Question Speculation: RDNA3 + CDNA2 Architectures Thread

[uzzi38]

Man I have been dying to make this one for a while now.

First rumours for RDNA3 are here so new thread time!

Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3 is much bigger than from RDNA1 to RDNA2. We should expect many big improvements in GFX11. 🤔" / Twitter

 

[biostud]

Looking forward to Computex presentations

 

[AtenRa]

And are we sure that is an official slide? Doesn't look like it tbh.

Yeah that's got to be a fake slide. They wouldn't mention Ada at this point.

Its official slide

 

[jpiniero]

Its official slide

Where'd you get it? Please don't say WTFTech.

 

[AtenRa]

Where'd you get it? Please don't say WTFTech.

AMD

 

[Aapje]

Do you have a link?

 

[AtenRa]

Do you have a link?

No its from a pdf file

 

[leoneazzurro]

https://twitter.com/x/status/1527632813943848960

Interesting

 

[jpiniero]

I will say that if Navi 31/32 are still 6 months away, AMD should be a bit nervous.

 

[gdansk]

I will say that if Navi 31/32 are still 6 months away, AMD should be a bit nervous.

Nervous? AMD can paper launch whenever Nvidia wants. Who could tell the difference compared to the 2020 releases anyway.

I'm only half joking...

 

[Panino Manino]

Qualcomm jumped ship and switched the current Snapdragon from Samsung to TSMC.
I fear that the jump to the new node alone will put Nvidia ahead of RDNA3.

 

[jpiniero]

Nervous? AMD can paper launch whenever Nvidia wants.

The timing might not work out to do something like that.

 

[Kepler_L2]

Qualcomm jumped ship and switched the current Snapdragon from Samsung to TSMC.
I fear that the jump to the new node alone will put Nvidia ahead of RDNA3.

? They are using the same node.

 

[exquisitechar]

? They are using the same node.

Yes, but he probably means that Nvidia has more to gain (going from Samsung’s 8nm to N5) than AMD does.

 

[Karnak]

Well, N7 for AMD with both RDNA and RDNA2 and considering that the perf/watt improvement for the latter was just insane.

Yes, the node does matter ofc, but the architecture itself is way more important.

 

[exquisitechar]

Yes, the node does matter ofc, but the architecture itself is way more important.

True, I’m just explaining the reasoning there.

We’ll find out soon enough how RDNA3 holds up. I’m fed up with Twitter leakers and their weekly specification changes at this point. Only thing I’ll say is that I hope AMD’s launch doesn’t lag too far behind Nvidia’s, if that mid July Ada rumor is true.

 

[Karnak]

It has been like Q4 for a Zen4 launch quite some time, then we were down to mid September and now we are at a point where Gigabyte and all the others will showcase their X670 boards already during the next week. That means probably that it's even earlier than September.

In terms of leakers and their "launch shedules" I just don't trust them.

 

[Saylick]

If I were AMD, the least I would do is paper launch before Nvidia hard launches Lovelace. That is, assuming RDNA 3 is competitive.

 

[DiogoDX]

If the slide is real RDNA3 will release only in 2023. And if Nvidia launching ADA in july rumor is true I preditc that AMD will reach ~5% market share by the time RDNA3 launches.

 

[ryanjagtap]

AMD may go with their perf/$ marketing and increase the value of their GPUs by reducing prices or give gaming bundles to keep the market share till hard launch of RDNA3.

 

[Saylick]

Nah, N33 will have products on shelves in 2022. I can see N31 launching a month or two after N33, but I would be surprised if it slipped to 2023. The amount of marketshare Lovelace can take might not be as drastic as you expect; Lovelace is still going to be a relatively high-end product, and the people who will be upgrading are likely people who already have a high-end Nvidia GPU. AMD should be more worried if cheaper Ampere GPUs flood the market because that would definitely eat into AMD's traditional "bang for buck" demographic.

 

[ryanjagtap]

Nah, N33 will have products on shelves in 2022. I can see N31 launching a month or two after N33, but I would be surprised if it slipped to 2023. The amount of marketshare Lovelace can take might not be as drastic as you expect; Lovelace is still going to be a relatively high-end product, and the people who will be upgrading are likely people who already have a high-end Nvidia GPU. AMD should be more worried if cheaper Ampere GPUs flood the market because that would definitely eat into AMD's traditional "bang for buck" demographic.

Well according to Techspot, Nvidia GPU (low end at least) are 24% above MSRP as of their May update whereas AMD GPU are 5% above MSRP. If this trend continues then even if they flood the market with low end Ampere GPUs, they will not be seen as an attractive buying prospect.

 

[Saylick]

Well according to Techspot, Nvidia GPU (low end at least) are 24% above MSRP as of their May update whereas AMD GPU are 5% above MSRP. If this trend continues then even if they flood the market with low end Ampere GPUs, they will not be seen as an attractive buying prospect.

That's assuming the price of Ampere remains elevated above MSRP, which is due to vendor/AIB markup. Apparently, AIBs don't want to sell at MSRP because they themselves have a ton of supply that they bought at elevated prices. If they sold at MSRP, some might actually lose money so it is in their best interest to milk their inventory as much as possible.

If Nvidia wanted, they could screw over their partners and issue an MSRP for Lovelace that puts serious downward pressure on the elevated pricing of Ampere. Each AIB would have to unload Ampere before other competing AIB companies does so first, and it might trigger a mad dash to clear Ampere inventory.

 

[beginner99]

The 6950 is a very inefficiently overclocked card

Not only prices are insane but also that basically undervolting is now the preferred method over overclocking. reduce power by 20-50% while losing 5% of performance. A no brainer really.

 

[biostud]

Well according to Techspot, Nvidia GPU (low end at least) are 24% above MSRP as of their May update whereas AMD GPU are 5% above MSRP. If this trend continues then even if they flood the market with low end Ampere GPUs, they will not be seen as an attractive buying prospect.

Looking at the prices in Denmark, it is now possible to get one 3080 from a single vendor in a single shop close to MSRP. The rest are still 20 percent over, and 6800XT are 30% over their MSRP. Obviously we don’t know the pricing and performance of next gen cards, but if the price/performance is better than current cards then I’ll get one.

 

[Mopetar]

My current theory is that Lisa Su felt so bad for the 'leakers' that she just called one of them up and told them straight up about the 1 compute die.

OK, not really.

If I were the CEO I'd call up tech tubers every night and tell them all kinds of fantastical things, occasionally given them the straight truth. If they ever figured out who I was by the sound of my voice I'd change it in an obvious way and feign ignorance. Then I'd offer to meet them in person, but I'd wear Groucho glasses and deny that it was actually me. I'd even give a clearly fake name made from swapping the first few letters of my first and last name.

When they talk to one of the other leakers about it later thinking they have this unbelievable story, the other leaker will just reply, "Yeah, Pete's kind of weird like that."