• 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.

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

I do hope AMD don't mess about and just push them out at $500 for initial reviews and get them out before Nvidia, aside from the good deal i'm interested to see how Nvidia would react, they could easily spoil AMD's party by slapping them in th face with a $500 to $550 RTX 5070 or stay on course for gradually ramping the price of GPU's up and pushing them out at $650, i would be interested to see which direction Nvidia take.
 
Last edited:
I'm curious about something and I think you might be able to answer it. How does a Ryzen 7900 compare in chip size with AMD GPUs?

That's a complicated question, the dual CCD Ryzen Zen 4 are 5nm for the two logic dies and 6nm for the IO die. The two logic dies are about 160mm combined, about 80mm each with IO die being about 120mm.

RDNA 3, is also 5nm for the logic die and 6nm for the memory dies, Navi 31 (7900 series) has a 300mm logic die and 6 37mm memory dies, about 220mm combined for the memory dies.

Navi 32 (7800 / 7700 series) has a 200mm logic die and 4 memory dies.

So....
Ryzen 7950X: 160mm 5nm dies + 120mm 6nm die.
7800 XT: 200mm 5nm die + 150mm 6nm dies.
7900 XTX: 300mm 5nm die + 220mm 6nm dies.
 
Last edited:
5nm / 6nm (5 and 6 mano metre) refers to the printing lithography of the die, another word for basically chip. How the actual chip is made, the smaller the number the smaller and better / later that is, all made at TSMC.

The first image is Navi 31, the 7900 series RDNA 3 GPU's

The second image is of a Ryzen 7900 series CPU with the heat spreader removed.

0GjSuei.jpeg

hbOMIdF.jpeg
 
Last edited:
Thank you Humbug.
So, what I'm thinking is: should AMD stop making big chips and focus on something the size of a Ryzen 9700?

Or, if slightly older nodes are price competitive, should AMD try to squeeze the best possible from nodes one generation behind and openly offer players a trade-off between price and efficency?
Basically declaring the best node available is for CPU and Instinct while GPU stays on the previous CPU node/cheaper alternative (Samsung?).

This way volume woes should be significantly lessened and price could be competitive.
There are also some R&D advantages: You set the microarchitecture with iGPU/APU and refine it on the same one when you actually switch it to dGPUs, basically pulling a tick/tock strategy having sinergy with fab node refinements.

Intel tried this and failed but AMD is much better positioned to take advantage of a full stack strategy.

The strategy is to go back to making workstation GPU's and gaming GPU's the same silicon, so if they don't sell them as gaming GPU's they stand a better chance moving them as workstation GPU's, i don't know if this is true for upcoming RDNA 4, probably not, possibly RDNA 5.

AMD wouldn't go Samsung, too far behind and they have a good relationship with TSMC that they would want to maintain, the best and most expensive node right now is TSMC N3, that's a 3nm node, its what Apple are currently using and Intel for their 200 series CPU's and APU's.
I think AMD are very likely to use TSMC N4, 4nm for RDNA 4, they use that for their Ryzen 9000 series, its a very solid node and as shown with Ryzen 9000 AMD are able to design very good efficiency with it.
They may also go back to it being monolithic, as you can see from my pictures and explanation current Radeon 7000 series (RDNA 3) is a multichip design, its a very successful design and the first MCM GPU but i think AMD want to go back to basics with RDNA 4, i think that's a shame, AMD are the best when it comes to inventing innovative architectural designs and i would have liked to have seen the next evolution of their designs but i think they don't want to spend the R&D for it anymore, they can't justify it. AMD are such a talented company in this way, many world firsts to their name, its a shame they don't have Nvidia money to play with.

The new GPU's will be smaller, they are not making high end GPU's anymore, highest end RDNA 4 will be the RX 8800 XT, equivalent in performance to an RX 7900 XTX with much better Ray Tracing and about the size on an RX 7800 XT, around 250 - 300mm, that's a bit larger than a Ryzen 9700, about 50 to 100mm larger but i don't see how you can get it any smaller with that level of performance and 300mm or less is quite small for a GPU.

You're right that AMD will continue to design GPU technology because their APU's are very successful.

The problem tho is this: An RX 7800 XT chip, just the chip costs about the same to make as a Ryzen 7950X, they currently sell that CPU for £500, bargain... its £100 cheaper than a Core Ultra 285K and better.
So take another look at the images i posted, what you see for the 7950X is pretty much how it sells, the only thing that is missing is the lid, the heat spreader, $5, they sell that at a supply chain who take their share who sell it to OCUK who take their share, so AMD probably sell this £500 CPU for £350 to £400, its costs less than £100 to make it retail ready, £200 to £250 profit, while that seems like a lot remember than they spent serval hundred million $ designing it, i don't know that but i would imagine so.
The same R&D cost applies to the GPU. AMD sell that to their partners, like Sapphire who design and manufacture their own PCB's and coolers for them, CPU's don't need to be shipped with coolers and they don't have PCB's, as such. So to sell what is now a £420 GPU they need to leave enough money for Sapphire to make a profit after designing and making the PCB and cooler.
Look at this thing, this is a 7800 XT Nitro PCB, i can't find the Pulse PCB but they are made to the same standard, believe it or not... there are thousands of individual components on this thing and some of them are quite expensive, costing multiple $ individually.
Sapphire have to design all of this, make it and then they sell it in to a supply chain, AMD are not selling these chips to Sapphire for £350 to £400, the thing costs £420 retail, they are selling it at a little above cost, that's fine if you're selling 30 million of them, but they aren't, that's why AMD profit from these in the last quarter was $12 Million, or in other words nothing, if they sold 120K units that's $10 profit on each one sold. Now ask your self how many hundreds of millions did AMD spend developing this thing? Ryzen is propping up Radeon in a very big way.


OKvXcud.jpeg
 
Last edited:
But AMD was actually in talks with Samsung and TBH I doubt Instinct has the volume to actually impact Radeon's costs.
AMD is not in an easy position TBH so either they manage to find a way to provide at least 80% of the value for 50% of the price or they won't be able to expand their market share.

Look at what's happening in the smartphone market.
Mediatek is basically playing AMD against Qualcomm but they're leaving the budget segment unguarded, what's happening is that Unisoc is starting to eat their lunch there.
Intel could potentially do the same if they can make Battlemage viable so AMD has to try something disruptive or they will find themselves Matroxed into Instinct and semi-custom.

I didn't know AMD was in talks with Samsung, its an interesting one to watch.

The way i see it, the thing about workstation GPU's is that segment is growing, more and more people want them and not just in huge batches, more small outfits and even individuals are looking for them, for the latter it is about cost and if instead of selling them for $2000 a pop what if they sold them for half that or even less? AMD could carve out a new market for it self, its probably not very politically correct to make this segment so affordable, Nvidia will hate them for it but so what AMD are opening up that market to more people.

AMD did this first with HEDT where a Ryzen 1800X was half the price of a 6900K and just as good if not better, they did it again in the server space, Intel was charging up to $50,000 per chip, AMD brought that down to $18,000 with better chips, its why Intel are in trouble, running fabs is hugely expensive and you can only do that as an individual by selling chips for $50,000 a pop to pay for it all.

They could disrupt the GPU workstation segment in the same way.

Honestly i don't follow the smartphone market, other than one Samsung collaboration i didn't think AMD have any presence there at all?

Intel have all but given up on the dGPU market, the rumour is Battlemage consists of one small SKU for laptops, i don't blame them, they thought they could just walk in and take AMD's market share but quickly found out just how brutal this segment is.
 
Last edited:
Back
Top Bottom