RDNA 3 rumours Q3/4 2022

[humbug]

@brostradamus believe it or not, replicating what Nvidia are doing in a different way is the easy bit.

AMD's main focus is to do what so far no one else can, to cheat Moore's law, that's difficult.

 

[Darujhistan]

350W maybe but I think 300W will not be enough power to match the 4090.



The 7600XT is supposed to be close to the 6900XT at 1080p. 7700XT which is using cut N32 with probably 5kish shaders should be a bit faster than the 3090Ti.

This may be one of the reasons the 4080 12GB was unlaunched because NV know that AMD are going to be releasing a 7700XT 12GB with similar / slightly better performance and it would look bad for AMDs 3rd tier to match the 4080 even if it is the castrsted one.

That might be a decent card as the 6700XT relative to Nv's last gen was somehwere around the 2080.

I'd buy something that matched 3090 or bettered it, but I'm not buying something that's only 1 tier up like 6800 performance. Not worth it. I might even just sit out the gen out altogether though.

 

[humbug]

That might be a decent card as the 6700XT relative to Nv's last gen was somehwere around the 2080.

I'd buy something that matched 3090 or bettered it, but I'm not buying something that's only 1 tier up like 6800 performance. Not worth it. I might even just sit out the gen out altogether though.

Its a bit more than that...

Between a 3060Ti and a 3070.

 

[Darujhistan]

Its a bit more than that...

Between a 3060Ti and a 3070.

Yeah, as far as I'm concerned it's basically a 2080ti [the 4% means nothing to me] , I misrembered and thought it was equivalent to the 2080.

 

[gpuerrilla]

what i meant was

Don't be a goalpost mover. Just move on its pointless debating when the moment passes eh.

@brostradamus believe it or not, replicating what Nvidia are doing in a different way is the easy bit.

AMD's main focus is to do what so far no one else can, to cheat Moore's law, that's difficult.

Yep. Get the context..

 

[brostradamus]

@brostradamus believe it or not, replicating what Nvidia are doing in a different way is the easy bit.

AMD's main focus is to do what so far no one else can, to cheat Moore's law, that's difficult.

yeah thats one way to look at it, but i have very strong conviction about dlss being the way forward and it would need some risk taking and intent. lets see how things move, interesting stuff as always

 

[EastCoastHandle]

AMD is now using a MCM approach to GPU gaming. A 1st that couldn't be achieved before because developers nor MS had a way to utilize mcm gpus for gaming at the time of writing this post.
So, AMD was able to make their MCM be seen as a single core in windows (so I've read).

The performance uplifts are going to be well above 2x. How much remains be seen. Making it clear that 7900xt should beat the 4090.
I find it hard to believe that AMD performance uplift with a MCM is about the same as if they went with a regular single die approach. Think about it for a minute. AMD knows the performance of the 4090 and potential 4090 ti.
Yet they still want to release a 7900xt and not the 7950xt (if they thought that the 7900xt would lose to the 4090).

I don't know man, it seems to me that these MCMs might pack more of punch then what's rumored and AMD has had no real leaks regarding RDNA 3 like they did the RDNA 1 and 2. There is certainly something peculiar this go around.

 

[humbug]

AMD is now using a MCM approach to GPU gaming. A 1st that couldn't be achieved before because developers nor MS had a way to utilize mcm gpus for gaming at the time of writing this post.
So, AMD was able to make their MCM be seen as a single core in windows (so I've read).

The performance uplifts are going to be well above 2x. How much remains be seen. Making it clear that 7900xt should beat the 4090.
I find it hard to believe that AMD performance uplift with a MCM is about the same as if they went with a regular single die approach. Think about it for a minute. AMD knows the performance of the 4090 and potential 4090 ti.
Yet they still want to release a 7900xt and not the 7950xt (if they thought that the 7900xt would lose to the 4090).

I don't know man, it seems to me that these MCMs might pack more of punch then what's rumored and AMD has had no real leaks regarding RDNA 3 like they did the RDNA 1 and 2. There is certainly something peculiar this go around.

I don't know about the performance of RDNA3 and i wouldn't get ahead of my self.

But i can expand on what i said earlier, we are getting to a point now where each die shrink offers less and less over the previous generation, a slowdown in More's law.

So the problem is to make more CPU or more GPU you have to keep making them bigger, look at the last few generations of Nvidia GPU's, aside from the cost of that you also have the laws of physics, the more transistors you pack in to these things the more electrical resistance you have, that drives up power consumption, again look at Nvidia, and Intel.

Now, wouldn't it be nice if you could split your big die up in to lots of little ones, your wafer yields would go up, they wouldn't have the electrical resistance problem so they are nice and efficient, like a small die chip, and if you do it right, like designing them to be modular, like logo, you can scale them infinitely, so the limits of how much CPU or GPU you can make no longer applies, the limit for how many cores you can have in a CPU is limited only by how big the PCB you glue them to is, 64 cores? no problem, 96? yeah, 128? done that..........
Can you imagine how big a monolithic CPU would need to be to accommodate 64 cores? Its why Intel can't do it, not even close, AMD have doubled that and they will double it again with Zen 5!

The problem with all that ^^^^ is there are A LOT of technical hurdles to overcome to make it work, AMD are the first to have done it, they might be the only ones....

 

[humbug]

If Intel and Nvidia aren't absolutly crapping themselves they are idiots.

 

[HACO]

Coming back down to earth, if the MCM approach proves effective there's no reason Intel and Nvidia wouldn't end up doing the same thing down the road.

 

[EastCoastHandle]

I don't know about the performance of RDNA3 and i wouldn't get ahead of my self.

To not know doesn't make it a positive nor a negative but remains as a unknown variable. So there is no getting ahead of the rumor curve to be had. However, to me, it makes sense that the performance uplifts are going to be more efficient thus greater then with a single die solution. It's clear we don't know what the performance is going to be. But IMO I believe it's going to be greater the 2x.

Coming back down to earth, if the MCM approach proves effective there's no reason Intel and Nvidia wouldn't end up doing the same thing down the road.

You really think that greater then 2x is beyond earth? I don't. And, it's not hard to understand why nv didn't do the same thing. You are under a false conception that AMD and Nv are the same and they are not. Their Uarch to Gpus are different. Not everything is the same because "your favorite company" is doing xyz. So come back down to earth, nv didn't do the same as AMD because they are on different Uarch roadmaps.

 

[Joxeon]

i believe dlss is going to be th e only path towards rendering efficiency as we start targeting higher than 4k resolutions, there are various approaches being tried atm and i willing to bet on this technique becoming the de-facto rendering method for future games


its nowhere close to CUDA though, the CUDA feature set dwarfs ROCm.. and AMD cards dont do matrix FMA in hardware


the point i was making is nvidia's on top of the innovation game, amd has been a follower - eventually amd will get there, but then nvidia would have moved on to something else. theres a big difference in company DNA. nvidia also employs many subject matter experts that have nothing to do with hardware but rather focus on the end user and industry problems that could be called deep-in-domain

I actually think MCM will be the future and will bring big gains like we have seen for CPUs, DLSS while ok is just a bandaid right now to help when turning on RT, as RT and raster improve there is less need for DLSS.

What nvidia is good at is creating problems and then selling you the solution, once RT is overcome then nvidia will be releasing next tech to make your current cards struggle so you have to buy their shiny new ones.

 

[Chuk_Chuk]

AMD is now using a MCM approach to GPU gaming. A 1st that couldn't be achieved before because developers nor MS had a way to utilize mcm gpus for gaming at the time of writing this post.
So, AMD was able to make their MCM be seen as a single core in windows (so I've read).

The performance uplifts are going to be well above 2x. How much remains be seen. Making it clear that 7900xt should beat the 4090.
I find it hard to believe that AMD performance uplift with a MCM is about the same as if they went with a regular single die approach. Think about it for a minute. AMD knows the performance of the 4090 and potential 4090 ti.
Yet they still want to release a 7900xt and not the 7950xt (if they thought that the 7900xt would lose to the 4090).

I don't know man, it seems to me that these MCMs might pack more of punch then what's rumored and AMD has had no real leaks regarding RDNA 3 like they did the RDNA 1 and 2. There is certainly something peculiar this go around.

Some slight correction.

The GPU core isn't split up in RDNA 3 only the cache pool (which is connected to the VRAM). In future they may be able to split the core, but that is speculation at the moment.

 

[Tired9]

If Intel and Nvidia aren't absolutly crapping themselves they are idiots.

You seem like those who went nuts over HBM.. calm down mate, it's not that cut and dry.

 

[HACO]

You really think that greater then 2x is beyond earth? I don't. And, it's not hard to understand why nv didn't do the same thing. You are under a false conception that AMD and Nv are the same and they are not. Their Uarch to Gpus are different. Not everything is the same because "your favorite company" is doing xyz. So come back down to earth, nv didn't do the same as AMD because they are on different Uarch roadmaps.

I don't understand your aggressiveness friend. You should try not to take things personally, I wasn't addressing you and my point wasn't even directed at you

1. Nvidia isn't my favorite company, they're probably my least favorite company in computer hardware. I like AMD much more.

2. 2x performance isn't beyond earth, believing competitors are crapping themselves over a product that isn't even launched yet is.

3. MCM is on Nvidia's uarch roadmap. It may or may not happen with Hopper, but it's on their roadmap and may (or may not) eventually happen depending on how effective it will be compared to alternative solutions.

 

[Calin Banc]

I actually think MCM will be the future and will bring big gains like we have seen for CPUs, DLSS while ok is just a bandaid right now to help when turning on RT, as RT and raster improve there is less need for DLSS.

What nvidia is good at is creating problems and then selling you the solution, once RT is overcome then nvidia will be releasing next tech to make your current cards struggle so you have to buy their shiny new ones.

GPUs are already massive parallel processors, is not the same issue as it was for the CPUs.
On the GPU side it will help a bit with costs as it will be cheaper to create smaller dies and then combine them. So performance wise, I don't think we'll see much happening. On the prices sector, AMD increase the prices for their CPUs quite a lot. They won't sell you anything cheap, but rather pocket every penny they can and transfer those savings into their account. Double that for nVIDIA .

The only way that I see in which performance can go up is by having multi GPUs working together properly - so the rebirth of CF and SLI. Some will get a 4080, others would have bought another 3080/3090 and double their performance.

DLSS or FSR you still need. My rtx2080 was once considered a 4k card, now it can't even maintain 60fps@1080p without turning down details or using DLSS - depending per game. Same will happen with every card.

 

[EastCoastHandle]

I don't understand your aggressiveness friend. You should try not to take things personally, I wasn't addressing you and my point wasn't even directed at you

1. Nvidia isn't my favorite company, they're probably my least favorite company in computer hardware. I like AMD much more.

2. 2x performance isn't beyond earth, believing competitors are crapping themselves over a product that isn't even launched yet is.

3. MCM is on Nvidia's uarch roadmap. It may or may not happen with Hopper, but it's on their roadmap and may (or may not) eventually happen depending on how effective it will be compared to alternative solutions.

There is nothing to understand about it really. You're just unusually sensitive to being replied to. Since I was the only one within that time frame discussing MCM and you prompted by saying:

Coming back down to earth, if the MCM approach proves effective there's no reason Intel and Nvidia wouldn't end up doing the same thing down the road.

It was only natural to assume you were referencing my post. Even though you didn't quote it directly. So, I replied to you. It's not the serious.


Having said that:
1. I didn't make a claim that nvidia was "your company". I said "Not everything is the same because "your favorite company" is doing xyz." That could be Intel, AMD or Nvidia in this context. It would take your reply to me to tell me which of the 3 you perfer. And it's usually the 1st thing you say.

2. I didn't say anyone was crippling themselves. So I am geniunely confused by your response. You went from understanding I was saying that the 7900xt is going to be over 2.x the performance to somehow being something else.

3. I didn't say that MCM wouldn't, eventually, be on Nv roadmap (for us as gamers). I said that AMD/Nvidia have different roadmaps. Lets face it AMD is 1st and will have an advantage moving forward with future releases. I don't think anyone here doesn't expect Intel and Nvidia to do the same thing a few generations from now.

 

[HACO]

There is nothing to understand about it really. You're just unusually sensitive to being replied to. Since I was the only one within that time frame discussing MCM and you prompted by saying:

It was only natural to assume you were referencing my post. Even though you didn't quote it directly. So, I replied to you. It's not the serious.

Because you're the only one posting here so everything anyone says must be about you right?

Trust me, it was never about you and it still isn't, no matter how much you want it to be.

Having said that:
1. I didn't make a claim that nvidia was "your company". I said "Not everything is the same because "your favorite company" is doing xyz." That could be Intel, AMD or Nvidia in this context. It would take your reply to me to tell me which of the 3 you perfer. And it's usually the 1st thing you say.

Oh so you're doing 4D chess to figure out people's favourite companies, lol. I only mentioned the name of a company that you threw at me.

2. I didn't say anyone was crippling themselves. So I am geniunely confused by your response. You went from understanding I was saying that the 7900xt is going to be over 2.x the performance to somehow being something else.

I didn't say that either. Maybe read better? Nobody said anything about crippling, you misread that word. My point was directed at the person who something about competitors crapping (not crippling) themselves.

You'd definitely be less confused if you actually read what you're replying to before rushing to respond

3. I didn't say that MCM wasn't, eventually, be on Nv roadmap. I said that AMD/Nvidia have different roadmaps. Lets face it AMD is 1st and will have an advantage moving forward with future releases. I don't think anyone here doesn't expect Intel and Nvidia to do the same thing a few generations from now.

Once you're less confused, you'll see that none of this has any relevance.

 

[EastCoastHandle]

Because you're the only one posting here so everything anyone says must be about you right?

Trust me, it was never about you and it still isn't, no matter how much you want it to be.

Not really. But it was the only post during that time frame. And since you didn't provide any other post to suggest that (by now) my reasoning is sound. Thanks for verifying it.

Oh so you're doing 4D chess to figure out people's favourite companies, lol. I only mentioned the name of a company that you threw at me.

Not at all. Again, it's all in your reply. I only ask the question. Now that I know do I understand what position you are coming from. However, it's no secret based on how you are responding at this point.

I didn't say that either. Maybe read better? Nobody said anything about crippling, you misread that word. My point was directed at the person who something about competitors crapping (not crippling) themselves.

2. 2x performance isn't beyond earth, believing competitors are crapping themselves over a product that isn't even launched yet is.

I see... However, you seem awfully offended by the notion to state that they are crapping themselves over 2x performance. I don't understand your aggressiveness friend. You should try not to take things personally. Since this is RDNA 3 rumor thread it's only natural to discuss ...RDNA 3 speculation. And, you really have no proof if Nv is or is not concerned over the RDNA 3.

Once you're less confused, you'll see that none of this has any relevance

That has nothing to do with my post. I stated:

3. I didn't say that MCM wouldn't, eventually, be on Nv roadmap. I said that AMD/Nvidia have different roadmaps. Lets face it AMD is 1st and will have an advantage moving forward with future releases. I don't think anyone here doesn't expect Intel and Nvidia to do the same thing a few generations from now.

I see some confusion in your response as well. AMD will be 1st with this type of Uarch. Others like Intel/Nv will eventually follow. The main aspect of all this is what kind of performance uplifts will we see? That will soon be answered within weeks.

Amd is already aware of what Nv put on the table and it's time for them to response. And since it's clear that they want to provide a product to attract buyers they have to entice the consumer based on price and performance. Having said that it's clear they are going for 50% efficiency for RDNA 3. However, it's be rumored that it's well over 50%.

We saw the same with RDNA 1 to RDNA 2. But that came later in the product life cycle with skus not mentioned much on this forum such as:
-6900xtx
-6900xtxh

Not many reviews were had with the 6900xtxh variants. They were common watercooled but they had some aircooled versions as well. Which overclocked well above the stock 6900xt. And, you got better results using a negative power limit (as I found out in recent drivers).

For someone like me I would have to see some serious uplifts to upgrade as I am getting exceptional performance with what I have already and I do play competitively. Even with the 4090 I am not impressed. As I don't use upscalers.

 

[HACO]

Not really. But it was the only post during that time frame. And since you didn't provide any other post to suggest that (by now) my reasoning is sound. Thanks for verifying it.


Not at all. Again, it's all in your reply. I only ask the question. Now that I know do I understand what position you are coming from. However, it's no secret based on how you are responding at this point.



I see... However, you seem awfully offended by the notion to state that they are crapping themselves over 2x performance. I don't understand your aggressiveness friend. You should try not to take things personally. Since this is RDNA 3 rumor thread it's only natural to discuss ...RDNA 3 speculation. And, you really have no proof if Nv is or is not concerned over the RDNA 3.


That has nothing to do with my post. I stated:

I see some confusion in your response as well. AMD will be 1st with this type of Uarch. Others like Intel/Nv will eventually follow. The main aspect of all this is what kind of performance uplifts will we see? That will soon be answered within weeks.

Amd is already aware of what Nv put on the table and it's time for them to response. And since it's clear that they want to provide a product to attract buyers they have to entice the consumer based on price and performance. Having said that it's clear they are going for 50% efficiency for RDNA 3. However, it's be rumored that it's well over 50%.

We saw the same with RDNA 1 to RDNA 2. But that came later in the product life cycle with skus not mentioned much on this forum such as:
-6900xtx
-6900xtxh

Not many reviews were had with the 6900xtxh variants. They were common watercooled but they had some aircooled versions as well. Which overclocked well above the stock 6900xt. And, you got better results using a negative power limit (as I found out in the later drivers).

Oh Kevin. All of this irrelevant nonsense and strawmans because you don't want to admit you goofed up by misreading a word and taking offence at something that wasn't even directed at you. Seriously, take a deep breath and you'll be able to move on. Happens to all of us

For someone like me I would have to see some serious uplifts to upgrade as I am getting exceptional performance with what I have already and I do play competitively. Even with the 4090 I am not impressed. As I don't use upscalers.

Something we agree on. 4000 series (and 4090) suck.