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

 

[ToTTenTranz]

PCI 5.0 x16 is surprise for N44? Also different TBP for 8 and 16GB

https://videocardz.com/newz/amd-radeon-rx-9060-xt-features-32-rdna4-cus-8-16gb-gddr6-memory-and-pcie-5-0x16

And all the HW codecs are there, contrary to previous reports.
It's a full GPU, not an iGPU companion.

x16 g5 roots are pretty dang chungus.

64GB/s duplex towards a DDR5 that provides between 64 and 140GB/s.

If properly implemented, I wonder if AMD could actually make the 8GB model work well enough by tapping into the DDR5 through PCIe for some data.
Vega's HBCC was supposedly able to do this, but AMD went silent on this feature right after launch.

FX 5800 and 9700 fight is back, but in AI now 😁

If they're comparing it to the RTX5080, does it mean MSRP will be $1000?

 

[adroc_thurston]

If properly implemented, I wonder if AMD could actually make the 8GB model work well enough by tapping into the DDR5 through PCIe for some data.

DMA latency is many microseconds which is poop even by GPU standards.
You're getting hammered with stutters on framebuffer spills either way.

 

[Thunder 57]

DMA latency is many microseconds which is poop even by GPU standards.
You're getting hammered with stutters on framebuffer spills either way.

That doesn't explain the 5060 Ti 8GB doing 10% or so better with PCIe 5 vs PCIe 3.

 

[itsmydamnation]

That doesn't explain the 5060 Ti 8GB doing 10% or so better with PCIe 5 vs PCIe 3.

what ?

 

[Thunder 57]

what ?

The 8GB model needs to access system RAM more and over a slower PCIe version it hurts performance:

https://cdn.videocardz.com/1/2025/05/Computerbase-5060Ti-768x501.png

 

[itsmydamnation]

The 8GB model needs to access system RAM more and over a slower PCIe version it hurts performance:

https://cdn.videocardz.com/1/2025/05/Computerbase-5060Ti-768x501.png

yes that's obvious and not what your statements says.

 

[jpiniero]

If they're comparing it to the RTX5080, does it mean MSRP will be $1000?

The 5080 equivalent Quadro's (RTX Pro 4500) MSRP is $2623. So yeah I'd expect it to be at least $1000.

 

[Heartbreaker]

Unfortunately, I presume it is a fake MSRP. At least for some time.

Especially since AMD doesn't have 1st party models this time, to do anything to hold the line on prices.

Big winner if it's ever available at MSRP. If I didn't already have my card. I would happily buy the 9060 16GB if you could get it at MSRP.

 

[SolidQ]

Big winner if it's ever available at MSRP.

there should be aviable soon after launch at MSRP. Some countries already get near MSRP prices for 9070, if we calculate without VAT


Microcenter from yesterday. So many 7900XT in stock

 

[jpiniero]

Microcenter from yesterday. So many 7900XT in stock

If Microcenter sold it online they would have been able to get rid of it a long time ago.

 

[Josh128]

Especially since AMD doesn't have 1st party models this time, to do anything to hold the line on prices.

Big winner if it's ever available at MSRP. If I didn't already have my card. I would happily buy the 9060 16GB if you could get it at MSRP.

$350 is obviously fake as hell MSRP when currently the cheapest 9070XT available on Amazon US is $829. Half the CUs and same VRAM amount will put this card at $429-$499 after the AIBs put their sparkly colors on it. Like the 9070XT, there will be a few thousand MSRP SKUs available in the first 2 minutes of day 1, after that its going to be $429-$499 for the 16GB version.

 

[SolidQ]

cheapest 9070XT available on Amazon US is $829

in different countries, different situation. In some countries it's slowly move to MSRP

 

[DAPUNISHER]

Yeah, here in the U.S. the cheapest 7600 XT I see is $350. No way a newer, faster, 16GB card with better upscaling, is going to sell widely for the same price in the present market.

 

[blckgrffn]

$350 is obviously fake as hell MSRP when currently the cheapest 9070XT available on Amazon US is $829. Half the CUs and same VRAM amount will put this card at $429-$499 after the AIBs put their sparkly colors on it. Like the 9070XT, there will be a few thousand MSRP SKUs available in the first 2 minutes of day 1, after that its going to be $429-$499 for the 16GB version.

Aspirational MSRP

The only positive I am taking from it is this was a staple nvidia marketing tactic and at least AMD is playing with the same tricks. I hate all the tricks but when one was playing straight and the other not so much it made reading reviews grate my nerves. Now if all reviews are like "so, it'll be priced how you find it in the wild, we'll review it for its performance today" at least it should tamp down on some of the "value" editorializing that so often didn't match reality. IMO. Obviously it'd be preferable if we weren't in this consumer hellscape of taking what you can find when you can find it.

*edit - I probably missed it, but when it the review embargo up? This is the magical and therefore not likely to be obtainable $350 card I was hoping for. Maybe. Honestly I just thought it would be at least $399 out the gate.