Question Speculation: RDNA2 + CDNA Architectures thread

[uzzi38]

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

 

[jpiniero]

RNDA 2 "pipeline" is the minimal architecture.

No I think nVidia is talking about the shader fallback method they are using with Pascal. Not about AMD at all.

 

[sontin]

The "minimal architecture" is for raytracing acceleration. That is not the software emulation.

 

[DiogoDX]

No I think nVidia is talking about the shader fallback method they are using with Pascal. Not about AMD at all.

The shader fallback doesn't have the separated harware to do ray and tri intersection like in the picture. It is talking about the RDNA2 pipeline talked by Microsoft in the hotchips XBOX presentation.

 

[GodisanAtheist]

What's the rush? Prices will plummet further once Ampere hits the streets.

- Or they'll go up thanks to scalpers/scarcity/underwhelming performance/lack of competition/mining/etc

All I know is I have a performance level and a price level for prior gen in my head where I wont regret anything regardless of what launched next.

If I come across one of those while AMD is still playing coy then they lost a potential sale.

 

[A///]

Not this time.

Usually doesn't mean all the time.

 

[itsmydamnation]

RNDA 2 "pipeline" is the minimal architecture.

funny how NV just admitted they suck at branching, isn't it nice how all that general scalar hardware inside the CU is growing along with issue width

 

[Kenmitch]

- Or they'll go up thanks to scalpers/scarcity/underwhelming performance/lack of competition/mining/etc

All I know is I have a performance level and a price level for prior gen in my head where I wont regret anything regardless of what launched next.

If I come across one of those while AMD is still playing coy then they lost a potential sale.

What are you babbling on about? You said you had your eye on used 2080Ti's. Once Ampere hits the street the 2080Ti's should drop lower in price. Nobody knows when AMD is going to announce anything yet.

 

[GodisanAtheist]

What are you babbling on about? You said you had your eye on used 2080Ti's. Once Ampere hits the street the 2080Ti's should drop lower in price. Nobody knows when AMD is going to announce anything yet.

- What are you getting so worked up about?

All I am saying is that there are a lot of factors that affect pricing.

Logically card prices should lower over time as volume increases and demand drops, but we have seen two very clear instances where this has not been the case with the first mining boom that drove GCN values through the roof (but at least crashed and left a glut of absurdly well priced cards on the market) then the second boom which drove GCN and Pascal prices through the roof where they never really came back down.

Technically, 2080ti's shouldn't be selling for $600+ dollars now based on what we know... But they are. Don't be so confident that the average buyer knows that they're doing or doesn't attach additional value to owning the prior gen's flag ship over the current gen's mid range.

 

[Geranium]

- What are you getting so worked up about?

All I am saying is that there are a lot of factors that affect pricing.

Logically card prices should lower over time as volume increases and demand drops, but we have seen two very clear instances where this has not been the case with the first mining boom that drove GCN values through the roof (but at least crashed and left a glut of absurdly well priced cards on the market) then the second boom which drove GCN and Pascal prices through the roof where they never really came back down.

Technically, 2080ti's shouldn't be selling for $600+ dollars now based on what we know... But they are. Don't be so confident that the average buyer knows that they're doing or doesn't attach additional value to owning the prior gen's flag ship over the current gen's mid range.

If you want cheap 2080Ti, then ask Nvidia to lower the price. AMD is not pricing the 2080Ti but Nvidia doing that. You are asking the wrong company.

And why AMD should care about you, when you are buying product from competitor not from them.

 

[eek2121]

??? Might just be fuel for the hype train?

I know a guy he sends a birdie my way occasionally to tweet in my ear ☺️. However you don’t have to believe me. Look at the clocks on the PS4. 36CUs and 2.23Ghz. I can tell you that they were only limited by thermals and to a lesser extent, power consumption.

 

[eek2121]

You do realize that they might just be sticking to their game plan?

If it was me in charge of their marketing I'd most likely use the PS5 and Xbone's advertising momentum/hype to my advantage.

Unconfirmed rumor says PS5 launch event is the 9th. I guess it's the 25th anniversary of the PS in the USA.

I guess we'll see how it turns out, but I don't really expect to see anything from AMD yet.

AMD will make an announcement when they are ready. They are in no hurry. I wouldn’t be either.

The shader fallback doesn't have the separated harware to do ray and tri intersection like in the picture. It is talking about the RDNA2 pipeline talked by Microsoft in the hotchips XBOX presentation.

Yet AMD’s console solution does RT just fine. The PC solution will have a slightly different approach from the Xbox.

 

[Glo.]

The less people expect from RDNA2, the better.

The more surprised they will be when AMD will release those GPUs.

 

[maddie]

I know a guy he sends a birdie my way occasionally to tweet in my ear ☺. However you don’t have to believe me. Look at the clocks on the PS4. 36CUs and 2.23Ghz. I can tell you that they were only limited by thermals and to a lesser extent, power consumption.

Thermals and power consumption, to use a metaphor, are two sides of the same coin, so saying thermals and to a lesser extent, power consumption is misleading.

 

[MrTeal]

Thermals and power consumption, to use a metaphor, are two sides of the same coin, so saying thermals and to a lesser extent, power consumption is misleading.

I get what you’re saying and they are, but you could run into thermal limits before you get to the point on the curve when power consumption skyrockets. In a GPU you have a lot more flexibility to control thermals than in a console that’s going to be shoved in someone’s enclosed entertainment cabinet.

 

[Glo.]

I can believe that RDNA2 will clock to 2.5 GHz in very specific scenarios.

I cannot believe in fairy tales that it will clock up to 3 GHz.

I cannot believe that we will get anything else than 2.3 GHz on lower ALU count GPUs, and at best 2.1 GHz on Navi 21.

Even that 2.1 GHz core clock on 80 CU GPU with 384 bit bus will result in 275W TDP and around 90-95% of performance of RTX 3090.

 

[maddie]

I get what you’re saying and they are, but you could run into thermal limits before you get to the point on the curve when power consumption skyrockets. In a GPU you have a lot more flexibility to control thermals than in a console that’s going to be shoved in someone’s enclosed entertainment cabinet.

I agree, but the implied here is that Sony has left performance untapped by specifying an inadequate cooling solution as power being not so critical as temps, mean that they are still within the efficient part of the freq/voltage curve.

Do we really think that with their CU disadvantage vs XBox, they will leave easy performance unused?

 

[Kenmitch]

I agree, but the implied here is that Sony has left performance untapped by specifying an inadequate cooling solution as power being not so critical as temps, mean that they are still within the efficient part of the freq/voltage curve.

Do we really think that with their CU disadvantage vs XBox, they will leave easy performance unused?

Let's not forget that Big Navi isn't the same as the SOC's in the PS5 and Xbone. They may have similar underlying architecture, but in the end they're not the same.

Big Navi might hit 3GHz? I wouldn't hold my breath, but there's always the crazy LN2 guys.

 

[raghu78]

I can believe that RDNA2 will clock to 2.5 GHz in very specific scenarios.

I cannot believe in fairy tales that it will clock up to 3 GHz.

I cannot believe that we will get anything else than 2.3 GHz on lower ALU count GPUs, and at best 2.1 GHz on Navi 21.

Even that 2.1 GHz core clock on 80 CU GPU with 384 bit bus will result in 275W TDP and around 90-95% of performance of RTX 3090.

Vega in Renoir already overclocks to 2.4 Ghz in desktop SKU. So I can see RDNA2 consistently hit 2.5 Ghz OC on the top bin N2x SKUs. In fact RDNA2 should clock better than Vega as its AMD's latest and best GPU architecture.

As for performance of N21 it depends on the perf/clock increase. My expectation is 10-15% based on the AMD slide at FAD 2020 where it seems to indicate that RDNA to RDNA2 perf/clock increase is like Zen to Zen 2. If its actually 15% then N21 80 CU could match RTX 3090. If Navi 21 is using 384 bit GDDR6 then the bandwidth will be 768 GB/s which is one area where RTX 3090 will have a decisive advantage. I do believe N21 will have 128 ROPs given the modular design of RDNA2 in which the SE is the building block.

RTX 3090 - 7 GPC, 41 TPC, 82 SM , 5248 cc, 112 ROPs, 916 GB/s
Navi 21 - 4 SE, 8 SA, 40 WGP, 80 CU , 5120 cc, 128 ROPs, 768 GB/s

The main factor is going to be pricing. I think AMD will target a $1000 price for N21 80CU top bin if they have the performance. If thats the case then NV is likely to bring a RTX 3080 Ti with 78-80 SM on a 352 or 384 bit memory bus with 11-12 GB memory and price it at $1000.

 

[Mopetar]

AMD is keeping quiet because they saw what the Ampere announcement did to the price of cards like the 2080 Ti.

Knowing AMD they still have some Polaris cards that they need to move before they drop any announcement about Navi.

Gotta clean out all that existing stock.

 

[Kenmitch]

AMD is keeping quiet because they saw what the Ampere announcement did to the price of cards like the 2080 Ti.

Knowing AMD they still have some Polaris cards that they need to move before they drop any announcement about Navi.

Gotta clean out all that existing stock.

WTF? no /s ???

It's AMD were talking about. They're not really into sticking it to the end user like Nvidia.

The poor OEM's holding the RTX 2xxxx's inventory is the one's to worry about.

 

[GodisanAtheist]

If you want cheap 2080Ti, then ask Nvidia to lower the price. AMD is not pricing the 2080Ti but Nvidia doing that. You are asking the wrong company.

And why AMD should care about you, when you are buying product from competitor not from them.

-It think you're misinterpreting the on going conversation for the individual post. I'll bring it back around and duck out, don't want to take up too much of the thread on bickering.

All I am saying is the longer AMD goes without releasing additional info, the more sales they lose to people that would normally upgrade to them this cycle (like me).

If I see a good RTX 2xxx card (used) for a good price, I will buy it and be out of the market for another 3-4 years. I know I'm not unique, so there must be other people like me out there.

So that being said, not sure why your posts are reading so aggressive and hostile.

 

[Konan]

Scaling..... Why isn't the XSX higher clocked?
52 CU Xbox Series X GPU @ 1.825Ghz
36 CU PS5 GPU can reach 2.23GHz

 

[Tarkin77]

I can believe that RDNA2 will clock to 2.5 GHz in very specific scenarios.

I cannot believe in fairy tales that it will clock up to 3 GHz.

I cannot believe that we will get anything else than 2.3 GHz on lower ALU count GPUs, and at best 2.1 GHz on Navi 21.

Even that 2.1 GHz core clock on 80 CU GPU with 384 bit bus will result in 275W TDP and around 90-95% of performance of RTX 3090.

Didn't AMD went a similar rout with Vega or Radeon VII? Relaxed density for higher Clocks? 505qmm is an awfull lot of space for only 80 CUs

 

[moinmoin]

Didn't AMD went a similar rout with Vega or Radeon VII? Relaxed density for higher Clocks? 505qmm is an awfull lot of space for only 80 CUs

That's a rumored size that may well turn out to be false. Also Renoir is very dense, and its iGPU clocks high nevertheless.

Scaling..... Why isn't the XSX higher clocked?
52 CU Xbox Series X GPU @ 1.825Ghz
36 CU PS5 GPU can reach 2.23GHz

Because Microsoft thinks it's better for the hardware performance to be predictable, which has been the console hardware MO up to now. Sony now thinks it's better to keep using the boosting power new hardware is capable of, despite it making the overall performance less predictable.

(It's an odd situation since at the same time Microsoft is all about compatibility across the Xbox ecosystem, meaning the high power of XSX can't be exploited without keeping the weaker Xbox devices in mind. Meanwhile Sony has the weaker system, but games can and should be dedicated to PS5.)

 

[jpiniero]

Because Microsoft thinks it's better for the hardware performance to be predictable, which has been the console hardware MO up to now. Sony now thinks it's better to keep using the boosting power new hardware is capable of, despite it making the overall performance less predictable.

No, Sony did that because they realized that the XSX was far more powerful and they wanted to close the gap as far as they could. The original intended spec was probably somewhere around a fixed 1.8-2 Ghz.