Question Speculation: RDNA2 + CDNA Architectures thread

Page 69 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html
 

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
AMD’s 3080 equivalent should edge out the 3080 by between 5 and 20% (depending on the game) if all of the various tidbits of information I have are correct.

Don’t expect a big price difference between either card.

AMD also apparently has something for the 3090, but it isn’t dropping yet.
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
AMD’s 3080 equivalent should edge out the 3080 by between 5 and 20% (depending on the game) if all of the various tidbits of information I have are correct.

Don’t expect a big price difference between either card.

AMD also apparently has something for the 3090, but it isn’t dropping yet.
Lol. Please. AMD can't even beat their own previous gen cards how will they beat Nvidia. RX580 still is much better than 5500XT overpriced card. Maybe AMD will beat 3090 with RDNA 3 with 5nm but I still doubt.
 
  • Haha
Reactions: spursindonesia

uzzi38

Platinum Member
Oct 16, 2019
2,705
6,427
146
Lol. Please. AMD can't even beat their own previous gen cards how will they beat Nvidia. RX580 still is much better than 5500XT overpriced card. Maybe AMD will beat 3090 with RDNA 3 with 5nm but I still doubt.
Damn, you got RDNA2 figured out, AMD were never intending on beating the 5700XT, they can't beat their last gen card after all.
 

eek2121

Diamond Member
Aug 2, 2005
3,100
4,398
136
Lol. Please. AMD can't even beat their own previous gen cards how will they beat Nvidia. RX580 still is much better than 5500XT overpriced card. Maybe AMD will beat 3090 with RDNA 3 with 5nm but I still doubt.
I don’t normally feed trolls, but to which cards are you referring to?

You don’t have to believe me, but be prepare to be disappointed. Keep in mind I nailed the 3080 positioning many months ago.
 

blckgrffn

Diamond Member
May 1, 2003
9,298
3,440
136
www.teamjuchems.com
If you don't mind me asking, are you a small biz retailer?

Not to derail the thread earlier, but I can't seem to PM you - yes. PM if you want/need more details, but as life gets stressful these forums are my go to for escapism. :cool:

Thankfully, there is some action going on. Earlier this spring/summer it was way to early to get excited about these launches, but we are in the thick of it now!

FWIW, the 5500 xt in my mind was always supposed to be the replacement for the RX 460/560 which it of course obliterates.

Given we never had full stack Vega refresh, we milked those for so long....

The RX 480/580 when they came out were about $200-$240 in 2016 and so are more the 5600 xt segment. The RX R9 380s and 390s are so competitive that... wait, you can't buy those anymore?

Polaris and the islands cards were around for so darn long. Man.
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
I don’t normally feed trolls, but to which cards are you referring to?

You don’t have to believe me, but be prepare to be disappointed. Keep in mind I nailed the 3080 positioning many months ago.
I don't know what you mean but AMD fans always hype things up to absurd degree and it always falls wrong. All the Vega hype killed it. It would be better if you said biggest Navi will not be able to beat 3070 so that when it comes close to 3080 atleast that will be a positive surprise.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,249
136
Lol. Please. AMD can't even beat their own previous gen cards how will they beat Nvidia. RX580 still is much better than 5500XT overpriced card. Maybe AMD will beat 3090 with RDNA 3 with 5nm but I still doubt.


I don't know what you mean but AMD fans always hype things up to absurd degree and it always falls wrong. All the Vega hype killed it. It would be better if you said biggest Navi will not be able to beat 3070 so that when it comes close to 3080 atleast that will be a positive surprise.

And your role in this is what? Twisting nipples? Why not do your part and refrain from speaking unneeded nonsense? Waiting for the official announcement/reviews before passing judgement is what the logical person will do.
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
And your role in this is what? Twisting nipples? Why not do your part and refrain from speaking unneeded nonsense? Waiting for the official announcement/reviews before passing judgement is what the logical person will do.
No sorry you're right. I just got a bit mad when he mentioned AMD has a 3090 competitor as history has shown AMD is unable to be competitive at the hi end since many years of promising and failing. But yes i agree lets wait for reviews and benchmarks before passing judgment.
 

blckgrffn

Diamond Member
May 1, 2003
9,298
3,440
136
www.teamjuchems.com
And your role in this is what? Twisting nipples? Why not do your part and refrain from speaking unneeded nonsense? Waiting for the official announcement/reviews before passing judgement is what the logical person will do.

Who wants their nipples twisted!?!?

Ah, this called forth the mustache ride scene from Super Troopers. 😂

I don’t want my nipples twisted, fwiw, but thanks for the chuckle.
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
I don't know what you mean but AMD fans always hype things up to absurd degree and it always falls wrong. All the Vega hype killed it. It would be better if you said biggest Navi will not be able to beat 3070 so that when it comes close to 3080 atleast that will be a positive surprise.
It doesn't take 150 IQ to figure out that a GPU with 100% more CUs, 25% higher core clocks, and at least 50% more memory bandwidth than RDNA1 GPU(RX 5700 XT) will be much faster than 45% faster GPU(RTX 2080 Ti), than that RDNA1 GPU.

How would a GPU with over 2X performance of RX 5700 XT lose to RTX 3070, which won't even be faster than RTX 2080 Ti, using RTX 3080 reviews, and scaling of SM's, is beyond me.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,340
5,464
136
It doesn't take 150 IQ to figure out that a GPU with 100% more CUs, 25% higher core clocks, and at least 50% more memory bandwidth than RDNA1 GPU(RX 5700 XT) will be much faster than 45% faster GPU(RTX 2080 Ti), than that RDNA1 GPU.

How would a GPU with over 2X performance of RX 5700 XT lose to RTX 3070, which won't even be faster than RTX 2080 Ti, using RTX 3080 reviews, and scaling of SM's, is beyond me.

He isn't saying it will perform like 3070, he is just saying it would be better to think that, so when it delivers ~3080 performance it will feel like a pleasant surprise.

If instead you start by thinking it's going to deliver 3090 performance, then ~3080 performance will be a disappointment.
 
  • Like
Reactions: Martimus

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
He isn't saying it will perform like 3070, he is just saying it would be better to think that, so when it delivers ~3080 performance it will feel like a pleasant surprise.

If instead you start by thinking it's going to deliver 3090 performance, then ~3080 performance will be a disappointment.
It won't perform like RTX 3080, unless you talk about 72 CU version that is cut down from N21. Then maybe.
 
  • Like
Reactions: Tlh97

Heartbreaker

Diamond Member
Apr 3, 2006
4,340
5,464
136
It won't perform like RTX 3080, unless you talk about 72 CU version that is cut down from N21. Then maybe.

No question 80 CUs could beat if 3080 if you ignore power usage.

But we have no idea how much AMD will have throttle 80 CUs to keep power in check.

You can't simply double Navi 10, without considering that also doubles power usage to 450 W. That's a lot of power to trim with minimal help from the process.

Marketing and Rumors are not iron clad truth to answer that question either.

The final proof when the cards are finalized and in third party hands. This is going to all come down to power usage, and how it is kept in check.
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
No question 80 CUs could beat if 3080 if you ignore power usage.

But we have no idea how much AMD will have throttle 80 CUs to keep power in check.

You can't simply double Navi 10, without considering that also doubles power usage to 450 W. That's a lot of power to trim with minimal help from the process.

Marketing and Rumors are not iron clad truth to answer that question either.

The final proof when the cards are finalized and in third party hands. This is going to all come down to power usage, and how it is kept in check.
We don't have to. We have seen already the consoles, and its much more efficient physical design. We have seen Renoir APUs and their efficiency, we also have seen how high those APUs IGPUs clock.

There is no reason why we cannot do some maths and with pretty good accuracy to extrapolate based on what already have the data, how RDNA2 GPUs will perform, knowing how RDNA1 scales.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,340
5,464
136
We don't have to. We have seen already the consoles, and its much more efficient physical design. We have seen Renoir APUs and their efficiency, we also have seen how high those APUs IGPUs clock.

There is no reason why we cannot do some maths and with pretty good accuracy to extrapolate based on what already have the data, how RDNA2 GPUs will perform, knowing how RDNA1 scales.

300W PSU in the new consoles, and the underclocked Ryzen sips power so most of it will be for the GPU, so not much info in that either.

We need to see the real card in third party hands to get the real answer.
 

Tup3x

Golden Member
Dec 31, 2016
1,086
1,084
136
We don't have to. We have seen already the consoles, and its much more efficient physical design. We have seen Renoir APUs and their efficiency, we also have seen how high those APUs IGPUs clock.

There is no reason why we cannot do some maths and with pretty good accuracy to extrapolate based on what already have the data, how RDNA2 GPUs will perform, knowing how RDNA1 scales.
Renoir APU has exactly as much to do with RDNA2 efficiency as Pascal with Ampere. We don't even know for sure that the consoles are 100% the same as upcoming cards. Judging by what Cerny said it might not be the case.

You are often so sure about the things you say yet more often you are wrong. I would refrain from being so sure about everything if I were you.

It's realistic to expect around RTX 3080 performance for AMD's top card. I'd expect that the second best will beat RTX 3070. NVIDIA will definitely launch SUPER/Ti variants after that though.
 

Glo.

Diamond Member
Apr 25, 2015
5,803
4,777
136
Renoir APU has exactly as much to do with RDNA2 efficiency as Pascal with Ampere. We don't even know for sure that the consoles are 100% the same as upcoming cards. Judging by what Cerny said it might not be the case.

You are often so sure about the things you say yet more often you are wrong. I would refrain from being so sure about everything if I were you.

It's realistic to expect around RTX 3080 performance for AMD's top card. I'd expect that the second best will beat RTX 3070. NVIDIA will definitely launch SUPER/Ti variants after that though.
It has so much to do with it that the same physical design team which made it possible is right now making RDNA2 GPUs.

If Vega iGPU is clocking at stock to 2100 MHz at stock, on desktop, and OC's to 2.4 GHz I see absolutely no problem for big GPU, like N21 to clock to 2.2 GHz.
300W PSU in the new consoles, and the underclocked Ryzen sips power so most of it will be for the GPU, so not much info in that either.

We need to see the real card in third party hands to get the real answer.
Yeah, we have seen. The power draw is the same as Xbox One GPU. So 52 CU's at 1.825 GHz sipping 130-140W of power.

No need to downplay.
 

pj-

Senior member
May 5, 2015
483
251
136

Up until now, our artists have always had to create approximations of reflections by hand to achieve that effect – but the ray tracing architecture in PlayStation 5 gives us “more power” to provide true reflections in real time. In DMC5SE, players have the choice to turn on ray tracing and prioritize resolution (targeting 4K @ 30fps), or frame rate (targeting 1080p @ 60fps).

For players who prefer to allocate their horsepower to frame rates over graphics, PS5 also gives console players the ability to play their games at frame rates up to 120fps (with a compatible display, of course), which is enabled in DMC5SE when players turn on High Framerate Mode.

RT reflections in a last gen port cuts framerate in half?
 
  • Like
Reactions: Olikan

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
The consoles will be close enough to RDNA2 for making a good comparison. Neither Sony or MS are going to bankroll any kind of custom solution that deviates too much. At that point why not just make your own chip if you don't want to leverage someone else's IP. Most of the customization is minor tweaks or more on the API side of things.

I don't know if NVidia will have as much leeway for any Super cards as they did before. The OC reviews for the 3080 suggest its already past the point of efficiency in terms of throwing more power at the cards and I even question why they pushed them as far as they did when 50W less wasn't even a 5% performance hit. The 3070 seems like it's pretty close to full die already, so all a 3070 SUPER could hope for is silicon improvements. There's at least room for a 3080 SUPER to have more shaders, but since the 3080 is on GA102 it butts heads with the 3090 pretty quickly. Unless the Samsung process improves considerably then the SUPER cards are going to see a minor boost to clock speed, very few additional shaders, and mainly an increase in the amount of memory.

If AMD actually comes out with a real competitor to NVidia I think that NVidia is going to be on the ropes in a way that they haven't been for almost a decade. They're at the mercy of the Samsung process and moving everything to TSMC would be both expensive and potentially alienate customers that have bought the initial cards.

The pricing alone suggests that NVidia is expecting a lot more from AMD than they've been able to deliver for several generations. How much AMD actually manages to deliver remains to be seen though.
 

blckgrffn

Diamond Member
May 1, 2003
9,298
3,440
136
www.teamjuchems.com



RT reflections in a last gen port cuts framerate in half?

Haha, have you seen what RT does in Minecraft? I don't the generation of the title has nearly so much to do with it - RT is just really expensive.

Oh heck yeah I'll take 1080p/60FPS/RT on. Sounds great!
 
  • Like
Reactions: Olikan

GoodRevrnd

Diamond Member
Dec 27, 2001
6,801
581
126
The consoles will be close enough to RDNA2 for making a good comparison. Neither Sony or MS are going to bankroll any kind of custom solution that deviates too much. At that point why not just make your own chip if you don't want to leverage someone else's IP. Most of the customization is minor tweaks or more on the API side of things.

I don't know if NVidia will have as much leeway for any Super cards as they did before. The OC reviews for the 3080 suggest its already past the point of efficiency in terms of throwing more power at the cards and I even question why they pushed them as far as they did when 50W less wasn't even a 5% performance hit. The 3070 seems like it's pretty close to full die already, so all a 3070 SUPER could hope for is silicon improvements. There's at least room for a 3080 SUPER to have more shaders, but since the 3080 is on GA102 it butts heads with the 3090 pretty quickly. Unless the Samsung process improves considerably then the SUPER cards are going to see a minor boost to clock speed, very few additional shaders, and mainly an increase in the amount of memory.

If AMD actually comes out with a real competitor to NVidia I think that NVidia is going to be on the ropes in a way that they haven't been for almost a decade. They're at the mercy of the Samsung process and moving everything to TSMC would be both expensive and potentially alienate customers that have bought the initial cards.

The pricing alone suggests that NVidia is expecting a lot more from AMD than they've been able to deliver for several generations. How much AMD actually manages to deliver remains to be seen though.
With where they are on the efficiency curve it seems like their only option is a Ti refresh on a refined process. Maybe they could get Supers out if 8nm tightens up but it doesn't seem like there's any room to maneuver there.
 

Hitman928

Diamond Member
Apr 15, 2012
6,186
10,693
136