*** The AMD RDNA 4 Rumour Mill ***

[Grim5]

100% this. David Wang said a few years ago that this was impossible due to the ridiculous bandwidth requirement needed to do chipsets for GPUs.

I still wonder why AMD even tried chiplet GPUs in the first place. They must have thought it would work at first otherwise the 7900xtx wouldn't exist and then suddenly they backtrack and it's gone. I can only assume AMD at first did not think the bandwidth requirements would be as ridiculous as it turned out to be, but also, how could they not know, are these things not able to be emulated in design software?

I do commend AMd for trying new things that Nvidia is too scared to do - AMD put HBM memory on gaming cards, Nvidia won't and AMD was first to chiplets too.


I do hope AMD revisits chiplet GPus again in future once TSMC is able to do fiber optic chiplet interconnects so they can get the bandwidth that's needed to make it work. We're about 5 years away from this. I'll be excited once we get there, the idea of having 2, 3, 4 or more GPUs on a single graphics card is salivating, imagine the performance potential

 

[harmattan]

I still wonder why AMD even tried chiplet GPUs in the first place. They must have thought it would work at first otherwise the 7900xtx wouldn't exist and then suddenly they backtrack and it's gone. I can only assume AMD at first did not think the bandwidth requirements would be as ridiculous as it turned out to be, but also, how could they not know, are these things not able to be emulated in design software?

I do commend AMd for trying new things that Nvidia is too scared to do - AMD put HBM memory on gaming cards, Nvidia won't and AMD was first to chiplets too.


I do hope AMD revisits chiplet GPus again in future once TSMC is able to do fiber optic chiplet interconnects so they can get the bandwidth that's needed to make it work. We're about 5 years away from this. I'll be excited once we get there, the idea of having 2, 3, 4 or more GPUs on a single graphics card is salivating, imagine the performance potential

Because the potential improved yields and modularity is exceedingly attractive.

Also' I haven't seen them, but expect a lot of the proposed chiplet tech is patented. If AMD had suceeded, it would be an area the competition would't have been able to touch (at least not in the same way).

I'd also assume, after the six months of R&D, sunk cost fallacy comes into play.

Edit: I'd also commend AMD for pushing the envelope on risk/reward over the years. Granted, a lot of it has been out of necessity... but you know what they say about necessity. And at least it panned out in the CPU space with dividends.

 

[CAT-THE-FIFTH]

RDNA4 uses an old node and GDDR6. It should have launched by now especially if it can't match an RX7900XT. Instead this is being released when the RTX5090, RTX5080 and RTX5070TI aew being launched. Unless this is another 4870,it's going to make AMD look even worse.

 

[JediFragger]

All I'm hoping for is some kind of hardware-accelerated upscaling from FSR4 for the RDNA3 cards, totally not bothered about the hardware (bar a possible 9800X3D upgrade, but it's hardly needed).

 

[Firegod]

RDNA4 uses an old node and GDDR6. It should have launched by now especially if it can't match an RX7900XT. Instead this is being released when the RTX5090, RTX5080 and RTX5070TI aew being launched. Unless this is another 4870,it's going to make AMD look even worse.

It is a mid range GPU.

 

[Bidley]

It is a mid range GPU.

That barely competes with 2 year old mid-range GPUs?

 

[rare]

That barely competes with 2 year old mid-range GPUs?

In true RTX 4060 fashion!

 

[Bidley]

In true RTX 4060 fashion!

Was the 4060 ever mid range?

 

[JediFragger]

5060 rumoured to be getting a decent core count bump though (over the 4060), hopefully it's true.

 

[Bidley]

5060 rumoured to be getting a decent core count bump though (over the 4060), hopefully it's true.

That'll be good with the 8gb of VRAM.

 

[rare]

Was the 4060 ever mid range?

It’s genuinely difficult to say these days I think. Used to be a lot clearer when 80 parts were the flagship, they have obviously been far superseded now with the 90 which is the clear flagship and the only part that genuinely pushes the envelope. It’s also harder because Nvidia have seemingly abandoned the low end market, with no 4050 or equivalent.

I still think of 4060-4070 as midrange though. But it’s obviously up for debate.

 

[TNA]

RDNA4 uses an old node and GDDR6. It should have launched by now especially if it can't match an RX7900XT. Instead this is being released when the RTX5090, RTX5080 and RTX5070TI aew being launched. Unless this is another 4870,it's going to make AMD look even worse.

In other words there is a chance they are just waiting for nvidia to price their cards so they can slot in $50 or so cheaper to maximise profits. All the while clearing out old stock.

Win win! In someone's mind over there anyway

 

[StarShock]

If amd doesn't release a 4090 tier card for under £200 they're through /s

People need to chill out we're just 2 weeks from real information, then we can decide. Let's face it. If nvidia decide they're gonna make the 5060 £500 and give it a fill 7% performance jump over the 4060 then regardless the 9070 xt is going to look good value.

 

[CAT-THE-FIFTH]

It is a mid range GPU.

It doesn't matter what it is - it should have launched by now as the RTX5000 series will hog all the news cycles next month. The RTX5000 series was delayed due to needing a re-design and GDDR7 being delayed too. If they had launched in November,AMD would at least get 100% exposure on the new design plus Christmas sales. Its why Intel pushed the B580 out this month.

Unless the RX9070XT is something special,it's going to get buried under all the RTX5000 articles.

In other words there is a chance they are just waiting for nvidia to price their cards so they can slot in $50 or so cheaper to maximise profits. All the while clearing out old stock.

Win win! In someone's mind over there anyway

This is basically what they are doing IMHO.

 

[JediFragger]

we're just 2 weeks from real information

 

[Bidley]

It’s genuinely difficult to say these days I think. Used to be a lot clearer when 80 parts were the flagship, they have obviously been far superseded now with the 90 which is the clear flagship and the only part that genuinely pushes the envelope. It’s also harder because Nvidia have seemingly abandoned the low end market, with no 4050 or equivalent.

I still think of 4060-4070 as midrange though. But it’s obviously up for debate.

I'm not sure it's that difficult, really - the only thing that has changed is the naming scheme and pricing. 90 is the new 80, 60 is the new 50.

Historically the current gen's 70 class has matched the previous gen's 80 class. It looks like this is also changing.

 

[gpuerrilla]

RDNA4 uses an old node and GDDR6. It should have launched by now especially if it can't match an RX7900XT. Instead this is being released when the RTX5090, RTX5080 and RTX5070TI aew being launched. Unless this is another 4870,it's going to make AMD look even worse.

Yeah I would agree with this. They now have little room though saying focus is on the affordable segment, if they fail to read the room and mess up the launch price it will complete the circle.

 

[humbug]

Theory is AMD botched their dual chip high performance design and instead put them chips to the 9070xt and then spin some bs about focusing on the mass market. No market disruption, the status quo it is and more market share loss.

I think "Botched" is hyperbole, they had an RDNA 4 MCM design but canaled it due to development costs and eventual retail cost, notice it has 6 GPU tiles, they already have something very similar to this in the CDNA range.
This is modern journalism, hyperbolic drama garbage for smooth brain consumption, click bait brain rot. The truth is much more boring, its unviable.
I mean, you don't think AMD could just make a larger version of Navi 48? make it 384Bit, add more CU's? is that harder than making it MCM? hm? like i said, Brain Rot.

 

[humbug]

Oh..... also, (Navi 10. Navi 21, 22, 23, 24. Navi 31, 32, 33.) The image i posted above was designated Navi 43, it no longer exists, what we do have is Navi 44 and Navi 48.

There are designs missing from Navi 2 and Navi 3 that never made it to retail, namely Navi 20 and Navi 30, there are a lot more missing from Navi 4, the only one that made it through was the smallest, Navi 44, the larger GPU is actually Navi 48 indicating it is a later design, possibly serval steps of it, Navi 45, Navi 46, Navi 47. and Navi 48 being the flagship. On top of that Navi 40 through to Navi 43 also never made it as designs.