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

#76
Super XP
medi01
2080Ti is about 46%/55% faster than 5700XT (ref vs ref) at 1440p/4k respectively in TPU benchmarks.
The rumored Big Navi based on RDNA2 should be about 30% to 40% faster over the current 2080Ti. (That's based on a Navi prototype) It's in direct competition over the upcoming RTX 3080 series. That's according to various sources and Geekbench. RDNA2 is so efficient, AMD has the ability to increase core clocks all while still maintaining a 250w power envelope. I'm sure they can achieve even more performance at 300w if required.
Posted on Reply
#77
efikkan
It would be wise to keep in mind that even if some of the lower clocked models achieve close to AMD's efficiency targets, this doesn't mean the entire lineup will achieve quite the same level of efficiency. These are best case scenarios to please investors, and deserves huge asterisks behind them.
Posted on Reply
#79
Valantar
sergionography
Yes that's true, but 2x performance is very likely non the less. Keep in mind that Navi 10/5700xt is a small 250mm2 chip. 50% performance per watt means AMD can scale up more shaders before running into a performance/power wall. If anything, this gives credit to big Navi being twice the size of navi10, a 500+mm2 chip with double the shaders. A 5120 Radeon core chip below the 300watt pci limit all for sudden becomes a possibility
Congratulations, you just made a textbook straw man argument. I never said there wouldn't be an RDNA 2 GPU 2x as fast as the 5700XT, I said that expecting 2x perf/w based on a slide naming a series of chips "Navi 2X" is stupid.
I also said this a few posts later:
Valantar
The main reason for perf/w improvements is to be able to cool a bigger/higher performing die in a PCIe form factor.
So, I don't know who it is you are arguing against, but it certainly isn't me. What you are saying bears no relation to the post you quoted when it's read in its proper context. It was commenting on something that related to an architecture and a series of chips (RDNA 1 vs 2 and Navi 1X vs 2X), not a specific chip, so talking absolute performance numbers (such as 2x 5700 XT) is meaningless in that context. AMD has said that they will be competing in the flagship space this generation, so at least close to 2x 5700XT is quite likely. But even then using such a card to say "RDNA 2 is 2x as fast as RDNA 1" would be stupid as you'd be comparing cards in different price ranges and power envelopes.
Posted on Reply
#80
sergionography
Valantar
Congratulations, you just made a textbook straw man argument. I never said there wouldn't be an RDNA 2 GPU 2x as fast as the 5700XT, I said that expecting 2x perf/w based on a slide naming a series of chips "Navi 2X" is stupid.
I also said this a few posts later:

So, I don't know who it is you are arguing against, but it certainly isn't me. What you are saying bears no relation to the post you quoted when it's read in its proper context. It was commenting on something that related to an architecture and a series of chips (RDNA 1 vs 2 and Navi 1X vs 2X), not a specific chip, so talking absolute performance numbers (such as 2x 5700 XT) is meaningless in that context. AMD has said that they will be competing in the flagship space this generation, so at least close to 2x 5700XT is quite likely. But even then using such a card to say "RDNA 2 is 2x as fast as RDNA 1" would be stupid as you'd be comparing cards in different price ranges and power envelopes.
My apologies I did not intend for my post to be "against" anybody, especially yourself. I was rather agreeing with you and adding perspective. I also agree that 2x might simply mean second gen, however it is a curious naming scheme that I don't remember from AMD before so it doesn't hurt to speculate. Last time such monikers were used they did so for dual chip cards. If we speculate based on this assumption then Navi 2X offers twice the performance of the first Navi/rx 5700xt, and NAVI 3x is triple the power. The interesting thing I noticed just now when I went back to the slides was that the architecture is RDNA 1, RDNA 2, and RDNA 3. But when specifically talking about the details of RDNA 2 they mention NAVI 2X. And we know Navi as the codename for the chips rather than the architecture. So when they describe RDNA 2 as NAVI 2X, along with the rumors we keep hearing about "big Navi", it all tends to be misleading in all sorts of way to indicate twice the top end performance.
Posted on Reply
#81
Super XP
sergionography
My apologies I did not intend for my post to be "against" anybody, especially yourself. I was rather agreeing with you and adding perspective. I also agree that 2x might simply mean second gen, however it is a curious naming scheme that I don't remember from AMD before so it doesn't hurt to speculate. Last time such monikers were used they did so for dual chip cards. If we speculate based on this assumption then Navi 2X offers twice the performance of the first Navi/rx 5700xt, and NAVI 3x is triple the power. The interesting thing I noticed just now when I went back to the slides was that the architecture is RDNA 1, RDNA 2, and RDNA 3. But when specifically talking about the details of RDNA 2 they mention NAVI 2X. And we know Navi as the codename for the chips rather than the architecture. So when they describe RDNA 2 as NAVI 2X, along with the rumors we keep hearing about "big Navi", it all tends to be misleading in all sorts of way to indicate twice the top end performance.
I have a feeling AMD called it NAVI 2x on purpose to spark speculative debate. Which means that they have a product that they are very confident about and could be a potential market disruptor.

I mean they could have easily named it NAVI 2 and NAVI 3. But they choose the "X" for a reason IMO.

For me anyway I never thought that 2X or 3X meant performance increase over 1X, though I can understand why some might read it that way. My speculation, RDNA2 is going to be more than 2x the performance of RDNA1.
Posted on Reply
#82
Valantar
sergionography
My apologies I did not intend for my post to be "against" anybody, especially yourself. I was rather agreeing with you and adding perspective. I also agree that 2x might simply mean second gen, however it is a curious naming scheme that I don't remember from AMD before so it doesn't hurt to speculate. Last time such monikers were used they did so for dual chip cards. If we speculate based on this assumption then Navi 2X offers twice the performance of the first Navi/rx 5700xt, and NAVI 3x is triple the power. The interesting thing I noticed just now when I went back to the slides was that the architecture is RDNA 1, RDNA 2, and RDNA 3. But when specifically talking about the details of RDNA 2 they mention NAVI 2X. And we know Navi as the codename for the chips rather than the architecture. So when they describe RDNA 2 as NAVI 2X, along with the rumors we keep hearing about "big Navi", it all tends to be misleading in all sorts of way to indicate twice the top end performance.
No problem, thanks for clearing that up :) You might have a point, though I'm still lening towards the simplest solution of X=second digit in the chip's code name, i.e. Navi 1X = Navi 10, Navi 14, etc., and Navi 2X = Navi 20, 21, 22 and so on. The reason we haven't seen this before is that these code names are typically not used publicly, at least not in this manner (they may be part of the specifications of a card, but I've never seen a range of code names used to denominate a generation of chips publicly like this).
Super XP
I have a feeling AMD called it NAVI 2x on purpose to spark speculative debate. Which means that they have a product that they are very confident about and could be a potential market disruptor.

I mean they could have easily named it NAVI 2 and NAVI 3. But they choose the "X" for a reason IMO.

For me anyway I never thought that 2X or 3X meant performance increase over 1X, though I can understand why some might read it that way. My speculation, RDNA2 is going to be more than 2x the performance of RDNA1.
Interesting theory. Might be true, or it might be that whoever made the slides didn't fully think this through. Either way I'm looking forward to... hopefully Computex?
Posted on Reply
#83
ARF
Radeon RX 5700 XT (Navi 10) = 219 W (average gaming consumption) = 100% performance (3840x2160)
GeForce RTX 2080 Ti = 273 W (average gaming consumption) = 156% performance (3840x2160)

50% better performance per watt in Navi 2* will mean 150% performance in the same 219 W as Navi 10.

If we assume that Navi 10 is memory bandwidth starved (only around 448 GB/s) and is overvolted at stock, then we could add additional 10-20% performance in considerably lower stock power consumption, for instance 160-170% performance in 180 W (average gaming consumption).

If Navi 21's average gaming consumption is 280 W and its performance scales linearly, then it should show around 55% higher consumption or 215-225% the performance of Navi 10.

So, around 45% higher performance than RTX 2080 Ti at the same consumption.

If, however, AMD decides to push the TDP further to 350 W, then the relative performance would be 250-260%.
Posted on Reply
#84
sergionography
Valantar
No problem, thanks for clearing that up :) You might have a point, though I'm still lening towards the simplest solution of X=second digit in the chip's code name, i.e. Navi 1X = Navi 10, Navi 14, etc., and Navi 2X = Navi 20, 21, 22 and so on. The reason we haven't seen this before is that these code names are typically not used publicly, at least not in this manner (they may be part of the specifications of a card, but I've never seen a range of code names used to denominate a generation of chips publicly like this).

Interesting theory. Might be true, or it might be that whoever made the slides didn't fully think this through. Either way I'm looking forward to... hopefully Computex?
Perhaps I missed your later posts or something but I think your right now that I think about it. That answers the X being referenced to the Navi name and not RDNA. And another argument to support this is that it makes no sense for AMD to publish vague performance expectations this far in advance anyways nor is it smart. And if they intended to mean performance then it actually would be X2 rather than 2X
Posted on Reply
#85
ARF
I can't know what the reason for this launch delay is, though.
They have already got working cards and as per reports are testing them right now.

Maybe a new revision to try to improve it even further?
Posted on Reply
#86
efikkan
sergionography
I also agree that 2x might simply mean second gen, however it is a curious naming scheme that I don't remember from AMD before so it doesn't hurt to speculate. Last time such monikers were used they did so for dual chip cards. If we speculate based on this assumption then Navi 2X offers twice the performance of the first Navi/rx 5700xt, and NAVI 3x is triple the power. The interesting thing I noticed just now when I went back to the slides was that the architecture is RDNA 1, RDNA 2, and RDNA 3. But when specifically talking about the details of RDNA 2 they mention NAVI 2X. And we know Navi as the codename for the chips rather than the architecture. So when they describe RDNA 2 as NAVI 2X, along with the rumors we keep hearing about "big Navi", it all tends to be misleading in all sorts of way to indicate twice the top end performance.
While it might be understandable that not everyone in this thread understood the Navi terminology, but those who have been deeply engaged in the discussions for a while should have gotten that Navi 1x is Navi 10/12/14 and Navi 2x is Navi 21/22/23*, we have known this for about a year or so. Even more astounding, I noticed several of those so-called "experts" on YouTube that some of you like to cite for analysis and leaks, who can ramble on about Navi for hours, still managed to fail to know this basic information about Navi. It just goes to show how little these nobodies on YouTube actually know.

*) I only know about Navi 21/22/23 so far.

ARF
I can't know what the reason for this launch delay is, though.
They have already got working cards and as per reports are testing them right now.

Maybe a new revision to try to improve it even further?
Which delay in particular are you thinking of?
Posted on Reply
#87
Valantar
ARF
I can't know what the reason for this launch delay is, though.
They have already got working cards and as per reports are testing them right now.

Maybe a new revision to try to improve it even further?
Delay? Sorry, but what delay? Has there been a launch date announced that I somehow missed? Most vendors have "working cards" half a year or more before launch for testing purposes, based off early silicon and on unfinished/massively overspecced engineering boards. This doesn't tell us anything about mass production or designs being finished. Silicon mass production can start as early as about half a year before retail launch too, as a wafer made on an advanced node can take a month or more to process, and it then needs to be packaged, tested, binned, mounted to a board, tested again, packaged for sales/shipping, shipped (3 weeks or more on a boat typically) - that's easily half a year in that process. For an AIB/non-reference design, add another month for additional design, binning and validation. So if mass production of these GPUs started today, they'd likely be up for a late Q3 launch. Tl;dr: there being working engineering sample cards in circulation has relatively little bearing on retail availability.
Posted on Reply
#88
ARF
efikkan
Which delay in particular are you thinking of?
Valantar
Delay? Sorry, but what delay? Has there been a launch date announced that I somehow missed?
2017, 2018, 2019 and now one quarter of 2020 is gone and still the best that AMD has got is a mid-range RX 5700 XT.

Where is the performance, high-end, enthusiast card?


Isn't this a delay of years?
Posted on Reply
#89
efikkan
In that sense, sure. Navi 1x was supposed to launch early 2018.
But I don't know if there ever was a "big Navi" for Navi 1x, if so it was scrapped long before tapeout.
Posted on Reply
#90
TKnockers
All I know is that 5700xt was the last amd gpu I bought... they can promise and create anything they like, won't get my money again.
Posted on Reply
#91
Valantar
ARF
2017, 2018, 2019 and now one quarter of 2020 is gone and still the best that AMD has got is a mid-range RX 5700 XT.

Where is the performance, high-end, enthusiast card?


Isn't this a delay of years?
I agree that it's about damn time AMD gets back into the high end GPU game, but "delay" is the entirely wrong word. A delay implies that something has been promised at a certain time and then has not appeared at that time, which isn't the case here - they simply haven't competed at all in that segment. Absence, failure to compete, sure, but not a delay.
TKnockers
All I know is that 5700xt was the last amd gpu I bought... they can promise and create anything they like, won't get my money again.
Care to expand on that?
Posted on Reply
#92
ARF
Navi 10's VCN 2.0 doesn't even support 8K video playback. People complain:
RX 5700 XT Nitro+ Special Edition here
Chrome: laggy - GPU load at 100%
Firefox: super extremely laggy - GPU load spiking like crazy from 0 to 90%
Edge: smooth - GPU load at 100%
Tested with this video (8k@60fps):

Amd/comments/cy9h72

Valantar
I agree that it's about damn time AMD gets back into the high end GPU game, but "delay" is the entirely wrong word. A delay implies that something has been promised at a certain time and then has not appeared at that time, which isn't the case here - they simply haven't competed at all in that segment. Absence, failure to compete, sure, but not a delay.
There is definitely a delay in Navi's launch. If it was originally supposed to launch in H1 2018, it was pushed back to H2 2019.
It's particularly interesting to hear Navi's story, was it originally intended for N14 node and then moved forward to N7, was simply N7 too late........




TKnockers
All I know is that 5700xt was the last amd gpu I bought... they can promise and create anything they like, won't get my money again.
The problem is that there is no alternative of AMD.
They offer some features in their graphics that no one else can or will because their IP competence is lower.

But AMD is way too late to implement 4K gaming for the masses, way too late to introduce ray-tracing, way too late to even compete in some segments of the market.

To be honest, I would be happier if Nvidia goes only for the low-end and mid-range markets, while AMD competes only with itself at the top high-end tier.
Posted on Reply
#93
efikkan
I don't believe AMD "promised" a "high-end" Navi until fairly recently.
I believe it was early last year that Lisa Su said something along the lines of "Vega level performance" for Navi (1x). So I believe it's the hype and expectations to blame here.
Posted on Reply
#94
Super XP
TKnockers
All I know is that 5700xt was the last amd gpu I bought... they can promise and create anything they like, won't get my money again.
Your statement makes no sense. Navi was released Q3 2019. What do you expect AMD to do? Keep launching more quarter after quarter? Not even Nvidia does such a thing. Navi is in a rotation. So expect a 5700XT upgrade by Q3 2020 exactly what the RDNA 2 roadmaps have stated since before 2019.

ARF
Navi 10's VCN 2.0 doesn't even support 8K video playback. People complain:


Amd/comments/cy9h72



There is definitely a delay in Navi's launch. If it was originally supposed to launch in H1 2018, it was pushed back to H2 2019.
It's particularly interesting to hear Navi's story, was it originally intended for N14 node and then moved forward to N7, was simply N7 too late........






The problem is that there is no alternative of AMD.
They offer some features in their graphics that no one else can or will because their IP competence is lower.

But AMD is way too late to implement 4K gaming for the masses, way too late to introduce ray-tracing, way too late to even compete in some segments of the market.

To be honest, I would be happier if Nvidia goes only for the low-end and mid-range markets, while AMD competes only with itself at the top high-end tier.
You are driving into a totally off topic situation. What people seem to not understand, in 2011 AMD took a chance on Bulldozer and it cost them. Almost all resources went into developing ZEN. 6 years later, 2017 ZEN launches and took the CPU world by storm. Only then did AMD put more resources into the RTG. While they were releasing what Radeon they can, they were quietly developing RDNA2 which would also stem to next generation gaming consoles. Microsoft had a hand in RDNA2. This spells good news for PC Gamers.
Not sure if there was any Sony involvement though.

Too late for Ray Tracing? Currently Ray Tracing is useless. 4K gaming? Radeon 7 takes care of that for those that didn't want to buy Nvidia. The meat and potatoes for the RTG is RDNA2, that's Rumor'd to be a market disrupter. And yes we ALL can't wait for stiffer competition.
Posted on Reply
#95
Valantar
ARF
Navi 10's VCN 2.0 doesn't even support 8K video playback. People complain:
How on earth is this relevant to this topic whatsoever? At this point you're just listing off things you don't like about RDNA (and seemingly by extension AMD). That is not what this thread is about. If you want a thread about that, go create one.

ARF
There is definitely a delay in Navi's launch. If it was originally supposed to launch in H1 2018, it was pushed back to H2 2019.
It's particularly interesting to hear Navi's story, was it originally intended for N14 node and then moved forward to N7, was simply N7 too late........

You have yourself gone to pains to underscore that RDNA 1 and RDNA 2 (Navi 1X and Navi 2X) are not the same thing. In this very thread, no less. So quit moving the goal posts please. When you first mentioned a delay it was very clear by your wording ("this delay" etc.) that you were saying that RDNA 2 was delayed - which is what you were asked to clarify. You haven't so far. Nobody here is disputing that Navi 1X was delayed. That doesn't mean that you can say that Navi 2X is delayed, as RDNA 2 builds heavily on RDNA 1 and could thus not have been designed before the major design elements of RDNA 1 were done. Is this a delay? No. The sins of the father and so on; you can't say that just because product 1 was delayed, product 2 that never had a launch date or timeframe or anything similar indicated was also delayed - that's faulty logic. The only time frame for RDNA 2 at the point RDNA 1 was delayed was "after RDNA 1". No delay there.

ARF
But AMD is way too late to implement 4K gaming for the masses, way too late to introduce ray-tracing, way too late to even compete in some segments of the market.
You need to work on your phrasing. "4k gaming for the masses" is not a feature that can be implemented, it is a performance goal that must be reached. AMD hasn't reached it yet (at least not 4k Ultra - you can play 4k medium-high just fine on a 5700 XT), mostly due to challenges with efficiency, as that sets a hard performance ceiling on what can be cooled in a PCIe form factor. "Way too late to introduce ray tracing" is also an absurd statement. Nvidia introduced this this generation. A one-generation feature gap is nothing at all, especially for a feature that barely exists in games. As long as the upcoming RTRT support on AMD cards performs well they will have delivered it in time. If not, we obviously have reason to complain. And "way too late to even compete in some segments of the market" - again, a statement that falls to pieces in terms of its internal logic. AMD/ATI has historically competed across the entirety of the GPU market. They have had a period of absence from the high-end/flagship space, yes, but how does that make it "too late" for them to return there? There's nothing stopping them from doing so as long as their architecture and technology allows them to do so.

Super XP
Your statement makes no sense. Navi was released Q3 2019. What do you expect AMD to do? Keep launching more quarter after quarter? Not even Nvidia does such a thing. Navi is in a rotation. So expect a 5700XT upgrade by Q3 2020 exactly what the RDNA 2 roadmaps have stated since before 2019.
I think the implication here was that they are/were very unhappy with their 5700XT. Which really needs explanation to be relevant to this thread whatsoever.
Posted on Reply
#96
ARF
Valantar
Nobody here is disputing that Navi 1X was delayed
Navi 2* depends on the Navi 1* launch. If one is delayed, the other is delayed too.

But I do expect the big Navi much sooner. According to me, it must already have been launched.

It is not and I do explain it in front of myself with bizarre political decisions.

In recent interviews, Mr. Papermaster from AMD says that they try to implement only right IPC improvements. What does "right" mean and if he is the person who decides, then these are subjective and wrong decisions.

See how many times the word "right" has been said by him:

https://www.anandtech.com/show/15268/an-interview-with-amds-cto-mark-papermaster-theres-more-room-at-the-top


Valantar
You need to work on your phrasing. "4k gaming for the masses" is not a feature that can be implemented, it is a performance goal that must be reached.
How do consoles with poor compared to the top PC hardware run 4K then and why?
Why are 4K TVs mainstream now?
Posted on Reply
#97
r.h.p
ARF
Navi 10's VCN 2.0 doesn't even support 8K video playback. People complain:


Amd/comments/cy9h72



There is definitely a delay in Navi's launch. If it was originally supposed to launch in H1 2018, it was pushed back to H2 2019.
It's particularly interesting to hear Navi's story, was it originally intended for N14 node and then moved forward to N7, was simply N7 too late........






The problem is that there is no alternative of AMD.
They offer some features in their graphics that no one else can or will because their IP competence is lower.

But AMD is way too late to implement 4K gaming for the masses, way too late to introduce ray-tracing, way too late to even compete in some segments of the market.

To be honest, I would be happier if Nvidia goes only for the low-end and mid-range markets, while AMD competes only with itself at the top high-end tier.
um I had no problem running the you tube video although some of it was a bit lagggy , although im using a 1440p monitor ?

also did the 8k benchmark




ran the 8k peru video no problem , but I don use any of those other browsers except brave for personal stuff not watching videos

Posted on Reply
#98
Valantar
ARF
Navi 2* depends on the Navi 1* launch. If one is delayed, the other is delayed too.
That's not how the concept of a delay works. A delay depends on something having some sort of timeframe attached to it. AMD published roadmaps showing Navi pre-2019, looking like early 2018. Navi arrived in mid-to-late 2019. Until the launch of Navi they had never published a roadmap showing RDNA 2/Navi 2(X). As such they had made zero promises about when Navi 2 was to arrive. "After Navi 1" might mean 6 months after or five years after; it's too vague to actually indicate anything at all. So while it is indeed somewhat reasonable to think that developmental delays for RDNA (1) delayed the development of RDNA 2, you cannot extend that into saying the RDNA 2 launch is delayed, simply because no time frame was given.

ARF
But I do expect the big Navi much sooner. According to me, it must already have been launched.
I don't know if you're trying for irony or sarcasm and a language barrier is mucking it up for you, or if this is just pure nonsense, but it comes off as the latter.

ARF
It is not and I do explain it in front of myself with bizarre political decisions.
This, on the other hand, is nothing other than pure, unadulterated nonsense. You'll need to try to use coherent sentences if you want what you are saying to be understood.

ARF
In recent interviews, Mr. Papermaster from AMD says that they try to implement only right IPC improvements. What does "right" mean and if he is the person who decides, then these are subjective and wrong decisions.

See how many times the word "right" has been said by him:
Uhh... The CTO of a company is supposed to be the one in charge of technological decision-making, no? It kind of makes sense that he is the one responsible for those decisions, even if the reality of the matter obviously is that the decisions are made based on the work and input of the engineering teams working under him. Also, "subjective"? What else are they supposed to be? Given that objectivity is an utopian ideal that humans are entirely incapable of reaching, every decision ever made is subjective. But even beyond that, what on earth makes you think the chief engineer of a company is making unfounded judgement calls rather than making decisions based on what are the best moves in terms of developing their technology? Of course it's possible for these choices to turn out to be completely wrong (Hello, Bulldozer architecture!), but that is largely down to the fact that nobody can predict the future, and that every decision is made within the constraints of what is possible in the circumstances in which the decision is made. I would therefore assume the "right" IPC improvements therefore means some balance of a) achieveable with the available resources and within the required time frame, b) high-yield compared to the engineering effort required, c) relevant to real-world workloads, and d) suited to underpin future development. Unless you have access to information to contradict this, your arguments here don't make sense.
Posted on Reply
#99
medi01
Looking at TSMC process chart, I simply do not see where the perf/watt jump should come from.
7N => 7NP/7N+ could give 10%/15% power savings, but the rest...
So, 35-40% improvement would come from arch updates alone?
And that following major perf/watt jump Vega=>Navi?

Vya Domus
RDNA is already worlds apart from GCN, the only real thing in common is that RDNA supports both wavefronts of 32 and 64, that's it. Well, that comes with the caveat that GPU architectures in general aren't very different one from another. GPUs have shallow pipelines, no out of order execution, no real branch prediction, they're mostly simple vector processors, there is just not a whole lot you can tweak and change.

In fact if you look throughout the history of GPUs you'll see that most of the performance typically comes from more shaders and higher clockspeeds, that's pretty much the number one driving factor for progress by far.
Welp, what about Vega vs Navi? Same process, 330mm2 with faster mem barely beating 250mm2 chip from the next generation.

TKnockers
5700xt
Messages 9 (0.07/day)

Ahaha, hi there, is it you, burnt fuse? :D
Posted on Reply
#100
ARF
Look, Valantar , I am talking about simple thing competitiveness, you are talking about utopia and how the CTO is always right.
The same people who introduced R600, Bulldozer, Jaguar, Vega and now have two competing chips Polaris 30 and Navi 14 covering absolutely the same market segment.

Please, let's just agree to disagree with each other and stop the argument here and now.

Thanks.
Posted on Reply
Add your own comment