Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

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...

forums.anandtech.com

 

[inquiss]

Exactly. AFMF/AFMF2? Nvidia has nothing like it.
FSR2 being available as open source and thus completely customisable? Go take in the DLSS binary and be quiet about it.
FSR 3 is objectively better than DLSS 3 pretty much across the board. Early frame pacing problems notwithstanding.

There is a general issue with AMD putting out software that is customisable but imperfect, while Nvidia idiot proofs everything and loves to control everything themselves.
I do agree with the NVers that having an AI corrected upscaler is very necessary, but I do not think that it is a sufficient enough argument to claim that "NV wins everything".

Frankly, it's down to a question of general communication competence. If you look at something like FSR, you'd assume you plug n play the binary, not at all. Everyone who touched it circa FSR 2.1 said that it just doesn't work well and you're expected to tune it. This is why you end up with something like FSR 1 being somehow able to run No Man's Sky on Switch pretty flawlessly at 720p, while FSR 2 implementations that aren't done with an FSR expert's touch just kinda look bad.

I think NV's approach is better than AMD's here, most studios won't have any finetuning past the bare minimum nor will they commit someone to become an upscaling expert to tune it to perfection.

AMD used to have frame interpolation hardware ages ago. Was never a discussion point though. Best in class hardware.

AMD used to be less power hungry, it wasn't a metric Nvidia fans cared about until they were more efficient. Then after that Nvidia made a big deal about how they could push more power and that was good then. Sometimes power efficiency good, sometimes power efficiency bad. Just depends which one Nvidia is.

A lot of fanfare was made of nvenc and why that was important when AMD lagged. Then it was ray tracing. Then DLSS.

AMD still has Radeon chill to extend battery life, another AMD first. But it's not important because the narrative is that only things Nvidia invents first are important.

The goalposts shift and people saying AMD needs an AI upscaler to compete will, without noticing, pivot to some other important feature that AMD needs to compete with to get sales once AMD hardware has it.

 

[ToTTenTranz]

If it was a fair fight yes, but Nvidia has brilliant marketing that weaponises their buyers into being their simps. Go on Reddit or most social media places and get insulted for even suggesting buying AMD. Ask someone about Nvidia with any kind of criticism and get told "go buy AMD, it's what you deserve".

When you don't have a great product or great stack, and the competition has effectively reined in the support of innumerable amounts of simps in a cult that expects all Cultists to keep buying, praising, thanking Nvidia, lest they get kicked out, you hunker down and sell what you can.

None of what you wrote is mutually exclusive with what I wrote.

Yes, AMD at this point is fighting a cult mentality.
Which makes it even more inexcusable that reviews of many of their new architecture releases ended up focusing on cooler noise,

It's just that AMD won't sell below 10% of Nvidia's price/perf, since they'd just lose money.

No, it's not because they'd lose money. It's because they've been trying to elevate AMD's brand perception.

It's literally what Dave Baumann said in the B3D forums while he was in the Radeon Strategy group.

 

[Mahboi]

No, it's not because they'd lose money. It's because they've been trying to elevate AMD's brand perception.

No. It's that they'd lose money. That's an insane take lol.
"We'll raise our brand perception by lowering sales" makes zero sense if you think about it for about .2 seconds.

There is no opening for Radeon to penetrate the market further, that's the problem. Pricing doesn't fundamentally change your brand perception when your brand perception is confirmed to be the same thing that people perceive since 10 years: poor man's Nvidia.

It's literally what Dave Baumann said in the B3D forums while he was in the Radeon Strategy group.

Never heard of him and won't read B3D since it is the heartland of Nvidia insanity.

 

[ToTTenTranz]

"We'll raise our brand perception by lowering sales" makes zero sense if you think about it for about .2 seconds.

You just described the marketing strategies of all luxury brands and Apple.
Apple takes 85% of the worldwide smartphone net profits out of a <20% marketshare.

All brands would love to be able to sell a 10% better product for a 100% larger margin and 50% higher ASP than the competition. That's Nvidia compared to AMD nowadays, or Apple compared to Samsung.

Never heard of him and won't read B3D since it is the heartland of Nvidia insanity.

Back in 2013, B3D was a really great forum. It started to turn into a nvidia cult circa 2018-2019 with the RTX marketing push and certain users made everything about raytracing and DLSS. Things really went downhill after DF's Alex Battaglia started posting there to feed the cult, under the "Dictator" username no less.

Dave Baumann was an original co-founder of the Beyond3D website (not sure if he co-wrote the B3D Benchmark Suite) who joined AMD in 2006.
He's the guy responsible for making the Radeon HD4850 as popular as it became by pushing for a lower price to drastically increase sales and mindshare. Many years later he was defending what became AMD's GPU pricing strategies of today (just a bit below Nvidia).

 

[adroc_thurston]

It started to turn into a nvidia cult circa 2018-2019 with the RTX marketing push and certain users made everything about raytracing and DLSS. Things really went downhill after DF's Alex Battaglia started posting there to feed the cult, under the "Dictator" username no less.

Technical and relevant people like sebbbi leaving didn't help either.

 

[Mahboi]

You just described the marketing strategies of all luxury brands and Apple.

Apple has no competition.
Luxury brands are a literal Ponzi. And also have no competition. You don't have a benchmark comparing Louis Vuitton to Chanel.

Dave Baumann was an original co-founder of the Beyond3D website (not sure if he co-wrote the B3D Benchmark Suite) who joined AMD in 2006.
He's the guy responsible for making the Radeon HD4850 as popular as it became by pushing for a lower price to drastically increase sales and mindshare. Many years later he was defending what became AMD's GPU pricing strategies of today (just a bit below Nvidia).

You mean that time when ATI had a top quality product and they squandered all their potential because despite having a better product than Nvidia all across the board, some idiot thought that margins were secondary and that they should just drive NV out of the market by pricing them out?
That moment where ATI started going down and would completely lose the race to NV and get sold to AMD?
If he's responsible for that, I would triple not listen to the guy.

 

[ToTTenTranz]

Technical and relevant people like sebbbi leaving didn't help either.

I still follow sebbbi on X. He's quite active there and still shares his SIGGRAPH presentations

You mean that time when ATI had a top quality product and they squandered all their potential because despite having a better product than Nvidia all across the board, some idiot thought that margins were secondary and that they should just drive NV out of the market by pricing them out?

Don't be ridiculous.

AMD was coming from the bad HD2900 and mediocre HD3870/50 launches. The HD4850 was in good part responsible for making AMD jump in marketshare by over 10% and put AMD within 10% from Nvidia, something they hadn't been able to do since the Radeon 9700 Pro. ASP went up, marketshare went up and profits went up. Brand recognition obviously went up as well, since they gained marketshare.

And AMD didn't have a clearly superior product to Nvidia either. At the same time Nvidia was invading the market with the overclocked G92b based Geforce 9800 GT/GTX which were super popular cards at the time, and the GTX280/60 were just around the corner which were competitive enough.

AMD’s CPU business is doing better than expected, but it’s still not doing well. Meanwhile everything is up with GPU sales; prices and units shipped are up compared to last quarter and last year. The Radeon 4000-series has been the smash hit AMD has been looking for from its graphics division, with Q3 marking the first full quarter for the high-end products of that generation. It needs to be kept in mind just how lop-sided AMD’s GPU and CPU business are in terms of revenue, so doing well here isn’t enough, but all-told the business is at a comfortable position.

It was a great decision, especially considering this happened in 2008, the year when the subprime crisis spread to the whole world.

That moment where ATI started going down and would completely lose the race to NV and get sold to AMD?

ATi had been bought by AMD 2 years prior to all of this.

 

[poke01]

Look I'm no NV fanboy. I go where the tech is best/efficient and right now its Nvidia for GPUs and if it takes Intel to beat Nv then so be it. I want cheaper GPU prices and for Nvidia to feel pressure. Their head has been too long in the clouds now.

 

[gdansk]

if it takes Intel to beat Nv

Looks like we got a hopeless optimist. Just don't be disappointed. BM might meet some actual goals this time, but not that.

 

[Mahboi]

And AMD didn't have a clearly superior product to Nvidia either. At the same time Nvidia was invading the market with the overclocked G92b based Geforce 9800 GT/GTX which were super popular cards at the time, and the GTX280/60 were just around the corner which were competitive enough.

It was a great decision, especially considering this happened in 2008, the year when the subprime crisis spread to the whole world.

ATi had been bought by AMD 2 years prior to all of this.

Then either I've mixed up cards either you're the one being ridiculous.

Either way, AMD's problems are what they are. I'm tired of hearing this ridiculous trope that "AMD needs to lower prices" from people that can't understand the first thing about markets.
If they lower prices, they don't make more money nor raise their market position (I.E get a competitive advantage). That's the end of that.

The one case in which this argument would make a modicum of sense is if AMD decides to flood the channel and market with as much as they can, with the goal of getting the product out at any cost. This is an extremely unlikely thing to happen since:
A/ Nvidia still sells much better even when the price/perf ratio is 20% worse
B/ That would put increased strain on TSMC which has no reason to happen since the pecking order is still Instinct/EPYC/Ryzen/Radeon
C/ In case they go for an older, less used node, then they're effectively making a poor man's GPU, which will completely walk into the NV propaganda of "we are the real GPU, they're the loser's GPU"

For AMD to actually try flooding the channel and lowering prices, they'd have to output more volume than normal while offering a cheaper, less margin-y product. NONE of these things make sense for them to do. No need to put GPU pressure on TSMC for gamers that prefer Nvidia, no need to kill margins, no need to raise volume and potentially fail and get a crappy GPU excuse.

Their actual options are:
A/ AMD gets a squarely better product than NV and the NV cult megaphone spam starts getting distorted. They win by hardware quality. Unlikely but the most likely out of the three.
B/ AMD raises their software offerings to the level of NV and it boils down to "same, but cheaper and more available" since I expect FSR to start digging DLSS out of the market by being FOSS. This is incredibly unlikely for every reason known to man: NV puts a lot more effort than AMD in SW, bespoke SW is easier than generic, FOSS at this level of complexity is very difficult to draw a large crowd of devs, etc. I'd say this is the least likely outcome of all.
C/ NV makes a serious blunder, something like a generation-wide 4060 Ti, and everyone is disappointed. Despite all the Cultists chanting for the Raytracing God and the 28th Coming of the Leather Jacket, people get disgusted and try AMD, and don't massively go back. NV is all about execution, so this is again unlikely. AMD would also need to retain those customers which is more difficult than one might think. I've heard people complain about AMD to the point that Adrenalin was "too complicated and poorly made" despite being a zillion times better than Nvidia's stuff. But people are used to NV and don't like to think about new things, they want a seamless, no thoughts head empty experience. They'll also be encouraged by the Cult to be extra critical of AMD while assuming that anything wrong at NV's is just an unfortunate accident. So we'd need the blunder to be a real one, a whole generation of disappointment, enough that they put real effort in walking away from Nvidia.

B won't happen. C is very unlikely. Which means A is the best bet, just pummel NV's position by showing that you have the Real Power. But even that is a hard sell. You can't just come out with 150 CUs in a 800W card and say "look, best Raster ever". You need the RT, you need a square beatdown with zero doubts that your stuff goes harder than their stuff. And I'm not seeing that happen anytime soon. Even that "RDNA 5 will be the real killer" thing is very much doubtful.

 

[steen2]

Back in 2013, B3D was a really great forum. It started to turn into a nvidia cult circa 2018-2019...

Nope. You forget Nvidia's AEG program, the rot had already set in. AT wasn't immune, either...
Nvidia had a craw in their throat about B3D pre-dating the DP/BB days. B3D discussion was mainly ImgTec/3dfx/Rendition focused while Nvidia's products/practices usually received scorn even back then.

Dave Baumann was an original co-founder of the Beyond3D website (not sure if he co-wrote the B3D Benchmark Suite) who joined AMD in 2006.
He's the guy responsible for making the Radeon HD4850 as popular as it became by pushing for a lower price to drastically increase sales and mindshare. Many years later he was defending what became AMD's GPU pricing strategies of today (just a bit below Nvidia).

The original founder was Kristof Beets with Dave Barron & The Reverend. Wavey Dave, Uttar, Ryszard came later.
Mind you, my recollection ain't what it used to be...

Technical and relevant people like sebbbi leaving didn't help either.

Many devs/industry peeps like Simon F. bailed much earlier.

 

[ToTTenTranz]

Nope. You forget Nvidia's AEG program, the rot had already set in. AT wasn't immune, either...

You know, I remembered this program and have been trying to find the news and discussions about it for ages but if you don't input the "AEG" initials there's absolutely no way that any search engine leads you to it. It's been super frustrating but you gave me what I needed, thanks!

Regardless, it wasn't as bad as it is nowadays. The whole forum has been turned into a nVidia ad since ~2018-19 in the hardware section, while the console section was turning into a Xbox ad. Seeing it go that way was really depressing for someone like me who posted there for like 18 years.

That thing should've stayed dead when BRiT nuked it earlier this year.

Nvidia had a craw in their throat about B3D pre-dating the DP/BB days. B3D discussion was mainly ImgTec/3dfx/Rendition focused while Nvidia's products/practices usually received scorn even back then.

Watching ImgTec engineers talk in the forum about Dreamcast's PowerVR GPU and how it compared to Series2 and Series3 was peak B3D.

Many devs/industry peeps like Simon F. bailed much earlier.

Simon F hasn't posted in like 10 years but it's still listed as a moderator. Reverend seems to appear as banned.

Then either I've mixed up cards either you're the one being ridiculous.

The former.

 

[igor_kavinski]

B won't happen. C is very unlikely. Which means A is the best bet, just pummel NV's position by showing that you have the Real Power. But even that is a hard sell. You can't just come out with 150 CUs in a 800W card and say "look, best Raster ever". You need the RT, you need a square beatdown with zero doubts that your stuff goes harder than their stuff. And I'm not seeing that happen anytime soon. Even that "RDNA 5 will be the real killer" thing is very much doubtful.

Very true. AMD just won't spend that much on R&D for a consumer product.

They are fine with the status quo.

 

[steen2]

You know, I remembered this program and have been trying to find the news and discussions about it for ages but if you don't input the "AEG" initials there's absolutely no way that any search engine leads you to it. It's been super frustrating but you gave me what I needed, thanks!

IIRC, I was contacted 1yr prior via Guru3D forums. Another Guru3D member actually posted the email. I ignored it. The story didn't blow up until it hit B3D, AT, [H], etc.

Watching ImgTec engineers talk in the forum about Dreamcast's PowerVR GPU and how it compared to Series2 and Series3 was peak B3D.

Heh, Gary Tarolli & Simon Fenney discussing the merit (or otherwise) of benchmarketing on usenet trumps anything on B3D, except perhaps some interesting posts by sa.

Reverend seems to appear as banned.

Try a search for why. You'll find it amusing.

Apologies to OP for the OT.

 

[Mahboi]

Yet another "our rumours vs MLID rumours".
When will RDNA 4 come out in the end? Answer in a month. If he turns out correct, I'll be vindicated that it made no sense for pretty much all RDNA 4 drivers to be already in Linux and yet the non-selling RDNA 3 to linger until 2025.

Also his perf predictions are embellished since he's an AMD proponent, but it's very much in the ballpark of our predictions and my earlier napkin math: Roughly between 7900 XT and XTX Raster (he says 4080, which I don't buy), about 4070 Ti Raytracing (and again he says 4070 Ti Super which I also don't buy).

 

[igor_kavinski]

Roughly between 7900 XT and XTX Raster (he says 4080, which I don't buy), about 4070 Ti Raytracing (and again he says 4070 Ti Super which I also don't buy).

Keep an open mind

 

[Mahboi]

Keep an open mind

It's a ~240mm² die. The size of a 6600 XT, from N7 to N4P.
Open mind or not, if they reach above a 7900 XT and get a 4070 Ti's RT, that's already amazing.

 

[igor_kavinski]

It's a ~240mm² die. The size of a 6600 XT, from N7 to N4P.
Open mind or not, if they reach above a 7900 XT and get a 4070 Ti's RT, that's already amazing.

When you put it like that, that's really gonna be a WOW moment for AMD.

 

[Kepler_L2]

When will RDNA 4 come out in the end? Answer in a month. If he turns out correct, I'll be vindicated that it made no sense for pretty much all RDNA 4 drivers to be already in Linux and yet the non-selling RDNA 3 to linger until 2025.

Release date was always a business decision, maybe with Blackwell delays AMD sees an opportunity to get easy market share.

Also his perf predictions are embellished since he's an AMD proponent, but it's very much in the ballpark of our predictions and my earlier napkin math: Roughly between 7900 XT and XTX Raster (he says 4080, which I don't buy), about 4070 Ti Raytracing (and again he says 4070 Ti Super which I also don't buy).

Clock target (as of a few months ago) was 3.46 GHz, which would put N48 at 56.69 TFlops, right inbetween 7900XT (51.48) and 7900 XTX (61.39). Although if MLID is right they missed clock targets yet again, so I would expect perf to land closer to 7900 XT.

As for RT perf, don't expect RDNA4 to do well on NVIDIA-sponsored titles as the HW still isn't as purpose built for RT like Lovelace.

 

[Mahboi]

Release date was always a business decision, maybe with Blackwell delays AMD sees an opportunity to get easy market share.

Mostly I was not buying 2025 from all the drivers being apparently ready and it makes no sense to release for CES.

Look at the November-early Dec spike. That's Christmas effect on buys, and it works every year since the 70s at least. Even if we are in a hard recession (and we are), there's no point holding out for AFTER that. Not unless you're seriously late. Even botched RDNA 3 came out early Dec.

Clock target (as of a few months ago) was 3.46 GHz, which would put N48 at 56.69 TFlops, right inbetween 7900XT (51.48) and 7900 XTX (61.39). Although if MLID is right they missed clock targets yet again, so I would expect perf to land closer to 7900 XT.

Dang, that would've been a really impressive ticking for a GPU.
I'm holding out for a personal copium of 3.3GHz. Because it sounds nice and fast.

As for RT perf, don't expect RDNA4 to do well on NVIDIA-sponsored titles as the HW still isn't as purpose built for RT like Lovelace.

Who in their right mind would expect that? The amount of light bounces is still designed by and for Nvidia, it won't be doing any good in say Cyberpunk. But if it gets out of being poor man's RT and becomes "RT but less good", its already great. Frankly, that "8700 XTX" or whatever it's called will be a gaming monster. 4K is still mostly a waste of money for 3D gaming (I say this sitting in front of my 4K monos) and 1440p is still the prime cost/perf target. If they're going for a cost/perf gen, there's no better target than a top tier 2K card, and this is it. 4070 Ti RT? 7900 xt and above raster? Yep, that's perfect, buy buy buy.

 

[marees]

Who in their right mind would expect that? The mount of light bounces is still designed by and for Nvidia, it won't be doing any good in say Cyberpunk.

I was shocked to discover during the release of the latest wukong game that there is a gimped nv version of UE5. the horror !!!

Soft Lumen is the future ...

 

[SolidQ]

Yep, that's perfect, buy buy buy.

only need right price

 

[Mahboi]

I wouldn't expect much there.
Frankly if they deliver around a 7800 xt's price, it's a good deal. It's a best quality 1440p/low quality 4K card for $500. About as good as we'll get.

 

[marees]

I wouldn't expect much there.
Frankly if they deliver around a 7800 xt's price, it's a good deal. It's a best quality 1440p/low quality 4K card for $500. About as good as we'll get.

If they are launching this "early" then price will be around $580 as RDNA 3 sales have cratered. So 7800xt will still be sitting around on shelves.

 

[SolidQ]

Frankly if they deliver around a 7800 xt's price, it's a good deal.

Mlid predicted they can go to 600$, because need to sell 7800XT/7900GRE