Discussion RDNA4 + CDNA3 Architectures Thread

Page 347 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DisEnchantment

Golden Member
Mar 3, 2017
1,774
6,757
136





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
Or Phoronix

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

 
Last edited:

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
It'll just end in Nvidia pricecuts.
It's a dead generation (for AMD, that is).
The next one is dead too. And the one after. All until they ship a halo that's 50% more shader cores than Nvidia.
Then they're alive!
RDNA2 showed this approach is a losing one.

RDNA2 didn't sell particularly well. RDNA3 sold better. People saw RDNA2 was a good generation, then ended up buying RDNA3 on launch instead which was mid AF. RDNA3 sales tapered off extremely hard by the end of the generation as people saw how mid the generation was, but still. This is despite RDNA2 being much closer to Nvidia's top end than RDNA3 was.

AMD needs to have a couple of generations where they consistently put out decent stuff worth looking at, then chase the Halo afterwards. Which, from all accounts, seems to be their strategy atm anyway.

The only way to change the rhetoric about Radeon at the moment is to price things decently well. If they don't avoid the "Nvidia -$50" meme this generation, after being every opportunity possible to make this generation an absolute win, then they're screwed in the long term too.
 

adroc_thurston

Diamond Member
Jul 2, 2023
5,493
7,681
96
RDNA2 showed this approach is a losing one.
They did not ship a 50% bigger config than GA102.
AMD needs to have a couple of generations where they consistently put out decent stuff worth looking at, then chase the Halo afterwards
Wroooooong ATi literally exists still because their first real move was The Beheading.
The only way to change the rhetoric about Radeon at the moment is to price things decently well. If they don't avoid the "Nvidia -$50" meme this generation, after being every opportunity possible to make this generation an absolute win, then they're screwed in the long term too.
The meme is just goymers coping about AMD not forcing NV pricing stack down.
And you know it.
 

Dezii90

Junior Member
Jan 9, 2025
17
30
46
Teletubers think amd shouldnt price it over $650.
"Now I know you want to sell that GPU for $650 US, but you have to know that's not a good idea. Deep down you know that will end in failure and you'll have to start over again. So please, learn from your mistakes, get it right the first time and everyone is happy."

I dont think any business will not milk if the competition has no products atm and priced higher...
 

Timorous

Golden Member
Oct 27, 2008
1,920
3,700
136
If AMD want market share they won't get it by getting into a price war and allowing people to buy the NV card they really want at a lower price. They get it by making something that is a lot faster at every price point and by having a halo product NV can't compete with.

There was a spell from RV770 to Hawaii where if AMD has made a 1 tier higher product for each generation then they would have utterly dominated the top tier NV card.

A single die 4970 with 1600 shaders would have crushed the GTX280. And RV770 was just 250mm ish.

A single die 5970 with 2400 shaders would have crushed the late GTX480 and probably GTX580. The top part was just 330mm ISH.

A 6970 was Blackwell tier but an actual top tier part with 2304 shaders would have been a bit faster than the 5970.

The 7970 with 3072 shaders would have destroyed the 680.

A 290X with 4096 shaders in 2013 would have crushed the gtx Titan and 780Ti and we know AMD could build something that big on 28nm because that was the spec of Fiji.

With an aggressive lineup over those years they probably realise the pitfalls of GCN years earlier so we end up with RDNA or similar years earlier as well. AMD would also have a lot more market share and mind share given they would have been. The absolute top dog for at least 5 generations.

Too late now though, AMD missed the boat, NV were able to match AMD performance for all those generations and then when AMD did try and go for the top spot with Fury X NV hit an absolute home run with Maxwell 2.0 and it was not enough and since Maxwell NV have been able to keep that top spot pretty easily, turn it into market share and mind share and are now leaning on software features that are creating a feature disparity that extends beyond pure performance.

So yea, AMD had a ton of opportunity to prevent this and they failed because they were overly cautious and that has directly led to the market we see today.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
It's not a meme. It's a tested way to keep Nvidia from price cuts for an entire year. And if you really don't want to make a lot of Navi 48...
In case you forgot, Radeon are already in the position of making nothing from RDNA3. Margins already look to be pretty goddamn awful.

N48 is the first time where AMD is probably actually fairly competitive from a BOM cost point of view. Lower peak perf than GB203, sure, but basic GDDR6 combined with a slightly smaller die is enough to put them ahead in that regard.

N48 should be cheaper than N32 was to manufacture. They should have some wiggle room for good pricing that doesn't kill their bottom line.
 

adroc_thurston

Diamond Member
Jul 2, 2023
5,493
7,681
96
They should have some wiggle room for good pricing that doesn't kill their bottom line.
They don't, NV has a comfortable margin cushion to force a price war if needed.
N48/44 are worthless products (waste of a really, really cool microarch, unfortunately).
Like any client gfx lineup without a halo part is worthless.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
They did not ship a 50% bigger config than GA102.

They didn't. And they shouldn't, at the moment. They're not really ready to yet. The market wouldn't be interested in a Radeon halo product at the moment.

Radeon has such a bad rep there's no point.

Wroooooong ATi literally exists still because their first real move was The Beheading.

The meme is just goymers coping about AMD not forcing NV pricing stack down.
And you know it.

The real meme is thinking a halo product is enough to suddenly make people change their minds.

Yes, for a lot of people that's all they want - cheaper Nvidia GPUs. For those people, it doesn't matter what AMD does anyway, all they care about is their green GPU. You can't convince those people without consistently putting out something compelling. Doesn't need to be the fastest, just needs to be something that can grab their attention, even if only to cope about.

In the short term, the focus has to be on grabbing people that are sitting on the fence, people that are inclined to give Radeon a try. You'll never do that by offering a product that looks worse at slightly worse prizes.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
They don't, NV has a comfortable margin cushion to force a price war if needed.
N48/44 are worthless products (waste of a really, really cool microarch, unfortunately).
Like any client gfx lineup without a halo part is worthless.

So explain to me where AMD falls behind in terms of BOM then. The die looks to be ~7-8% smaller, GDDR6 is hugely cheaper than GDDR7 and both companies are shipping the same amount of the stuff. AMD didn't even go for full UHBR20 on the display side again just like last gen, no doubt to save on cost. Probably going to be the same split where consumers get UHBR13.5, PRO cards get the full UHBR20.
 
Reactions: Joe NYC and Tlh97

adroc_thurston

Diamond Member
Jul 2, 2023
5,493
7,681
96
And they shouldn't, at the moment
yeah they should, that's their last chance before Lisa cuts dGFX guys altogether.
The market wouldn't be interested in a Radeon halo product at the moment.
It'll be interested in whatever beheads top-end NV part.
Radeon has such a bad rep there's no point.
Radeon had catastrophic rep before R300, young one.
The real meme is thinking a halo product is enough to suddenly make people change their minds.
indeed it does, that's how it works.
For those people, it doesn't matter what AMD does anyway, all they care about is their green GPU
yeah everyone wanted a cheaper GeForce 4 Ti. Until they didn't.
it took years and a goddamn G80 to remove the market inertia R300 created.
In the short term, the focus has to be on grabbing people that are sitting on the fence, people that are inclined to give Radeon a try. You'll never do that by offering a product that looks worse at slightly worse prizes.
no one's giving Radeon a try until they win.
You only win by shipping a part comp refuses to ship.
So explain to me where AMD falls behind in terms of BOM then.
zero volume to coast R&D amort.
It's a completely worthless product, they should've killed the lineup altogether tbf, just like they did with RDNA4.5 APUs.
 

Thunder 57

Diamond Member
Aug 19, 2007
3,489
5,783
136
So explain to me where AMD falls behind in terms of BOM then. The die looks to be ~7-8% smaller, GDDR6 is hugely cheaper than GDDR7 and both companies are shipping the same amount of the stuff. AMD didn't even go for full UHBR20 on the display side again just like last gen, no doubt to save on cost. Probably going to be the same split where consumers get UHBR13.5, PRO cards get the full UHBR20.

He can't. He states things as fact without any proof.
 

Timorous

Golden Member
Oct 27, 2008
1,920
3,700
136
They didn't. And they shouldn't, at the moment. They're not really ready to yet. The market wouldn't be interested in a Radeon halo product at the moment.

As I said up thread, AMD had their chance from RV770 through to Hawaii and they blew it. If they had made halo tier products for those 5 generations they would not be in this position now.

Radeon had catastrophic rep before R300, young one.

Yea. Back before the 9700Pro the 8000 series was regarded very much like Intel Arc is now. Not bad but full of pitfalls. Then suddenly with the 9700Pro with a large performance advantage and the total flop of FX propelled ATi into a competitive position until they dropped the ball with the 2900X and then AMDs reticence to actually make a bigish die (so 500mm region) until Fiji really kicked them in the teeth. Had they actually made 500mm dies from RV770 to Hawaii the market today would be utterly different and AMDs position would be much better.

So explain to me where AMD falls behind in terms of BOM then. The die looks to be ~7-8% smaller, GDDR6 is hugely cheaper than GDDR7 and both companies are shipping the same amount of the stuff. AMD didn't even go for full UHBR20 on the display side again just like last gen, no doubt to save on cost. Probably going to be the same split where consumers get UHBR13.5, PRO cards get the full UHBR20.

They don't but when they cannot supply enough 9800X3D to keep it in stock and that on its own sells for $500 how do AMD justify a $500 9070XT when they are supply limited in another product that has far fatter margins? It would be robbing Peter to pay Paul and it does not look like Paul is going to have any opportunity to pay something back. Especially now NV are differentiating on feature lock in to create a captive market.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
yeah they should, that's their last chance before Lisa cuts dGFX guys altogether.

It'll be interested in whatever beheads top-end NV part.

You are very far from the average gamer.

Radeon had catastrophic rep before R300, young one.

indeed it does, that's how it works.

yeah everyone wanted a cheaper GeForce 4 Ti. Until they didn't.
it took years and a goddamn G80 to remove the market inertia R300 created.

No one's giving Radeon a try until they win.
You only win by shipping a part comp refuses to ship.

Intel's a clear example - if a very extreme one - that price is what it takes to win over gamers. Nobody cares about driver issues or anything else, B580 is seen as a clear winner product for a simple reason: 4060Ti performance at below 4060 pricing.

AMD can pull that off without resorting to the lengths Intel has. They don't need to ship a die twice as large with more VRAM to provide compelling value at an even lower price. I don't think they should be selling 9070XTs for $550. But I don't see how $599-$629 wouldn't be both a huge improvement for Radeon over last gen, whilst simultaneously being a huge step up in popularity.

zero volume to coast R&D amort.
It's a completely worthless product, they should've killed the lineup altogether tbf, just like they did with RDNA4.5 APUs.

And your solution to the issue of no volume to spread out R&D costs, is to laser focus on producing a part you know would have no volume.

Sheer genius.
 

adroc_thurston

Diamond Member
Jul 2, 2023
5,493
7,681
96
You are very far from the average gamer.
Average gamer buys whatever has the brand rep and you could only build the brand rep by winning.
Intel's a clear example - if a very extreme one - that price is what it takes to win over gamers. Nobody cares about driver issues or anything else, B580 is seen as a clear winner product for a simple reason: 4060Ti performance at below 4060 pricing.
no one cares about Arc outside of "their GFX IP will be good enough to force NV pricecuts eventually". Be. Serious.
But I don't see how $599-$629 wouldn't be both a huge improvement for Radeon over last gen, whilst simultaneously being a huge step up in popularity.
one NV pricecut away from irrelevancy.
And your solution to the issue of no volume to spread out R&D costs, is to laser focus on producing a part you know would have no volume.
YESSSSSSSSSSSSSSSSSSS.
when you're behind, you have to gamble! Stake yourself on being bold, and ambitious and ship something you think comp can't. Something that pulls a decisive win on top.
That's how AMD won server CPUs ffs. And client CPUs, really.
That's where Radeon guys have been failing since the goddamn R420.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,637
146
As I said up thread, AMD had their chance from RV770 through to Hawaii and they blew it. If they had made halo tier products for those 5 generations they would not be in this position now.

Well the reality is, they did, and they have to live with a new market now that looks at Intel more favourably than they do at AMD.

Y'all need to stop living in the past and start looking out the window as to what's happening right now.

They don't but when they cannot supply enough 9800X3D to keep it in stock and that on its own sells for $500 how do AMD justify a $500 9070XT when they are supply limited in another product that has far fatter margins? It would be robbing Peter to pay Paul and it does not look like Paul is going to have any opportunity to pay something back. Especially now NV are differentiating on feature lock in to create a captive market.

Best to talk about what features could potentially lock someone out of buying Radeon after we've seen FSR4, and some other stuff that will become clear once the launch is done. All I'll say for now is AMD is attempting to break down that software moat one bit at a time.

But as for the 9800X3D point, we've already established that currently AMD isn't limited by wafer throughput. And given that regular Zen 5 isn't supply limited at the moment, it seems clear that their main bottleneck is SoIC.

Which is totally irrelevant to dGPUs.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |