more advanced than 7 nm (likely 6 nm or 5 nm, AMD won't say); and that it will come out some time between 2021 and 2022. RDNA2 will extensively power AMD client graphics products over the next 5-6 calendar quarters, at least.
Add your own comment

242 Comments on AMD RDNA2 Graphics Architecture Detailed, Offers +50% Perf-per-Watt over RDNA

#201
Master Tom
R0H1T
Oh in that case AMD should fire the wise guy who made that slide, I mean it makes little sense even now!
What is the problem?
Posted on Reply
#202
Super XP
EarthDog
I'm not at all. Nothing I've said even infers such a thing. Look at my previous posts! I used the same information you did, but others here are talking your points down off a ledge, not mine. I play both sides to middle...my posts support that. Your highly optimistic opinion is the one rooted in rumor and assumption. I have no expectations of this product outside of it being competitive.

It's like groundhog day with you, lol.
My highly optimistic opinions? Not my opinions but from actual tech sites that are quite optimistic for RDNA2. Here's a few, TechPowerUp, Wccftech, PCWorld, Extreme Tech, Guru3D, videocardz, Anandtech, techradar etc.,

What we currently know today and verified by Microsoft is the XBox Series X console specifications. Of course by the time this console is released, things could change articetually to a certain extent, but right now, the performance figures via RDNA2 are quite impressive coming from a APU none discrete GPU ya know? :D :laugh:
Posted on Reply
#203
EarthDog
Super XP
Here's a few, TechPowerUp, Wccftech, PCWorld, Extreme Tech, Guru3D, videocardz, Anandtech, techradar etc.,
Anand - https://www.anandtech.com/show/15591/amds-rdna-2-gets-a-codename-navi-2x-comes-this-year-with-50-improved-perfperwatt
guru3d - https://www.guru3d.com/news-story/amd-talks-rdna2raytracingroadmaps-and-a-possible-teaser-of-big-navi.html
TPU - https://www.techpowerup.com/264538/amd-rdna2-graphics-architecture-detailed-offers-50-perf-per-watt-over-rdna
techradar - https://www.techradar.com/news/amd-big-navi-isnt-coming-until-the-end-of-2020
videocardz - https://videocardz.com/newz/amd-speaks-rdna2-rdna3-zen3-and-zen4-announces-new-roadmaps

Weird how one man's sites that are "quite" optimistic is another man's "they just regurgitated the rumor we've been talking about and nothing more". o_O :rolleyes:

If I missed an article showing this high level of optimism you say some had, please link them. I gave up after those five. :)

EDIT: The best piece of news you have is the Xbox able to run 4K 60 fps....which is easier on a console than it is on a PC...but also puts it on 2080 Super/2080 Ti level.

[ATTACH alt="groundhog-day-bill-murrayjpg-26978fbed7794bca.jpg"]148610[/ATTACH]
Posted on Reply
#204
Super XP
EarthDog
Anand - https://www.anandtech.com/show/15591/amds-rdna-2-gets-a-codename-navi-2x-comes-this-year-with-50-improved-perfperwatt
guru3d - https://www.guru3d.com/news-story/amd-talks-rdna2raytracingroadmaps-and-a-possible-teaser-of-big-navi.html
TPU - https://www.techpowerup.com/264538/amd-rdna2-graphics-architecture-detailed-offers-50-perf-per-watt-over-rdna
techradar - https://www.techradar.com/news/amd-big-navi-isnt-coming-until-the-end-of-2020
videocardz - https://videocardz.com/newz/amd-speaks-rdna2-rdna3-zen3-and-zen4-announces-new-roadmaps

Weird how one man's sites that are "quite" optimistic is another man's "they just regurgitated the rumor we've been talking about and nothing more". o_O :rolleyes:

If I missed an article showing this high level of optimism you say some had, please link them. I gave up after those five. :)

EDIT: The best piece of news you have is the Xbox able to run 4K 60 fps....which is easier on a console than it is on a PC...but also puts it on 2080 Super/2080 Ti level.


Well of course they are reporting from AMD's own Financial Day with the information that was available. Then each and every single site takes that information and applies there opinions and tech expertise on the matter. Microsoft claims 4K60FPS, and no it is not easier to achieve that on a console, its easier to achieve that on a custom built gaming PC. No console that exists today can do 4K60FPS guaranteed. NONE, unless you are aware of some mysterious console which nobody heard of.
Posted on Reply
#205
EarthDog
Super XP
Well of course they are reporting from AMD's own Financial Day with the information that was available. Then each and every single site takes that information and applies there opinions and tech expertise on the matter. Microsoft claims 4K60FPS, and no it is not easier to achieve that on a console, its easier to achieve that on a custom built gaming PC. No console that exists today can do 4K60FPS guaranteed. NONE, unless you are aware of some mysterious console which nobody heard of.
I'm waiting for "then" links you speak of where the sites say they are optimistic. I went and looked at 5 of the sites you mentioned but didnt see it....

Under a similar premise as iphones, when you have one configuration to code for and work with, you can get more out of what hardware you have. So in that respect, it is in fact easier to get 4k60 out of a console. I wasnt talking about hardware specs but being able to do more with the hardware due to having a single configuration like a console. "Whoooosh" as @Vayra86 says! :)

Again, its 4k60 capable and on par with a 2080 super/2080ti. That's a start! I'd love to see what the discrete cards will do. Is it 80-100% like someone else said? Is it between a 2080ti and Ampre's flagship leaning towards 2080ti? Is it as fast as Ampre's flagship? I guess we'll find out in several months. :)
Posted on Reply
#206
Super XP
EarthDog
I'm waiting for "then" links you speak of where the sites say they are optimistic. I went and looked at 5 of the sites you mentioned but didnt see it....

Under a similar premise as iphones, when you have one configuration to code for and work with, you can get more out of what hardware you have. So in that respect, it is in fact easier to get 4k60 out of a console. I wasnt talking about hardware specs but being able to do more with the hardware due to having a single configuration like a console. "Whoooosh" as @Vayra86 says! :)

Again, its 4k60 capable and on par with a 2080 super/2080ti. That's a start! I'd love to see what the discrete cards will do. Is it 80-100% like someone else said? Is it between a 2080ti and Ampre's flagship leaning towards 2080ti? Is it as fast as Ampre's flagship? I guess we'll find out in several months. :)
All we have right now is what AMD has said about RDNA2 and various rumors and speculations. There writing style seems optimistic, not that they actually said we are optimistic. lol,
I do agree on some of your points, :D I think RDNA2 will be faster than the 2080Ti, the question is by how much? SemiAccurate sources say AMD is aiming to compete with Ampere and not nesesarily with the 2000 series. We are probably about 9 to 10 months away from both RDNA2 & Ampere. :) Some paid subscriber at SemiAccurate took a picture of this. If there is one site that I would trust, it's Charlie from SemiAccurate.
[ATTACH alt="1eb01708da20cde7d6ab1b0d61d12f060770c0e3f9a7a5c4ec1737aca32d2463.jpg"]148619[/ATTACH]
Oh and this was back in November 2019 I think, it may no longer apply. Not really sure, but thought I post it.
Posted on Reply
#207
ARF
Super XP
All we have right now is what AMD has said about RDNA2 and various rumors and speculations. There writing style seems optimistic, not that they actually said we are optimistic. lol,
I do agree on some of your points, :D I think RDNA2 will be faster than the 2080Ti, the question is by how much? SemiAccurate sources say AMD is aiming to compete with Ampere and not nesesarily with the 2000 series. We are probably about 9 to 10 months away from both RDNA2 & Ampere. :) Some paid subscriber at SemiAccurate took a picture of this. If there is one site that I would trust, it's Charlie from SemiAccurate.

Oh and this was back in November 2019 I think, it may no longer apply. Not really sure, but thought I post it.
Didn't AMD say "later this year", not "late this year" ?
At Financial Analyst Day, AMD talked about future products and about the so-called RDNA 2 architecture. Information was modest, but we learned that video cards with a big Navi chip would be released later this year.
https://engnews24h.com/amd-has-revealed-new-product-launch-plans-for-big-navi-in-%E2%80%8B%E2%80%8Blate-2020/



https://forums.tomshardware.com/threads/amd-says-its-upcoming-rdna-2-and-navi-2x-will-boost-performance-per-watt-by-50.3579275/
Posted on Reply
#208
Super XP
ARF
Didn't AMD say "later this year", not "late this year" ?
I must have gotten those terms confused. Later this year sounds earlier over late this year which seems like end of year Nov/Dec 2020.
Posted on Reply
#209
Midland Dog
oxrufiioxo
Yeah, I figured this slide would confuse people even though I don't think that was the intention..... They clearly stated 50% more performance per watt in the live stream.
its die codenames guys, navi 10 navi 14 navi 12, navi 20 21 etc
Posted on Reply
#210
ARF
Midland Dog
its die codenames guys, navi 10 navi 14 navi 12, navi 20 21 etc
Navi 12 probably doesn't even exist, or maybe it was but later renamed to Navi 10 while the real Navi 10 got cancelled.

With Navi 2* there should be more chips, so that the whole range from bottom to top gets covered with recent DX12.2 feature level.

For example:

Navi 21 - Radeon RXI 6900 XT and Radeon RXI 6900
Navi 23 - Radeon RXI 6800 XT and Radeon RXI 6800
Navi 24 - Radeon RXI 6500 XT and Radeon RXI 6500

so that
Navi 10 gets rebranded and relegated to the entry - low-end market like Radeon RXI 6300 XT
Navi 14 gets rebranded and relegated to the entry - low-end market like Radeon RXI 6100 XT.

Just wishful thinking but let's hope AMD has finally got some sense in naming its products.

And of course, no more Polaris and Vega, we are already tired of them!
Posted on Reply
#211
Valantar
ARF
Navi 12 probably doesn't even exist, or maybe it was but later renamed to Navi 10 while the real Navi 10 got cancelled.

With Navi 2* there should be more chips, so that the whole range from bottom to top gets covered with recent DX12.2 feature level.

For example:

Navi 21 - Radeon RXI 6900 XT and Radeon RXI 6900
Navi 23 - Radeon RXI 6800 XT and Radeon RXI 6800
Navi 24 - Radeon RXI 6500 XT and Radeon RXI 6500

so that
Navi 10 gets rebranded and relegated to the entry - low-end market like Radeon RXI 6300 XT
Navi 14 gets rebranded and relegated to the entry - low-end market like Radeon RXI 6100 XT.

Just wishful thinking but let's hope AMD has finally got some sense in naming its products.

And of course, no more Polaris and Vega, we are already tired of them!
RXI?

Also there's no way on earth a 250mm2 215W GPU (or even a cut down version) hits x300 naming in its second outing. My money would be on Navi 14 possibly surviving in the low end x300 range, but otherwise I think we'll see a wholesale move to RDNA 2 to ensure feature parity with consoles and to make use of the efficiency gains of the updated arch - though it might be 6+ months from the launch of the high end and upper midrange cards to the rest of the series being filled out.
Posted on Reply
#212
EarthDog
Super XP
We are probably about 9 to 10 months away from both RDNA2 & Ampere. :)
New AMD should be out in late Q3 (4-5 months) with with Nvidia to follow. Initial launches will happen well before the end of the calendar year. 9 months+ puts that into 2021.

Midland Dog
its die codenames guys, navi 10 navi 14 navi 12, navi 20 21 etc
That was corrected 200 posts ago. :)
Posted on Reply
#213
R0H1T
This thread doesn't have 200 pages :wtf:
Posted on Reply
#214
EarthDog
R0H1T
This thread doesn't have 200 pages :wtf:
I said POSTS... o_O :nutkick:

EarthDog
That was corrected 200 posts ago. :)
Posted on Reply
#215
Super XP
EarthDog
New AMD should be out in late Q3 (4-5 months) with with Nvidia to follow. Initial launches will happen well before the end of the calendar year. 9 months+ puts that into 2021.

That was corrected 200 posts ago. :)
I'm factoring in the COVID-19 disaster. Hopefully all works out well, the world goes back to normal and both companies release it's new GPUs before Christmas 2020.
Posted on Reply
#216
ARF
Valantar
RXI?

Also there's no way on earth a 250mm2 215W GPU (or even a cut down version) hits x300 naming in its second outing. My money would be on Navi 14 possibly surviving in the low end x300 range, but otherwise I think we'll see a wholesale move to RDNA 2 to ensure feature parity with consoles and to make use of the efficiency gains of the updated arch - though it might be 6+ months from the launch of the high end and upper midrange cards to the rest of the series being filled out.
XI is 11. X is 10, previous was 9, you know R9, R10, R11.

RDNA 1 doesn't support Variable Rate Shading, Ray-tracing, DX 12 FL 12.2, etc, so yes, normally it should be lower in the product stack.
Posted on Reply
#217
EarthDog
Super XP
I'm factoring in the COVID-19 disaster. Hopefully all works out well, the world goes back to normal and both companies release it's new GPUs before Christmas 2020.
oh... good to know. Consider saying your complete thoughts in a post. You'll have less follow up that way, lol.
Posted on Reply
#218
ARF
Actually, it goes like this:

Radeon X700,
Radeon X800,
Radeon X1800,
Radeon X1900,
Radeon HD 2000,
Radeon HD 3000,
Radeon HD 4000,
Radeon HD 5000,
Radeon HD 6000,
Radeon HD 7000,
(Radeon HD 8000 ?)
Radeon R9 200, (from 7/8000 to 200 ?)
Radeon R9 300,
Radeon RX 400,
Radeon RX 500,
Radeon RX 5000,

and now what? RX 6000 or? RXI 6000?

Or RX 1280?
1 meaning 11, 2 that random number denoting the generation and 80 the performance tier? :laugh:


X0.
X1.
HD2.
HD3.
HD4.
HD5.
HD6.
HD7.
HD8.
R9.2.
R9.3.
RX.4.
RX.5.
RX.50.
Posted on Reply
#219
Valantar
ARF
XI is 11. X is 10, previous was 9, you know R9, R10, R11.

RDNA 1 doesn't support Variable Rate Shading, Ray-tracing, DX 12 FL 12.2, etc, so yes, normally it should be lower in the product stack.
ARF
Actually, it goes like this:

Radeon X700,
Radeon X800,
Radeon X1800,
Radeon X1900,
Radeon HD 2000,
Radeon HD 3000,
Radeon HD 4000,
Radeon HD 5000,
Radeon HD 6000,
Radeon HD 7000,
(Radeon HD 8000 ?)
Radeon R9 200, (from 7/8000 to 200 ?)
Radeon R9 300,
Radeon RX 400,
Radeon RX 500,
Radeon RX 5000,

and now what? RX 6000 or? RXI 6000?
No. The previous series was Rx with x ranging from 3 to 9 to indicate the tier of a GPU within its generation. I.e. R3 was entry level, R5 was lower midrange, R7 was upper midrange, and R9 was high-end. RX does not mean R-ten, but is an entirely new naming scheme that dropped the numbered tiers, seeing how this is all indicated in the model number anyhow (lower number = lower tier). If one GPU is called x300 and one is called x500, it doesn't require a numbered prefix to show that the latter is higher up the performance ladder, after all.

As you bring up a long history of naming it's also worth pointing out that - as your list clearly shows! - there have been several such shifts in naming. The Xxxx series followed after ATI ran out of numbers from their 9xxx series. In that case, X actually did mean 10, but also represented a break in naming that was then abandoned after just a single refresh, with the HD series then taking over. Roman numerals have not been seen here since (probably due to mixing two numbering systems in the same name being a terrible idea). The Rx series then took over as the Radeon HD naming was again nearing running out of numbers, and following the move from VLIW5 to GCN - making a significant change in naming make sense.

Thirdly, when they've used RX (for the entire lineup) for three whole generations across two architectures why would they now suddenly say "oh, this meant ten the whole time, and now we're going to start climbing the ladder of roman numerals"? Sorry, but that isn't happening. Either the new series is RX 6xxx, or they move to something new entirely to indicate that RDNA 2 is something new again.


As for RDNA (1) belonging lower in the product stack due to missing features, this is true, but it makes no sense whatsoever for a relatively large 251mm2 die. That is not a cheap die to produce, and selling an x300 tier GPU above $100 is near impossible. As such, Navi 10 is highly likely to be discontinued (partly to free up 7nm capacity I would guess) and replaced by new chips. Even accounting for the several million dollar cost of taping out a new die it makes very little sense to keep Navi 10 around when RDNA 2 is supposed to be more efficient, clock higher, and comparably dense in terms of CUs while also adding a lot of new features. There's very little chance there won't be a ~250mm2 RDNA 2 die coming (as that's where most sales tend to happen), and keeping two such dice in production at the same time doesn't make sense economically. Navi 14 is small enough that it might be kept around for a while longer as it is thus much cheaper to make and can be sold at sufficiently low prices to make sense in a product tier like that.
Posted on Reply
#220
ARF
Valantar
No. The previous series was Rx with x ranging from 3 to 9 to indicate the tier of a GPU within its generation. I.e. R3 was entry level, R5 was lower midrange, R7 was upper midrange, and R9 was high-end. RX does not mean R-ten, but is an entirely new naming scheme that dropped the numbered tiers, seeing how this is all indicated in the model number anyhow (lower number = lower tier). If one GPU is called x300 and one is called x500, it doesn't require a numbered prefix to show that the latter is higher up the performance ladder, after all.

As you bring up a long history of naming it's also worth pointing out that - as your list clearly shows! - there have been several such shifts in naming. The Xxxx series followed after ATI ran out of numbers from their 9xxx series. In that case, X actually did mean 10, but also represented a break in naming that was then abandoned after just a single refresh, with the HD series then taking over. Roman numerals have not been seen here since (probably due to mixing two numbering systems in the same name being a terrible idea). The Rx series then took over as the Radeon HD naming was again nearing running out of numbers, and following the move from VLIW5 to GCN - making a significant change in naming make sense.

Thirdly, when they've used RX (for the entire lineup) for three whole generations across two architectures why would they now suddenly say "oh, this meant ten the whole time, and now we're going to start climbing the ladder of roman numerals"? Sorry, but that isn't happening. Either the new series is RX 6xxx, or they move to something new entirely to indicate that RDNA 2 is something new again.


As for RDNA (1) belonging lower in the product stack due to missing features, this is true, but it makes no sense whatsoever for a relatively large 251mm2 die. That is not a cheap die to produce, and selling an x300 tier GPU above $100 is near impossible. As such, Navi 10 is highly likely to be discontinued (partly to free up 7nm capacity I would guess) and replaced by new chips. Even accounting for the several million dollar cost of taping out a new die it makes very little sense to keep Navi 10 around when RDNA 2 is supposed to be more efficient, clock higher, and comparably dense in terms of CUs while also adding a lot of new features. There's very little chance there won't be a ~250mm2 RDNA 2 die coming (as that's where most sales tend to happen), and keeping two such dice in production at the same time doesn't make sense economically. Navi 14 is small enough that it might be kept around for a while longer as it is thus much cheaper to make and can be sold at sufficiently low prices to make sense in a product tier like that.
R3, R5, R7 were mobile or integrated graphics, no?
I can't remember any desktop except R7...

Let's just stick with desktop models.

Either RX 6000, RXI 6000 or something like RX 1290/1280.

X0.
X1.
HD2.
HD3.
HD4.
HD5.
HD6.
HD7.
HD8.
R9.2.
R9.3.
RX.4.
RX.5.
RX.50.
Posted on Reply
#221
Super XP
ARF
R3, R5, R7 were mobile or integrated graphics, no?
I can't remember any desktop except R7...

Let's just stick with desktop models.

Either RX 6000, RXI 6000 or something like RX 1290/1280.

X0.
X1.
HD2.
HD3.
HD4.
HD5.
HD6.
HD7.
HD8.
R9.2.
R9.3.
RX.4.
RX.5.
RX.50.
What the heck is a RXI? Where did the 'I' come from? Lol
Posted on Reply
#222
ARF
Super XP
What the heck is a RXI? Where did the 'I' come from? Lol
Why do we hate the "I" ? :D I find it quite sexy!
Posted on Reply
#223
Valantar
ARF
R3, R5, R7 were mobile or integrated graphics, no?
I can't remember any desktop except R7...

Let's just stick with desktop models.

Either RX 6000, RXI 6000 or something like RX 1290/1280.

X0.
X1.
HD2.
HD3.
HD4.
HD5.
HD6.
HD7.
HD8.
R9.2.
R9.3.
RX.4.
RX.5.
RX.50.
No, this is entirely wrong.

Radeon R7 360.
Radeon R7 350.
Radeon R5 230.
Etc.


R3 was AFAIK only used for integrated graphics both on desktop and mobile. Beyond that the lower tiers weren't seen all that much in retail/media coverage simply due to AMD being at a significant economic disadvantage at the time and that most of these GPUs were rebrands/refreshes of previous cards with new names for OEMs to use. Just go to the TPU GPU database and search for "Radeon R[3/5/7/9]" and have a look for yourself. You are only looking at high-tier models, which doesn't show the whole picture of the naming scheme by any means.
Posted on Reply
#224
ARF
Valantar
No, this is entirely wrong.

Radeon R7 360.
Radeon R7 350.
Radeon R5 230.
Etc.


R3 was AFAIK only used for integrated graphics both on desktop and mobile. Beyond that the lower tiers weren't seen all that much in retail/media coverage simply due to AMD being at a significant economic disadvantage at the time and that most of these GPUs were rebrands/refreshes of previous cards with new names for OEMs to use. Just go to the TPU GPU database and search for "Radeon R[3/5/7/9]" and have a look for yourself. You are only looking at high-tier models, which doesn't show the whole picture of the naming scheme by any means.
Ok, now I see that R5 230 was a direct rebrand of HD 8450.

Then AMD went from HD 8000 to R0 200.
Posted on Reply
#225
Valantar
ARF
Why do we hate the "I" ? :D I find it quite sexy!
It looks absolutely terrible, and changes the prefix from a series prefix (i.e. a name) to somehow being a count of something. Either it's counting the generation of the GPU (which the model number already does, i.e. it's redundant and only confusing - why is 11 (XI) the same as 6(xxx)?) or it's counting something else entirely, in which case the question becomes what?

RX is an equivalent of GTX for Nvidia, which is currently simply the prefix to what all their gaming-focused GPUs are named (there is also the entry-level, mobile-only MX series). RTX is then of course an extension of this - a gaming card with Ray tracing support. Or do you think Nvidia's X also stands for 10?
Posted on Reply
Add your own comment