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

 

[Schmide]

It's been like this since two or three years ago, just set DX9 inside zoom and you'd be good to go
View attachment 122448

All I have to do is uncheck hardware acceleration receiving data. It hasn't been a problem for me till now.

It doesn't present itself as garbled video, it stutters. I looks like network errors. First thing I did was check the network. Reboot router etc.

Some streams were fine. My friend on an iPhone was perfect. I think the freezing video streams were HP laptops.

In the notes

Lower than expected performance may be observed when using YouTube on the Microsoft Edge web browser immediately after a driver install. Users experiencing this issue are recommended to restart Microsoft Edge and relaunch YouTube as a temporary workaround.

There is something wrong with the drivers, you don't pay $800 to use software rendering. If that was the case I'd use the integrated graphics and call it a day.

 

[gdansk]

If that was the case I'd use the integrated graphics and call it a day.

That's another possible workaround. You can specify per application GPU. Assign zoom to your iGPU for hardware acceleration.

But AMD is slower about fixing problems, and releases in general. When people were pointing out all the Blackwell problems I noted that Nvidia has a good reputation because usually they fix things quickly (comparatively). Not necessarily because they're problem free (as that hadn't been my experience with Turing or Ampere at launch).

 

[Hail The Brain Slug]

Some streams were fine. My friend on an iPhone was perfect. I think the freezing video streams were HP laptops.

This sounds like its an out-of-spec video stream issue. Apple tends to be the leader with regards to having the most in-spec encoding, and also the most robust player that can handle lots of out-of-spec issues playing back video so it stands to reason youd be receiving a less error/quirk prone stream from iphones.

In a way, if its due to the video stream being out of spec on some platforms, its not exactly AMD's fault.

However, the reality is nearly every streaming video will have errors or quirks that put it out of specs, so your decoder being capable of gracefully handling them is important.

I'm not attempting to absolve AMD here, their new codec engine on RDNA4 is just not going to be as mature as previous gen, so it likely doesn't have the same level of robust handling of these issues yet. I just wanted to share this possibility to say it's likely not a straightforward "AMD has a bug and isn't fixing it" issue like you seem to want it to be.

 

[Schmide]

This sounds like its an out-of-spec video stream issue. Apple tends to be the leader with regards to having the most in-spec encoding, and also the most robust player that can handle lots of out-of-spec issues playing back video so it stands to reason youd be receiving a less error/quirk prone stream from iphones.

In a way, if its due to the video stream being out of spec on some platforms, its not exactly AMD's fault.

However, the reality is nearly every streaming video will have errors or quirks that put it out of specs, so your decoder being capable of gracefully handling them is important.

I'm not attempting to absolve AMD here, their new codec engine on RDNA4 is just not going to be as mature as previous gen, so it likely doesn't have the same level of robust handling of these issues yet. I just wanted to share this possibility to say it's likely not a straightforward "AMD has a bug and isn't fixing it" issue like you seem to want it to be.

In the same vernacular. AMD is unable to handle edge cases so they are out of spec.

You can use the software version of the codec and it works. You can use a different card and it works. (yes I have more than one AMD card and a couple nvidias)

There is one odd player out and it's 9070 xt with AMD's drivers.

As posted above.

AMD's own drivers work, just not with the 9070 xt. You can't blame the other streams and call quirks if everyone handles it but AMD's 9070 xt. That's diverting and not productive.

They're having trouble with edge and youtube. (see post 3 above) Though there we can definitely blame anyone who uses edge.

What is this innate need to shift the blame? I personally think this hyper defensive posture hurts AMD's reputation. They are top 3, they should handle even the most *edge cases gracefully. They should aggressively log and categorize bugs. They've been around for 55 years (ATI 39). Though I am older. Edit: actually they got me by a month.

* yes even the edge browser. Have you tried edge its based on chrome.

 

[Hail The Brain Slug]

In the same vernacular. AMD is unable to handle edge cases so they are out of spec.



As posted above.

AMD's own drivers work, just not with the 9070 xt. You can't blame the other streams and call quirks if everyone handles it but AMD's 9070 xt. That's diverting and not productive.

They're having trouble with edge and youtube. (see post 3 above) Though there we can definitely blame anyone who uses edge.

What is this innate need to shift the blame? I personally think this hyper defensive posture hurts AMD's reputation. They are top 3, they should handle even the most *edge cases gracefully. They should aggressively log and categorize bugs. They've been around for 55 years (ATI 39). Though I am older. Edit: actually they got me by a month.

* yes even the edge browser. Have you tried edge its based on chrome.

I apologize for attempting to ensure you at least have a single clue what you're talking about before you fly off the handle on an internet forum. My bad.

Continue being an ignorant piece of work.

 

[Schmide]

I apologize for attempting to ensure you at least have a single clue what you're talking about before you fly off the handle on an internet forum. My bad.

Continue being an ignorant piece of work.

If you're going to call me "an ignorant piece of work" you should reflect on your positions and think about the motivation behind them? None of us are insiders. You artificially praised apple disparaged HP and defended AMD all in the same post. But its ok if you say

I'm not attempting to absolve AMD here

 

[Schmide]

Just to be clear. This Friday with the same streaming participants I will have.

The original computer Windows 10 that had problems with the 9070 xt driver cleaned and forced zoom update.

A fresh install linux computer with a 9070 xt in it.

A fresh install windows 11 computer with a forced zoom update. 7900xtx. I may switch the 9070 xt from the linux computer to it.

On hand I will have a 6700xt, 5700 and 3070 ti along with a 4060 laptop with rembrant+ switchable graphics to maybe switch.

You can't be ignorant if you're willing to put in the leg work. (never skip leg day)

 

[Shmee]

Please get along all. Thanks.

 

[SolidQ]

https://videocardz.com/newz/amd-radeon-rx-6500-non-xt-spotted-features-1024-cores-4gb-64-bit-memory-and-55w-tbp

Interesting what AMD going is announce today

 

[poke01]

https://videocardz.com/newz/amd-radeon-rx-6500-non-xt-spotted-features-1024-cores-4gb-64-bit-memory-and-55w-tbp

Interesting what AMD going is announce today

Now this is irrelevant to this thread 😝



(Cause it’s not RDNA4 related)

 

[Gideon]

a LOOOONG but quite interesting post on AMD raytracing patents:

https://www.reddit.com/r/hardware/comments/1k4e14o/analysing_amds_ray_tracing_patents

 

[soresu]

a LOOOONG but quite interesting post on AMD raytracing patents:

https://www.reddit.com/r/hardware/comments/1k4e14o/analysing_amds_ray_tracing_patents

They could even have a viable competitor to ....... and a ReSTIR path tracer.

ReSTIR despite being funded by nVidia is open tech that is well documented.

Not sure what they mean here, unless they are talking about renderer sample code for developers to use so that they aren't anchored to nVidia's own implementation a la RTX Remix.

There is mention of "reuse" and "spatiotemporal" in the "SPATIOTEMPORAL ADAPTIVE SHADING RATES FOR DECOUPLED SHADING" patent description, but that's about it

 

[marees]

N44 launch around Computex in June

N48 12gb launch (worldwide only or Chinese too?) indefinitely postponed

AMD 被曝 RX 9060 XT 显卡 5 月 18 日发布，RX 9070 GRE 延后 - IT之家

RX 9070 GRE 显卡原计划 5 月 8 日上市，目前暂无准确发售时间，可能会推迟到四季度。

m-ithome-com.translate.goog

https://twitter.com/x/status/1914904114523435108

 

[SolidQ]

N44 launch around Computex in June

In twit, 18th may
AMD should rename 9600XT 8gb to 9600, and sell for 250$ max. We saw already how bad 5060ti 8gb

 

[marees]

In twit, 18th may
AMD should rename 9600XT 8gb to 9600, and sell for 250$ max. We saw already how bad 5060ti 8gb

From opera translate:

According to news on April 23, channel source Bobantang said today that according to information obtained from AMD internally, the Radeon RX 9060 XT graphics card will be released on May 18, the eve of the COMPUTEX 2025 Taipei International Computer Show; it was originally planned to be launched on May 8 ​

​

The launch of the RX 9070 GRE graphics card will be postponed, and there is no accurate release time yet, may be postponed to the fourth quarter​

My interpretation is that reveal is in May & Launch in June

 

[SolidQ]

from MLID. Seems reasonable to me make 8gb as 9050XT and sell itfor 200-230$ max.

 

[Ranulf]

AMD could win some support if they put out a 9050XT at a good price.

 

[marees]

I doubt AMD is making such drastic last minute changes. Lets see

 

[GodisanAtheist]

from MLID. Seems reasonable to me make 8gb as 9050XT and sell itfor 200-230$ max.

-This honestly sounds like a ridiculous rumor.

Why would AMD not launch an 8gb 9060xt? Nvidia is asking $400 for their 8gb card and it *will sell*.

$299 9060xt 8gb absolutely will sell in this tits up market.

If there is a $229 9050xt with like 28 CU or something below it, even better.

 

[gdansk]

$299 9060xt 8gb absolutely will sell in this tits up market.

RTX 5060 is "$299". So N44 8GB ought to be "less".

 

[Timorous]

-This honestly sounds like a ridiculous rumor.

Why would AMD not launch an 8gb 9060xt? Nvidia is asking $400 for their 8gb card and it *will sell*.

$299 9060xt 8gb absolutely will sell in this tits up market.

If there is a $229 9050xt with like 28 CU or something below it, even better.

Because a better alternative is a 96bit 12GB 9050XT.

That could easily sit in the $250-$300 range with 6700XT perf or thereabouts and do very very well. We saw this with the B580 reception.

The reason not to sell an 8GB part is two fold, 1) it will get slated in the press unless it is sub $250 at best. 2) it is already obsolete and is a bad product.

I know a lot of people have said a 96bit 12GB part would be slower than an 8GB 128bit part in the titles that are not Vram bound, correct it would be and despite that, due to how Vram bound 8GB can be a 96bit 12GB part would still be the better product for the latest games and upcoming games because it won't require the level of settings tuning an 8GB part will.

 

[Win2012R2]

Why would AMD not launch an 8gb 9060xt?

Because when Nvidia gets big flak for doing it, then AMD will get 10 fold.

8 GB should have died a death at least 2 years ago, minimum should be 16 GB, it costs very little over - what's there to argue about, 25-30 bucks and it's a proper card.

 

[TESKATLIPOKA]

Because a better alternative is a 96bit 12GB 9050XT.

That could easily sit in the $250-$300 range with 6700XT perf or thereabouts and do very very well. We saw this with the B580 reception.

The reason not to sell an 8GB part is two fold, 1) it will get slated in the press unless it is sub $250 at best. 2) it is already obsolete and is a bad product.

I know a lot of people have said a 96bit 12GB part would be slower than an 8GB 128bit part in the titles that are not Vram bound, correct it would be and despite that, due to how Vram bound 8GB can be a 96bit 12GB part would still be the better product for the latest games and upcoming games because it won't require the level of settings tuning an 8GB part will.

This 96-bit 9050XT 12GB is not a better alternative than 9060XT 128-bit 8GB.

By having only 96-bit wide controller, It will have even less BW than RX 7600(240 vs 288 GB/s) and I think this will also reduce available Infinity cache from 32MB -> 24MB, and I am not sure If ROPs are decoupled or not from the memory controller.
You will clearly lose a lot of performance while BOM won't be lower than 9060XT 8GB and then there is still the question how long will 12GB be good enough.

To me both of them are not good products, one will sacrifice performance in current games and the other one in the future games.
I would pay a few bucks more for the full 9060XT 16GB to be honest.

 

[TESKATLIPOKA]

Because when Nvidia gets big flak for doing it, then AMD will get 10 fold.

8 GB should have died a death at least 2 years ago, minimum should be 16 GB, it costs very little over - what's there to argue about, 25-30 bucks and it's a proper card.

12GB should be the new minimum. I don't really mind If there were still some 8GB cards, I myself could live with It for some time thanks to my Steam collection, but the price needs to be reasonable and 5060Ti 8GB's price is not reasonable at all.

 

[waffleironhead]

People are still buying new 4GB cards, there is no reason to believe a properly priced 8GB card wont sell.
A sub 150 watt 6 pin 8GB card would sell just fine. Obsolete or not.