• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD RDNA 4 thread

humbug said:
Its too late for RDNA 4, but they will take what they learned from it and apply it to RDNA 5. :)
Click to expand...

Honestly what excites me the most about the concept of the chiplet design is if they can get it working on a single board level. You can have the 9600 as 1 chiplet, 9700 and so on, but the idea that it could be possible to get near perfect scaling which could in theory bring back crossfire/sli and the like. The problem with them before was deminishing returns, but if they can solve that on a board level it's possible they could also solve it on a multiboad level. I doubt that will ever happen in all honesty. But the idea that a 9700 is 2x chiplets but 2x 9600 could perform almost as good as a single 9700 is a pretty interesting concept to me.

In all likely they will gimp the living hell out of it but i love the idea of simply doubling the chiplets for each tier of card so you can actually have a way of categorising them. If the lessons from Zen can carry over it could have very good scaling.
 
kieran_read said:
Honestly what excites me the most about the concept of the chiplet design is if they can get it working on a single board level. You can have the 9600 as 1 chiplet, 9700 and so on, but the idea that it could be possible to get near perfect scaling which could in theory bring back crossfire/sli and the like. The problem with them before was deminishing returns, but if they can solve that on a board level it's possible they could also solve it on a multiboad level. I doubt that will ever happen in all honesty. But the idea that a 9700 is 2x chiplets but 2x 9600 could perform almost as good as a single 9700 is a pretty interesting concept to me.

In all likely they will gimp the living hell out of it but i love the idea of simply doubling the chiplets for each tier of card so you can actually have a way of categorising them. If the lessons from Zen can carry over it could have very good scaling.
Click to expand...
Yeah...

From the R23 thread:

Score 32,393: AMD Ryzen R9 5950X at 4.8Ghz, RSR

Score 16,142: AMD Ryzen R7 5800X at +150Mhz, humbug

Both running about 4.8Ghz all core. 100% scaling for dual vs single CCD. :)

Windows doesn't see them as dual CPU's, because of the engineering work they put in to it, anything less than that causes complications, its only worth doing if its done right, like 64Bit wasn't a new thing by the time AMD came in to it, but they were the ones to do it right, its why now everything in high performance computing runs on the AMD64 instruction set, including the latest and greatest Intel CPU's, 25 years later...

If they have shelved it for now its because it isn't right.
 
Last edited:
humbug said:
Yeah...

From the R23 thread:

Score 32,393: AMD Ryzen R9 5950X at 4.8Ghz, RSR

Score 16,142: AMD Ryzen R7 5800X at +150Mhz, humbug

Both running about 4.8Ghz all core. 100% scaling for dual vs single CCD. :)

Windows doesn't see them as dual CPU's, because of the engineering work they put in to it, anything less than that causes complications, its only worth doing if its done right, like 64Bit wasn't a new thing by the time AMD came in to it, but they were the ones to do it right, its why now everything in high performance computing runs on the AMD64 instruction set, including the latest and greatest Intel CPU's, 25 years later...

If they have shelved it for now its because it isn't right.
Click to expand...
True.

Considering they take nearly a year to release a full lineup of one generation could also just be that it's coming a year latter and they will just name it a different gen.

Time will tell but its and exciting concept if it works
 
Apex said:
Wait...does anyone actually believe 12GB is going to be fine for the next few years?

You cannot be serious.
Click to expand...

I think it will be fine for me for the next couple of years, because I play at 1440 and I'm fine with medium settings and most of the games I play are lower budget indie games anyway.
 
CAT-THE-FIFTH said:
But apparently now 12GB of VRAM is fine for 4K for the next few years,because Nvidia wants to sell you an £800 RTX4070TI. Yet none of the people defending 12GB has fine for 4K for years,own a 12GB dGPU. If these people bought 12GB dGPUs to play games at 4K for the 3~5 years then it might show some actual belief.

But the issue is that Nvidia is not only skimping on VRAM but relative transistor increase per tier for their around 300MM2 dGPUs. If you compare their
Click to expand...
Nvidia are 100% selling you a card that should be worth a max of £500 for closer to £800, it’s a scam product.
 
Joxeon said:
Nvidia are 100% selling you a card that should be worth a max of £500 for closer to £800, it’s a scam product.
Click to expand...
Exactly - this thread is a prime example of how well Nvidia marketing works. As KompuKare said Apple spun 8gb was fine because reasons. The memory amount might be OK under £600 but not £800. Even the barely 20% faster RTX4080 has 16GB:

As an £800 card people are using "you are holding it wrong" logic to justify it is fine. GDDR6 is cheap now. Nvidia could easily have used more VRAM or used a die salvaged AD103 instead(to enable a 256 bit memory controller) and it would have been a better fit for the price.

AMD rightly got the flack it did with the RX7900XT nonsense, but it's clear the RTX4080 12GB..sorry... RTX4070TI 12GB is now being "rehabilitated". Soon wait for the rehabilitation of the RTX4060TI.

It makes me wonder at times. Think we are pretty much screwed both ways with PCMR!
 
Last edited:
It looks like RDNA4 was meant to be a substantial push forward with Chiplets

MLID has leaked a diagram of the now cancelled Navi41. Where as Navi31 (7900xtx) has a single monolithic GPU die flanked by 6 memory chiplets, Navi41 (8900XTX) was planned to consist of 14 to 20 chiplets, most of which would be Shader Chiplets and the rest Memory chiplets and there would also be an I/O chiplet like Ryzen CPUs have

This design would mean AMD needed to come up with a seriously fast interconnection method between the shader engines and memory chiplets - the interconnect used on rdna3 is hundreds of times too slow for shader chiplets. So I suspect this is the issue AMD has run into which is causing this design to get pushed back to rdna5



videocardz.com

AMD's canceled Radeon RX 8000 "Navi 4C" GPU diagram has been partially leaked - VideoCardz.com

AMD Navi 4C: The GPU that will never be? This would have been the AMD flagship RX 8000 GPU if not for the fact that it was canceled, allegedly. Alleged AMD Navi 4C, Source: Moore’s Law is Dead Moore’s Law is Dead recently shared a diagram of AMD’s unreleased RX 8000 GPU, codenamed Navi 4C. In […]
videocardz.com videocardz.com
 
Last edited:
Grim5 said:
It looks like RDNA4 was meant to be a substantial push forward with Chiplets

MLID has leaked a diagram of the now cancelled Navi41. Where as Navi31 (7900xtx) has a single monolithic GPU die flanked by 6 memory chiplets, Navi41 (8900XTX) was planned to consist of 14 to 20 chiplets, most of which would be Shader Chiplets and the rest Memory chiplets and there would also be an I/O chiplet like Ryzen CPUs have

This design would mean AMD needed to come up with a seriously fast interconnection method between the shader engines and memory chiplets - the interconnect used on rdna3 is hundreds of times too slow for shader chiplets. So I suspect this is the issue AMD has run into which is causing this design to get pushed back to rdna5



videocardz.com

AMD's canceled Radeon RX 8000 "Navi 4C" GPU diagram has been partially leaked - VideoCardz.com

AMD Navi 4C: The GPU that will never be? This would have been the AMD flagship RX 8000 GPU if not for the fact that it was canceled, allegedly. Alleged AMD Navi 4C, Source: Moore’s Law is Dead Moore’s Law is Dead recently shared a diagram of AMD’s unreleased RX 8000 GPU, codenamed Navi 4C. In […]
videocardz.com videocardz.com
Click to expand...

Yup, makes sense, more engineering work needed.
 
humbug said:
Yes.

Which is why it needs to work in a way where the OS and the game doesn't even know its multiple dies.

But damn that was a monster GPU wasn't it? Look how much faster it is even compared to the dual GPU GTX 690.
Click to expand...

Yeah, it was super nice. If money would not be a problem, then definitely was nice card to game on and add to a collection. :)
But, since I'm more sensitive to that, I would have probably went with a couple of r290s in CF (or 3x290 for about the same money :p ). Those drivers where at their peak when it was about multi GPU scaling...

kieran_read said:
Honestly what excites me the most about the concept of the chiplet design is if they can get it working on a single board level. You can have the 9600 as 1 chiplet, 9700 and so on, but the idea that it could be possible to get near perfect scaling which could in theory bring back crossfire/sli and the like. The problem with them before was deminishing returns, but if they can solve that on a board level it's possible they could also solve it on a multiboad level. I doubt that will ever happen in all honesty. But the idea that a 9700 is 2x chiplets but 2x 9600 could perform almost as good as a single 9700 is a pretty interesting concept to me.

In all likely they will gimp the living hell out of it but i love the idea of simply doubling the chiplets for each tier of card so you can actually have a way of categorising them. If the lessons from Zen can carry over it could have very good scaling.
Click to expand...

Sadly, I think is quite different from chiplets to multi GPU. With chiplets you have / need a lot of bandwidth and low latency to make it work in gaming, PCI-E can't offer that. Maybe to somehow catch the game code and do distribute as needed could work, a bit like nVIDIA did for their dx11 drivers, but... is a lot of work. If my memory serves me right, I think there's nothing stopping a studio in DX12/Vulkan to enable multi GPU, even combined AMD+nVIDIA. Is just that there isn't much interest in that...
 
Cancelled Navi 4C, this will likely be similar to Navi 5 now.

They are still developing the technology but no longer for RDNA 4, its going to take longer.

6o1AB8D.jpg


lTlcPBR.png


 
Last edited:
Jay85 said:
God knows but this is some rumour being spun lol. Lots and lots of hopium.


Click to expand...


Lol MLID said Navi 41 was cancelled and rdna4 would mostly compete with rtx5000 midrange, then few days later this guy makes a video saying rdna4 is going to beat the rtx5090 and it's going to launch early before Nvidia can
 
Grim5 said:
Lol MLID said Navi 41 was cancelled and rdna4 would mostly compete with rtx5000 midrange, then few days later this guy makes a video saying rdna4 is going to beat the rtx5090 and it's going to launch early before Nvidia can
Click to expand...
I'm sure this guy has admitted he just churns out clickbait and plays the algorithm game.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom