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

 

[moinmoin]

It isn't as though the would have done that knowing in advance that it wouldn't work out. If they do have some large cache designed to help offset the smaller bus the simple explanation is that their approach didn't work out as well in reality as they had planned.

Chip developers use simulations to avoid getting into exactly such situations. The only way it then can still happen is bad management (Koduri) or over-promised capabilities for the node used (Ampere on Samsung). AMD's track record with TSMC is pretty flawless so far, and RDNA2 isn't even on a new node, so simulations should have become even better and not worse than with RDNA1.

 

[alexruiz]

6500XT	40 CU	6GB GDDR6	~250mm²	~175w TDP
6600XT	56 CU	8GB GDDR6	~350mm²	~215w TDP
6700XT	60 CU	8GB GDDR6	~350mm²	~235w TDP
6800XT1	72 CU	12GB GDDR6	~450mm²	~255w TDP
6900XT	80 CU	16GB GDDR6	~450mm²	~275w TDP
6950XT2	96 CU	24GB HBM2E	~550mm²	~300w TDP

1 Not part of initial line-up; preparatory response to 3070/80 Ti/Su.
2 Marketed as "ultimate prosumer" card, appears to be RVII 2.0.

Interesting info.
How confident are you in these specs?

 

[Shivansps]

Why wouldn't cache be counted as part of "IPC"? AMD's claim/goal is 50% perf/watt improvement for RDNA2. Everything that gets it there is fair play.

Kinda inane article, why would AMD challenge its own long time customers in the console field? It's far more likely that Project Quantum, if ever released commercially, is supposed to challenge Intel's gaming NUCs.

WELL... there is this supposed Van Gogh APU with RDNA2 IGP... If they are able to do something with that and also add a PCI-E for a optional dgpu, that would be interesting.

It would also show that AMD has no sold out to MS and Sony.

 

[Glo.]

Interesting info.
How confident are you in these specs?

Since we know the specs of the dies, and there is no 96 CU die, I'd say zero confidence in them.

Navi 21 - 80 CUs/256 bit GDDR6,
Navi 22 - 40 CUs/192 bit GDDR6,
Navi 23 - 32 CUs/128 bit GDDR6.

 

[Shivansps]

256bit is not a bit too low for what could potencionally be a 3090 level card or better?

 

[Glo.]

Paul says that the SKU demoed had between 64 and 72 CUs, according to his sources.

 

[HurleyBird]

Paul says that the SKU demoed had between 64 and 72 CUs, according to his sources.

If true, then wow. They'll probably beat the 3090 in vanilla rasterized performance. If true.

 

[Saylick]

If true, then wow. They'll probably beat the 3090 in vanilla rasterized performance. If true.

I would love to see the flagship 80 CU card matching a 3090 at $999 or less.

 

[Kuiva maa]

Paul's video is talking about an nvidia-like arrangement. 6800 series is the consumer focus and 6900 will be quite a bit more expensive, just like what nvidia is doing with 3070-3080 vs 3090. This actually agrees with what our Russian new poster is saying (Radeon VII like prosumer top card). I will curb my enthusiasm and still expect an 80CU as top SKU with performance anywhere between 3080 and 3090 , with a price according to said performance (3080 price for equivalent performance). If however, such a 24HBM2 550+ beast exists and it beats the 3090, I will get it 100%.

 

[Vope45]

We are talking about Raja's RTG, or David Wang's/Lisa Su's RTG?

Its not the same RTG anymore, guys.

Navi is still Raja's child.

 

[raghu78]

Paul says that the SKU demoed had between 64 and 72 CUs, according to his sources.

I believe AMD showed off RX 6800XT (aka Navi 21 XL 16 GB) with 64 CU. The way I see it even teasing RX 6000 was not necessary unless AMD were very confident of Navi 21. I believe RX 6900XT will be roughly 25% faster than RX 6800XT. So for rasterization I do expect RX 6900XT to take the crown with lower power and lower price. Its the trifecta which any company would dream of for a new product launch - better perf, perf/watt and perf/$ than competition.

 

[itsmydamnation]

Navi is still Raja's child.

it was sept 17 when Raja took his sabbatical. So RDNA/RDNA 2 might be Raja's teams direction or it could be completely different we will never know. RDNA launched with a very large amount of errata so it does seem like it was "rushed" , make of that what you will.

 

[Glo.]

Navi is still Raja's child.

Nothing points to this, especially considering that Zen team engineers worked on it way after Raja's leave.

 

[Vope45]

Nothing points to this, especially considering that Zen team engineers worked on it way after Raja's leave.

His twitter on Sep 1st.

 

[Glo.]

His twitter on Sep 1st.

Which is quite ironic, because he hasn't twitted ANYTHING on September 1st.

There is a RT from 31 August, and next twitt is on September 2nd.

Kek.

 

[Vope45]

Which is quite ironic, because he hasn't twitted ANYTHING on September 1st.

There is a RT from 31 August, and next twitt is on September 2nd.

Kek.

https://twitter.com/x/status/1301028641652965376

 

[Glo.]

https://twitter.com/x/status/1301028641652965376

And this says that RDNA2 is his child?

What the ... ?

 

[eek2121]

Navi is still Raja's child.

I disagree. Vega was Raja's child and Navi was another team. All the evidence points to it. AMD had more than one team working on Radeon, even when they were broke. That is why Raja went out the way he did.

I believe AMD showed off RX 6800XT (aka Navi 21 XL 16 GB) with 64 CU. The way I see it even teasing RX 6000 was not necessary unless AMD were very confident of Navi 21. I believe RX 6900XT will be roughly 25% faster than RX 6800XT. So for rasterization I do expect RX 6900XT to take the crown with lower power and lower price. Its the trifecta which any company would dream of for a new product launch - better perf, perf/watt and perf/$ than competition.

I've told people this countless times. There are a LOT of NVIDIA fanboys here, and they are only countered by equally annoying AMD fanboys. People like me that simply have an objective look (and, in my case, experience in owning and operating a successful business) can clearly see what happened. AMD demoed a 72 CU card (the rumored 6800XT). They don't want to giveaway the 6900XT, which will soundly beat the 3080. I reject the argument that it's a 64 CU (or less) card because Lisa did imply Big Navi, though she DID leave that open to interpretation...I haven't looked at the fine print, so I don't know whether it offers any hints confirming the card to be 'Big Navi' or the middle tier chip. As a bonus, the 72 CU card may not have even been running at stock speeds.

What I can tell you with absolute certainty is that anyone claiming that Navi2X had an IPC regression should be kicked off this forum, because IMO they are trolling. The cards also don't have memory bandwidth issues. This much is evident because they DROPPED the bus size for the 40CU card from 256 bit to 192 bit. The only reason they'd do that is if the card performed too close to the next card up the stack...the 52 CU card. Yes, there IS a 52 CU card. I'm still curious if I'll end up being pretty close on the specs I estimated a while ago...or if the speculation at ChipHell will be.

 

[eek2121]

And this says that RDNA2 is his child?

What the ... ?

No, it says he used the term 'BFGPU'.

 

[Vope45]

And this says that RDNA2 is his child?

What the ... ?

It implies Raja knew about GA102 and therefore already had plans to counter with RDNA. Raja being Raja of course likes running his mouth like a coyote with Vega being a prime example. If Raja doesnt know about GA102 he wouldn't have tweeted about it on the same launch day. Remember RDNA1 has a plethora of issues ? I reckon it was done in the same manner as Vega, good ideas but with low budget and poor execution. People in the industry talk with each other, quite a lot actually but I'm sure you are fully aware of this already.

 

[Vope45]

I disagree. Vega was Raja's child and Navi was another team. All the evidence points to it. AMD had more than one team working on Radeon, even when they were broke. That is why Raja went out the way he did.



I've told people this countless times. There are a LOT of NVIDIA fanboys here, and they are only countered by equally annoying AMD fanboys. People like me that simply have an objective look (and, in my case, experience in owning and operating a successful business) can clearly see what happened. AMD demoed a 72 CU card (the rumored 6800XT). They don't want to giveaway the 6900XT, which will soundly beat the 3080. I reject the argument that it's a 64 CU (or less) card because Lisa did imply Big Navi, though she DID leave that open to interpretation...I haven't looked at the fine print, so I don't know whether it offers any hints confirming the card to be 'Big Navi' or the middle tier chip. As a bonus, the 72 CU card may not have even been running at stock speeds.

What I can tell you with absolute certainty is that anyone claiming that Navi2X had an IPC regression should be kicked off this forum, because IMO they are trolling. The cards also don't have memory bandwidth issues. This much is evident because they DROPPED the bus size for the 40CU card from 256 bit to 192 bit. The only reason they'd do that is if the card performed too close to the next card up the stack...the 52 CU card. Yes, there IS a 52 CU card. I'm still curious if I'll end up being pretty close on the specs I estimated a while ago...or if the speculation at ChipHell will be.

Recalling back to 2017 when Vega was released, people were saying Vega was already in development before Raja joined back and Navi was his child.

But of course AMD has a history of disappointment starting with Hawaii so there's that.

 

[Konan]

^^^There is a pretty good thread here on the same topic

https://www.reddit.com/r/Amd/comments/buql4h/_/epg75uu

 

[PhoBoChai]

AMD might be borrowing a page from NV marketing. Ampere is upto 1.9X Turing Perf/Watt after all (at 60 fps on some game). Without knowing at which level of performance AMD is talking about, the % increase is just marketing talk.

AMD's recent marketing events under Lisa Su is quite accurate to final perf or power efficiency claims.

Navi is still Raja's child.

RDNA 1 is Raja's development when he was hired back to AMD, to give them a new post-GCN architecture.

RDNA 2 is after his departure to Intel, thus, the lead would be David Wang, assisted by Suzanne Plummer's Zen team.

Vega features what were missing, seem to be Raja's attempt to back-port these Navi/RDNA1 features into the last GCN iteration, without success.

 

[dzoni2k2]

Recalling back to 2017 when Vega was released, people were saying Vega was already in development before Raja joined back and Navi was his child.

But of course AMD has a history of disappointment starting with Hawaii so there's that.

Hawaii was not a disappointment. Reference cooling solution was. Hawaii was great and aged far better than Kepler.

 

[Vope45]

AMD's recent marketing events under Lisa Su is quite accurate to final perf or power efficiency claims.


RDNA 1 is Raja's development when he was hired back to AMD, to give them a new post-GCN architecture.

RDNA 2 is after his departure to Intel, thus, the lead would be David Wang, assisted by Suzanne Plummer's Zen team.

Vega features what were missing, seem to be Raja's attempt to back-port these Navi/RDNA1 features into the last GCN iteration, without success.

RDNA 1 is a stepping stone for RDNA 2 therefore Raja must have involved to RDNA 2 to a degree ( unknown ). I still remembered someone here (?) suggested that Zen was named after Suzanne lol. I lurked for decades, no jokes. Only coming out of the woodwork due to Covid.

To @dzoni2k2

Hawaii is in itself rather problems free however everything else was a complete disaster.

It is also the first sign showing AMD was cash starved at the time.