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

 

[SolidQ]

when AMD doesn't 'release' this card then he'll accuse AMD of lying !!

nah, he's saying it might be not release card. So i think it's fine.

AMD show some graphic vs 7600/XT and 5060ti 8/16

 

[soresu]

MILD is a butthead and I dont understand why people watch him therefore giving him money.

Unless you watch the whole of his super long 3rd person reviews of his latest rumors with random Youtubers it's not going to add much to his Premium revenue, and AdSense revenue only happens when people actually click on the ads, not if they skip or just watch them and do nothing.

That being said yes, I would advocate not adding to it, which is why I would prefer that people do not directly link his videos in this forum.

In this case I only watched long enough to understand where the 9090 number came from, and he gave it away almost immediately, so he's not even good at monetising himself either 😆

 

[soresu]

nah, he's saying it might be not release card. So i think it's fine.

AMD show some graphic vs 7600/XT and 5060ti 8/16

With the increased price of the 9070 series since its initial release I'm probably going to be getting this.

Won't make a huge difference over my 6600 XT in raster gfx, but should give me a nice boost in Blender Cycles when their AMD HIP RT backend is eventually tuned for RDNA4.

 

[SolidQ]

Won't make a huge difference over my 6600 XT in raster gfx

i think ~50% is fine?

 

[soresu]

i think ~50% is fine?

It's OK, but I don't really care that much about it anyway these days vs offline RT/PT rendering.

 

[DAPUNISHER]

MILD is a butthead and I dont understand why people watch him therefore giving him money.

His guests episodes are often excellent. He had Wendell on the other day and it was informative and interesting. I listen while cooking and such. I don't watch the rumor mill stuff.

 

[maddie]

His guests episodes are often excellent. He had Wendell on the other day and it was informative and interesting. I listen while cooking and such. I don't watch the rumor mill stuff.

It's important in this increasingly fantasized world, consuming itself with unproven suppositions and a modeled reality as defined by biased assumptions, to return to basic reasoning at times, as a reality check.

To that end, a reasonably intelligent person would at the very least wonder, "why would these respected industry voices waste hours of their time with him?". Might it be that they see things differently to some here?

The almost herdlike behavior by some to belittle and destroy is quite disturbing.

 

[Heartbreaker]

No Surprises. Very close to 5060 ti.

 

[jpiniero]

No Surprises. Very close to 5060 ti.

At Compubase, there's a bunch of games where it's faster even.

 

[coercitiv]

You know why you came here, here's the verdict! /s

 

[Timorous]

7700XT performance with 16GB of VRAM. Not bad. Lets see how many are available at MSRP.

Performance is exactly what I expected to be honest.

 

[coercitiv]

GN wins the review title award:
AMD Needs to Just Shut Up: AMD Radeon RX 9060 XT 16GB GPU Review

LTT is asking AMD why they did not send them an 8GB to roast properly, the card that most gamers around the world want!

 

[Heartbreaker]

At Compubase, there's a bunch of games where it's faster even.

Well, yeah. This is always the case unless they are in completely different classes. Some games are fairly neutral while other strongly favor AMD or NVidia. So cards in the same class trade blows.

 

[gdansk]

It's a cheaper 7700 XT at a lower price with more VRAM and better upscaling support.

That's not awful, on paper. But I think buyers will consider DLSS worth the price difference that shows up in practice.

 

[Heartbreaker]

It's a cheaper 7700 XT at a lower price with more VRAM and better upscaling support.

That's not awful, on paper. But I think buyers will consider DLSS worth the price difference that shows up in practice.

Unfortunately I don't think you will get either 16GB (5060 Ti or 9600 XT) at MSRP. Both are reasonable at MSRP IMO.

If I was buying to today, I would be happy with either one at MSRP. (I bought an 4070 near MSRP and I will be sticking with that for years to come).

 

[SolidQ]

It's a cheaper 7700 XT at a lower price with more VRAM and better upscaling support.

Rt also a lot faster for small N44 comapre to 7700. 7600 just demolished

 

[blckgrffn]

You know why you came here, here's the verdict! /s

View attachment 124866
View attachment 124867

2.5 stars rom Tom's? I am just guessing now, I'll head over.

Same negative points, that's my bet.

And... no review yet. I guess furrowing my eyebrows at JW's justifications for wonky ratings will have to wait.

 

[Paratus]

Unfortunately I don't think you will get either 16GB (5060 Ti or 9600 XT) at MSRP. Both are reasonable at MSRP IMO.

If I was buying to today, I would be happy with either one at MSRP. (I bought an 4070 near MSRP and I will be sticking with that for years to come).

I got my Asus 9600XT at MSRP. Even came with a free copy of HL2!


Not sure about the 9060XT though.

 

[Ranulf]

From TPU:

"If I was gaming on 1080p only, I'd probably consider the RTX 5060 Ti 8 GB, if it was priced at $350 and below. At its current $380 price point, I'd probably rather just get a $350 RX 9060 XT 16 GB—for the extra VRAM peace of mind, or save a bit more for the RTX 5060 Ti 16 GB. If money is tight, which means you should expect to make compromises, the $300 RX 9060 XT 8 GB could be a viable choice, our upcoming review will have all the details. It will also tell us whether NVIDIA's RTX 5060 8 GB for the same $300 is the better pick. In terms of last-generation cards, the RTX 3070 could be interesting, if you can find one at around $300. A used RX 6800-series card will probably be too expensive, don't bother unless it's $300 too, and be aware of the much lower RT performance. The same is true for RX 7000-series cards. There is no way I'd pick a RX 7700 XT, unless I could get it for $300—the lack of FSR 4 and lower RT perf are the main reasons here. RTX 3080? 10 GB VRAM could be good enough, the wide 320-bit memory helps, you do get DLSS Transformer—maybe if the price is right."

I love how he's gonna do a 8gb card review when they'll be the same price. Its obvious he will favor the 5060 for dlss etc..

 

[blckgrffn]

From TPU:

"If I was gaming on 1080p only, I'd probably consider the RTX 5060 Ti 8 GB, if it was priced at $350 and below. At its current $380 price point, I'd probably rather just get a $350 RX 9060 XT 16 GB—for the extra VRAM peace of mind, or save a bit more for the RTX 5060 Ti 16 GB. If money is tight, which means you should expect to make compromises, the $300 RX 9060 XT 8 GB could be a viable choice, our upcoming review will have all the details. It will also tell us whether NVIDIA's RTX 5060 8 GB for the same $300 is the better pick. In terms of last-generation cards, the RTX 3070 could be interesting, if you can find one at around $300. A used RX 6800-series card will probably be too expensive, don't bother unless it's $300 too, and be aware of the much lower RT performance. The same is true for RX 7000-series cards. There is no way I'd pick a RX 7700 XT, unless I could get it for $300—the lack of FSR 4 and lower RT perf are the main reasons here. RTX 3080? 10 GB VRAM could be good enough, the wide 320-bit memory helps, you do get DLSS Transformer—maybe if the price is right."

I love how he's gonna do a 8gb card review when they'll be the same price. Its obvious he will favor the 5060 for dlss etc..

The mental gymnastics on display there are pretty grand.

So, the real "issue" with either of these new 16GB cards is that you can reliably get a 5070 for under $600 (where I am at) and so when a 5060Ti is already ~$500 street price... what is the point again?





If AMD can really hold the line under $400 retail for the 16GB 9060XT I think it'll be just as successful as the 9070/9060XT, which appear to be doing pretty well and have positive mindshare.

Obviously, AMD should never have made a 8GB version of the "XT" - they could just take the name off and price it still at $300 and it would likely quell a lot of nerd rage. A 9060"XT" should be 16GB, full stop. That makes shopping and comparing so much easier with a clear delineation. Again and as always, IMO.

 

[DAPUNISHER]

Rich's review is excellent within its scope. For games like 2077 his test loop is too short and in the wrong area to expose how bad 8GB is. Probably a couple of other games with similar caveats. Still, I am liking Rich's approach and the way he shows his bench passes; even if most are automated. 8GB repeatedly 💩 the bed when 16GB models are gliding along. Interestingly, it depended on the pass as to whether the 5060ti or 5060 would be the one to go belly up. It's like playing craps...

 

[Timorous]

Rich's review is excellent within its scope. For games like 2077 his test loop is too short and in the wrong area to expose how bad 8GB is. Probably a couple of other games with similar caveats. Still, I am liking Rich's approach and the way he shows his bench passes; even if most are automated. 8GB repeatedly 💩 the bed when 16GB models are gliding along. Interestingly, it depended on the pass as to whether the 5060ti or 5060 would be the one to go belly up. It's like playing craps...

Their automation is along the lines of selenium UI interaction which mimics human input. It allows them to automate without using built in benchmarks.

 

[Kronos1996]

So GB206 is about 10% smaller vs Navi 44 while being slightly more efficient and having better RT. All else being a wash. I guess Nvidia still has a slight PPA advantage, GDDR7 can’t account for all of that.

 

[marees]

So GB206 is about 10% smaller vs Navi 44 while being slightly more efficient and having better RT. All else being a wash. I guess Nvidia still has a slight PPA advantage, GDDR7 can’t account for all of that.

You are sounding like Jarred Walton 😀

 

[SolidQ]

So GB206 is about 10% smaller vs Navi 44 while being slightly more efficient and having better RT. All else being a wash. I guess Nvidia still has a slight PPA advantage, GDDR7 can’t account for all of that.

Hard to say who have advantage, because it's GDDR6 vs GDDR7.