Goodbye NVIDIA. – AMD RDNA3 Announcement

[AlexTheGreatish]

AMD just announced their RDNA3 GPUs and dang.. it looks amazing for gamers.

 

 

 

[Senzelian]

AMD this generation: We are not retarded. Our GPUs just make sense.

Consumers: Wooohooo! GO AMD!

[DoctorNick]

Damn I would like me some 7900XT, even if I don't need it.

[hishnash]

it is an interesting approach AMD have gone for, doing McM but just for cache/mem controllers allows them to get the savings of die area cost without the complexities of dealing with multiple compute dies.   but the media calling it the first consumer McM gpu is not correct as the GPU is not McM and even if it were it is not the first, apples M1 Ultra gets that label as it truly has gpu compute on both dies, in fact each die is made out of 4 seperate gpus all working in tandem

[Error 52]

Sick of these shoddy production errors from LTT. It lists "Modern Warfare 2" as a benchmark, which is the wrong game! It's meant to be "Modern Warfare II"! Totally different!

 

Sarcasm aside (that was sarcasm, for the record) it's unfortunate that RT performance and productivity is going to continue to be a thorn in AMD's side, but otherwise this is a really damn compelling GPU for the money. The XTX anyway, the XT maybe a little less so. Hopefully AMD's lower-end cards are able to also chokeslam Nvidia price-to-performance wise as they roll out.

[DarkSwordsman]

2 minutes ago, Error 52 said:

it's unfortunate that RT performance and productivity is going to continue to be a thorn in AMD's side

Agreed, but I am at least happy that AMD is taking advantage of Nvidia's silliness by just selling decent, normal GPUs that are relatively affordable, rather than trying to compete specifically to top them.

[hishnash]

3 minutes ago, Error 52 said:

Hopefully AMD's lower-end cards are able to also chokeslam Nvidia price-to-performance wise as they roll out.

the solution of separating out the cache and memory controllers means we might well see the same compute die used for the entier product stack, with lower and lower end models having lower yields of compute dies and less cache/meme controler dies so less bandwidth this could save AMD a lot of money compared to needing to create yet another design for the smaller die. 

[TJB]

They are tempting me away from my 3060ti, but ive heard bad things about their drivers, and as others have said, their RT performance is an issue for them. Will need to see some proper in depth benchmarks.

[JogerJ]

NVIDIA is still king in machine learning use cases, unfortunately. I'm really hoping AMD could catch up in that area, but at least with cheaper competition hitting the markets hopefully NVIDIA would cut prices sooner

[ForboJack]

1 hour ago, TJB said:

They are tempting me away from my 3060ti, but ive heard bad things about their drivers, and as others have said, their RT performance is an issue for them. Will need to see some proper in depth benchmarks.

As Someone who has used an AMD RX 6900 XT for the past 1.5 years now, I can say that their drivers got so much better in that time. Not gonna lie, it was pretty ruff over that time and I had some huge issues with my drivers. But for the past 6 months they have been super stable. While in the past I had regular errors and even system crashes, I basically had no major issues recently. The only Problems I still have is performance in Fortnite. The FPS is just all over the place in that game. No problems with other games though. I especially love how easy it is to overclock your GPU in the Adrenalin software. 

Before that card I used Nvidia cards for like the past 15+ years and their drivers weren't without problems either. So of course waiting for some real reviews and benchmarks is always a good idea, but I think AMD is really committed to fix and improve their software. So I wouldn't say this is a no go anymore for potential buyers.

[Luscious]

This new product announcement makes two things here very clear:

 

1. EK absolutely cannot ignore team red any more. They have been in bed with nVidia for years now. It's time for them to make water blocks for these cards.

2. EVGA needs to bring AMD on board and make Hydro Copper versions of these cards. I would buy four of them in a heartbeat

 

And there is a third if you really go between the lines here - nVidia totally went down the wrong road with their 4090. Melting connectors, going outside the ATX size spec, blatant overpricing, abusing their leadership position... It really was only a matter of time before their reign fell.

[Marmelade-Maki]

So with the 4080 you get

• Less memory (eventhough faster)
• Less performance (if the XTX comes as close to the 4090 as people expect)
• less connectivity (Display Port 1.4)
• the same size as a 4090 (according to Nvidia). Why though? It has the same Power consumption as the 3090 Ti or the XTX

That's going to be a hard sell. The 4090 flew off the shelfs, even if we don't know how many were actually in the wild. But the 4080 might be more difficult to sell considering it's lacking the insane numbers of the flagship. Seriously, if i wasn't so CUDA bound in a lot of stuff that i do outside of gaming i would jump over.

[Man]

4 hours ago, Senzelian said:

AMD this generation: We are not retarded. Our GPUs just make sense.

Consumers: Wooohooo! GO AMD!

There's no such thing as bad hardware, only bad pricing. 

 

Pretty sure we all would be seething with range; had AMD set the MSRP of 7900XTX at anything beyond $1,000.

 

 

[Man]

While I'm far from being an Nvidia fanboy (current GPU is an RX6600 while previous one was an R9-380), I think AMD can't quite compete with Nvidia on the MSRP front... yet. 

 

RDNA2 has inferior RT and on my vanilla RX6600, it's practically a gimmick. And I very much doubt RDNA3 is going to finally catch-up with Ampere - let alone Ada - when it comes to RT.

 

Then there's the matter of DLSS. As much as I love the open-source nature of FSR 2.0, it just can't compete with DLSS. And FSR 3.0's quality and performance remain to be seen.

 

I'm not holding my breath, suffice to say. 

[Man]

19 minutes ago, Marmelade-Maki said:

So with the 4080 you get

• Less memory (eventhough faster)
• Less performance (if the XTX comes as close to the 4090 as people expect)
• less connectivity (Display Port 1.4)
• the same size as a 4090 (according to Nvidia). Why though? It has the same Power consumption as the 3090 Ti or the XTX

1. I despise (hate is a strong word) GDDR6X with a passion. It just runs too bloody hot for its own good. I'm glad AMD hasn't bothered with it. 
2. Different architectures have different memory bandwidth requirements. 
3. I very much doubt the XTX is going to come anywhere near close to the 4090, given the amount of shader cores and their core frequency. Ada is touching nearly 3GHz while RDNA3 appears to be capped at around 2.5GHz, for whatever reason. 
4. DP1.4 doesn't matter to 99.99% people out there. No one in their right mind would bother with an 8K 120Hz display. 
5. The 4090 is a halo product. So obviously, rich blokes are going to line-up to get their mitts on one. It's all about bragging rights.
6. AMD's flagships rarely manage to match NVidia's finest. Why? Because - apparently - AMD doesn't like to make large dies. It's not like they can't design a 600mm² die, they absolutely can. It's just that they've never bothered with it... for some reason. 

It's been like that ever since I remember. In my late teens, the GTX280 was all the rage. The fastest bloody GPU in the world with an enormous (for the time) GT200 die @ 576mm² and a humongous 512-bit wide GDDR3 vRAM. 

 

Meanwhile, AMD was taking pot-shots at the "Yamato class" GTX280 with their much, much smaller 282mm² RV790 - in the guise of HD4890 - with a teeny, tiny 256-bit GDDR5 and a much lower price tag. 

 

That's AMD/ATI's strategy for the past decade!

[BuckGup]

4 hours ago, hishnash said:

it is an interesting approach AMD have gone for, doing McM but just for cache/mem controllers allows them to get the savings of die area cost without the complexities of dealing with multiple compute dies.   but the media calling it the first consumer McM gpu is not correct as the GPU is not McM and even if it were it is not the first, apples M1 Ultra gets that label as it truly has gpu compute on both dies, in fact each die is made out of 4 seperate gpus all working in tandem

The M1 wasn't the first by far either. My profile pic is the die shot of an MI250. Which is truly 2 GPUs slapped into one die 

[kumicota]

1 hour ago, Man said:

AMD's flagships rarely manage to match NVidia's finest. Why? Because - apparently - AMD doesn't like to make large dies. It's not like they can't design a 600mm² die, they absolutely can. It's just that they've never bothered with it... for some reason. 

 

Big dies are a PiTA to make and manufacture, when you're doing chips for something(besides the maintenance and pre-production cost), you basically pay it by wafer, for example a 300mm² wafer 5n from TSMC is roughly about 17k per wafer, besides that you have about 80% of yield per wafer, so you basically lose 20% of the chips on every 17k wafer that you buy and this yield rate decreases more if you ask for bigger dies.

 

Another problem with wafers is that they are round and this yield don't take it in consideration(IIRC), so you lose more a few chips that can't be made because it's on the corners of a circle. This make that Nvidia big monolithic dies to be expensive and hard to make, which is why AMD and others is pushing to MCM and chiplets(which has it's own series of drawbacks too) designs, because you retain the easier manufacturing of smaller dies but you have almost the same performance as a monolithic design for less.

[Beerzerker]

1 hour ago, Luscious said:

This new product announcement makes two things here very clear:

 

1. EK absolutely cannot ignore team red any more. They have been in bed with nVidia for years now. It's time for them to make water blocks for these cards.

2. EVGA needs to bring AMD on board and make Hydro Copper versions of these cards. I would buy four of them in a heartbeat

 

And there is a third if you really go between the lines here - nVidia totally went down the wrong road with their 4090. Melting connectors, going outside the ATX size spec, blatant overpricing, abusing their leadership position... It really was only a matter of time before their reign fell.

EVGA is getting out of the GPU business period. EVGA never had a relationship with AMD and even if they were still wanting to make GPUs they'd probrably never do business with them.
That's just the business model they went by and since they are now bailing on GPUs all together, ain't happening for sure.

[themrsbusta]

Theory: They don't put a general purpose AI on the GPUs because of the AVX-512 on the Ryzen 7000 lineup.

[Middcore]

LTT will still continue to exclusively use Nvidia GPU's in all their build videos. 

[ToboRobot]

Why does LTT not upload a regular video on the day of launch events, and then rush to release the launch video the next day? 

[Ubersonic]

The fact they only talked about gaming not compute is a good thing IMO.  The last thing the world needs right now is for AMD to launch a GPU with insane mining performance.

[Zando_]

4 hours ago, Luscious said:

EK absolutely cannot ignore team red any more. They have been in bed with nVidia for years now. It's time for them to make water blocks for these cards.

Have they been ignoring them? When I got my Radeon VII only EK had a good block for it, was originally waiting for a Heatkiller but I don't know if those ever ended up being made or no. 

3 hours ago, Beerzerker said:

EVGA never had a relationship with AMD and even if they were still wanting to make GPUs they'd probrably never do business with them.

They do. They make AM4 (chipset X570) motherboards, I don't know if they have any AM5s out but they do have a relationship with AMD on the mobo-side for at least one modern gen of boards. They also made AMD boards back when AM2/Nvidia chipsets were a thing. But yeah, they said they don't intend to make GPUs for anybody anymore, so very little chance of seeing an EVGA RX card unless they get desperate and/or AMD is actually better to work with than Nvidia. 

[papajo]

10 hours ago, AlexTheGreatish said:

AMD just announced their RDNA3 GPUs and dang.. it looks amazing for gamers.

 

 

 

As far as 3:53 of the video how stupid does Jensen think that we are?...  

1) Moore's law is kind of dead for the CPUs... GPUs are far simpler logic and different (bigger) die sizes and power budgets... for them Moore's law is pretty much very real! 

2) How can a company claim that the prices are like this for things out of their control (so money doesnt go into their pockets) but constantly increasing its REVENUE multiple times since the first time it launched GPUs with stupid prices like 1000 usd per piece? (so about 2018) 

I mean if the prices were such because shortages (they dont make as much as they should cause the factory cant produce) or other reasons (tarrifs, taxes etc) they should make the same money  as they used to make more or less...  yet on 2019 they had DOUBLE the profits they had on 2016 and on 2022 they made x3 (THREE TIMES) the profit they made on 2019

And we are talking about NVIDIA so no man in the middle no notorious scalpers buying the stock etc... they increased their profits directly from selling the CHIPs (after paying TSMC for producing them!) 

Like I really would like linus to ask Jensen in a video interview or like say a forum user Papajo -or in such a case I can give my real name etc-  asks "How much stupid you really think we are Jensen?" 

[Motifator]

8 hours ago, Error 52 said:

Sick of these shoddy production errors from LTT. It lists "Modern Warfare 2" as a benchmark, which is the wrong game! It's meant to be "Modern Warfare II"! Totally different!

 

Sarcasm aside (that was sarcasm, for the record) it's unfortunate that RT performance and productivity is going to continue to be a thorn in AMD's side, but otherwise this is a really damn compelling GPU for the money. The XTX anyway, the XT maybe a little less so. Hopefully AMD's lower-end cards are able to also chokeslam Nvidia price-to-performance wise as they roll out.

Haven't you seen? It's actually neither Modern Warfare 2 nor Modern Warfare II. It is Modern II Warfare. Whoever came up with the ideas of spec ops looking like Rob Zombie movie characters and this abbreviation needs to be bagged...