AMD Radeon RX 9060 XT to Roll Out 8 GB GDDR6 Edition, Despite Rumors

[lexluthermiester]

The only constant in time is change...

And yet, the more things change the more they stay the same.

Also in a world where people use terms such as '2K' liberally to denote any resolution within 40% of 2000 horizontal pixels at this point... good luck getting a true definition to last more than a year lol.

Exactly. By the numbers, 2k is the same as 1080p because 1920 is closer to 2000 than 2560. 2560x1440p is 2.5k and yet every dingbat on the net calls it 2k.. Nitwits..

 

[Vayra86]

And yet, the more things change the more they stay the same.

Paradoxically true

 

[ARF]

Then don't buy it (I don't mean to disagree with you, just saying).


Good. Maybe then, AMD (and Nvidia) will learn.

I have a question - which one do you think will be faster - RX 7600 XT 16 GB or the "new" RX 9060 XT 8 GB ?
And another - if it proves that the latter is slower, how will AMD explain this generational regression ?

 

[Mrgravia]

I have a question - which one do you think will be faster - RX 7600 XT 16 GB or the "new" RX 9060 XT 8 GB ?
And another - if it proves that the latter is slower, how will AMD explain this generational regression ?

Should that not be compared to the 9060xt 16gb? The 8gb would be like the 7600 8 gig variant .

 

[AusWolf]

I have a question - which one do you think will be faster - RX 7600 XT 16 GB or the "new" RX 9060 XT 8 GB ?

It'll depend on the situation.
In any case where the 8 GB VRAM is not exceeded, or the game engine isn't bothered by swapping data with the system RAM too much, the 9060 will be faster. Any other cases, the 7600 XT of course.

And another - if it proves that the latter is slower, how will AMD explain this generational regression ?

I don't care. I'm not gonna buy one, and neither should anyone slightly concerned by the above. Thankfully, the 16 GB variant exists.

 

[Craptacular]

If you don't understand what's being said, can't help you man. It's pretty clear.


Let me quote my self

Because the card with more vram will cost more than the card with less

Sure man.

So, in conjunction with this statement: vram has nothing to do with the cards being expensive.

You are saying the cards will be expensive regardless of the vram amount. So why even have the lower vram amount then? Saying it will cost less isn't an argument because if you can already afford an "expensive" gpu why would you ever go with the card with less vram that is also "Expensive? Unless of course you can't afford an "expensive" gpu and the lesser vram brings it down to the "non-expensive" price like the B570 and B580 gpus.

 

[JustBenching]

So, in conjunction with this statement: vram has nothing to do with the cards being expensive.

You are saying the cards will be expensive regardless of the vram amount. So why even have the lower vram amount then? Saying it will cost less isn't an argument because if you can already afford an "expensive" gpu why would you ever go with the card with less vram that is also "Expensive? Unless of course you can't afford an "expensive" gpu and the lesser vram brings it down to the "non-expensive" price like the B570 and B580 gpus.

Well, at least you understood what I was saying.

It depends on the price, if it's only 30$ then sure youll go for the higher vram card. But still - I don't see a reason for the 8gb not existing at all.

 

[ARF]

A 16GB version is also coming, but the question remains - why did AMD make decisions to have both Navi 48 and Navi 44 with equal amounts of VRAM?
Couldn't they design the Navi 48 to have 24 GB, while Navi 44 16 and 12 GB versions?
Is it so difficult for them to follow more logic and sense in order to offer according to the gaming market realities, without artificially plaguing their cards with insufficient VRAM amounts ?

https://videocardz.com/newz/gigabyte-radeon-rx-9060-xt-gaming-with-16gb-memory-listed-in-brazil

 

[AusWolf]

A 16GB version is also coming, but the question remains - why did AMD make decisions to have both Navi 48 and Navi 44 with equal amounts of VRAM?
Couldn't they design the Navi 48 to have 24 GB, while Navi 44 16 and 12 GB versions?
Is it so difficult for them to follow more logic and sense in order to offer according to the gaming market realities, without artificially plaguing their cards with insufficient VRAM amounts ?

View attachment 398414

https://videocardz.com/newz/gigabyte-radeon-rx-9060-xt-gaming-with-16gb-memory-listed-in-brazil

It's because of the memory controller. To have 24 GB VRAM, you need either a 192-bit bus, which is not enough, or a 384-bit one, which makes the GPU bigger and more expensive. Neither of these are good solutions in the current market. 16 GB on 256 bits is the price-to-performance sweet spot.

Similarly on Navi 44. To have a 12 GB variant, you need to cut down the 128-bit bus to 96 bits. You don't want that.

 

[ARF]

It's because of the memory controller. To have 24 GB VRAM, you need either a 192-bit bus, which is not enough, or a 384-bit one, which makes the GPU bigger and more expensive.

A wider memory bus doesn't mean a larger die. You have enough space on the 360 mm2 Navi 48 to widen the bus over the current 256-bit.
Also, 192-bit bus if cleverly designed with a larger Infinity Cache to 128 MB or better 256 MB could have been a solution.

Similarly on Navi 44. To have a 12 GB variant, you need to cut down the 128-bit bus to 96 bits. You don't want that.

Or to 192-bit, which you definitely want.

 

[AusWolf]

A wider memory bus doesn't mean a larger die. You have enough space on the 360 mm2 Navi 48 to widen the bus over the current 256-bit.

Where?

Also, 192-bit bus if cleverly designed with a larger Infinity Cache to 128 MB or better 256 MB could have been a solution.

I disagree. Infinity Cache doesn't seem to give enough advantage to substitute a wider memory bus.

Or to 192-bit, which you definitely want.

Again, larger die. You don't want that in the current overpriced state of the market.

(I mean I'd want the chip, I just wouldn't want the price of it)

 

[ARF]

Where?

On the sides.
If you look at the die, you will see that the 16 16-bit memory bus controllers are placed on two of the four sides of the die.
You have plenty of space to include more 16-bit controllers.

 

[AusWolf]

On the sides.
If you look at the die, you will see that the 16 16-bit memory bus controllers are placed on two of the four sides of the die.
You have plenty of space to include more 16-bit controllers.

View attachment 398454

That's just a diagram, it doesn't scale with the die. Please mark it on the die shot that I attached. You'll see that there is no empty space on it.

Edit: Here's a better one.

 

[lexluthermiester]

It's because of the memory controller. To have 24 GB VRAM, you need either a 192-bit bus, which is not enough, or a 384-bit one, which makes the GPU bigger and more expensive. Neither of these are good solutions in the current market. 16 GB on 256 bits is the price-to-performance sweet spot.

Similarly on Navi 44. To have a 12 GB variant, you need to cut down the 128-bit bus to 96 bits. You don't want that.

Or 1.5/3GB memory chips...

 

[AusWolf]

Or 1.5/3GB memory chips...

On GDDR6? Do they exist?

 

[lexluthermiester]

On GDDR6? Do they exist?

Um, it's technically possible. Honestly, don't know though.

 

[GodisanAtheist]

AMD was obviously expecting this gen to be a complete wipe-out thanks to their MCM designs getting scrapped, so N44 was always supposed to be a bottom bin die and N48 was cobbled together last minute (Hence N48 instead of N43 or N42 which is roughly where it belongs on the stack, because the numbering goes up the later the die is designed) to be a lean mean clocked to the sky machine.

Trying to extract logic from AMD's current product stack is going to be an exercise in frustration, because the product stack itself is a last minute gamble to make sure that they didn't miss this gen entirely. Turns out Nvidia is too busy making a quintillion dollars on AI to care about consumer Blackwell and now we have to act like RDNA4 was anything else other than a desperate gamble for AMD not to get stuck with a xx50 series part as their only new offering on the market.

I'm sure AMD wishes they did a lot of things differently for RDNA 4 knowing what they know now, they could have come in absurdly strong with a top to bottom monolithic stack, but they had to go to war with the cards they have not the ones they wished they'd designed knowing what we know now.

 

[AusWolf]

AMD was obviously expecting this gen to be a complete wipe-out thanks to their MCM designs getting scrapped, so N44 was always supposed to be a bottom bin die and N48 was cobbled together last minute (Hence N48 instead of N43 or N42 which is roughly where it belongs on the stack, because the numbering goes up the later the die is designed) to be a lean mean clocked to the sky machine.

Trying to extract logic from AMD's current product stack is going to be an exercise in frustration, because the product stack itself is a last minute gamble to make sure that they didn't miss this gen entirely. Turns out Nvidia is too busy making a quintillion dollars on AI to care about consumer Blackwell and now we have to act like RDNA4 was anything else other than a desperate gamble for AMD not to get stuck with a xx50 series part as their only new offering on the market.

I'm sure AMD wishes they did a lot of things differently for RDNA 4 knowing what they know now, they could have come in absurdly strong with a top to bottom monolithic stack, but they had to go to war with the cards they have not the ones they wished they'd designed knowing what we know now.

They needed something to release their RT improvements and FSR 4 on. Waiting until UDNA is ready wouldn't have been a very good option.