RDNA 5 speculation

Marees

2[H]4U
Joined
Sep 28, 2018
Messages
2,091
Top RDNA 5 skus to have 135/180 WGPs. AMD targetting release by 2025 end

~RedGamingTech
 
The previous hype cycle did not remove the talk of the 300% jump for a while ?

If it is 3x-4x why even mention ancient history like a 4090, it will destroy the 5090 and could beat the 6090 as well, we are talking better than 2 generation achieving 75% gen on gen in a row, Nvidia doing again the 3090 to 4090 jump with the 5090 and once again for the 6090 and that not fully reach 4 times a 7900xtx
 
Last edited:
Calling a 4090 destroyer when it comes out 3 years afterwards lol... I would hope so.

If that "priced at $2000" wasn't there and it was a <$1K card, I'd say it would be a killer card.

Beating a 4090 in 2025 at $2K? Ok. Not very excited. Not even "Meets or exceeds expectations". More of "Needs Improvement".
 
New AI in RDNA 5 is capable of manufacturing up to 3 Nividia 5090 clones (requires "the fab" add-on) all while just using 100W. Or so I hear.
 
The previous hype cycle did not remove the talk of the 300% jump for a while ?

If it is 3x-4x why even mention ancient history like a 4090, it will destroy the 5090 and could beat the 6090 as well, we are talking better than 2 generation achieving 75% gen on gen in a row, Nvidia doing again the 3090 to 4090 jump with the 5090 and once again for the 6090 and that not fully reach 4 times a 7900xtx

Nope, AMD rumors will always have the "300%" gain rumor every single time. I'm guessing someone just keeps accidently adding the extra 0
 
AMD has been working through the turmoils of their chiplet GPU's. We see the hiccup that the 7000 series is. RDNA 4 is said to be a limited arc that focuses on the mid range, as to advance the chiplet designs with a focus of lower power usage, manufacturing costs and the ability of the chiplet design to faster incorporate GPU advancements like dedicated A.I chiplets.
I would think by RDNA 5's release the fruits of AMD's tactic should arise.
Nvidia may have to go through a similar struggle yet the $$ they can throw at the problem it probably won't set them back.
AMD has executed the chiplet tech quite well on the CPU side so they have obviously learned a lot and are a market leader in these regards. I don't see why then in two gens from today's GPU tech they wouldn't follow up with the same successes.
 
I would think by RDNA 5's release the fruits of AMD's tactic should arise.
I wish the best for AMD but I hope RDNA 5 is not a multi GCD chip. Then there is no chance in hell of AMD ironing out the "bugs" by 2025 and it will definitely postpone to 2026

I hope it is a single GCD like RDNA 3 (Navi 31 & Navi 32)
 
AMD has been working through the turmoils of their chiplet GPU's. We see the hiccup that the 7000 series is. RDNA 4 is said to be a limited arc that focuses on the mid range, as to advance the chiplet designs with a focus of lower power usage, manufacturing costs and the ability of the chiplet design to faster incorporate GPU advancements like dedicated A.I chiplets.
I would think by RDNA 5's release the fruits of AMD's tactic should arise.
Nvidia may have to go through a similar struggle yet the $$ they can throw at the problem it probably won't set them back.
AMD has executed the chiplet tech quite well on the CPU side so they have obviously learned a lot and are a market leader in these regards. I don't see why then in two gens from today's GPU tech they wouldn't follow up with the same successes.

Even as the current market leader, the 7800x3d still tend to beat the 7950x3d in games and the next GPU launch is rumoured to be pure monolithic for good reasons.

The ways game are made and run is quite the challenge to be run with multiplie GPUs even that close to each other, end of the days some stuff need to be sync 100 times per second and a lot of effect like to have the complete current buffer to work with, which made the easiest SLI solution of the past like each gpu render part on the screen out of the window and made fast FPS but with stutter issue the norm.

Maybe they launch a real chiplet GPU and not just having cache-io chips surrounding a monolithic logic one by 2025, but maybe not. I am not sure if simulators are good enough for them to know before they actually make it work in the real world if it will.

But there no doubt they have a plan A (or B) for it and will try to do it like last time, which will make the rumours mill turn like the previous gen (and I imagine the previous before that)
 
Last edited:
They want to be literally, nvidia - same or similar pricing tactics - yet so many fanboys out there.
Who would not want to be Nvidia.... maybe Apple and that it.

But there hype video and reality, Nvidia does not have a 7800xt product for the moment.

bs marketing about having hardware for content creation.

This post would be quite more effective if it has shown real world result significantly different from the marketing, stuff like up to 27% faster is not a big claim.
 
Rumors always start off huge then at the launch gets closer you start seeing the usual increases in performance.
 
AMD are notorious liars, just like Nvidia. This is AMD's bs marketing about having hardware for content creation.

View: https://www.youtube.com/watch?v=INyWSrCV8ys

View: https://www.youtube.com/watch?v=L5vGhT4j66I

View: https://www.youtube.com/watch?v=BKzty6YdRNM

Yeah they just needed the software makers to support their cards better. AMD works really well for photo and video apps, now:


View: https://youtu.be/xmsPhb6Tugw?si=pNm-XKR2yNfHeKUm

View: https://www.youtube.com/watch?v=JIsFqX4LgZI


3D apps still have yet to take the time to make better AMD specific code. CUDA translation to ROCm is only serviceable, until then.
 
Yeah they just needed the software makers to support their cards better. AMD works really well for photo and video apps, now:


View: https://youtu.be/xmsPhb6Tugw?si=pNm-XKR2yNfHeKUm

View: https://www.youtube.com/watch?v=JIsFqX4LgZI


3D apps still have yet to take the time to make better AMD specific code. CUDA translation to ROCm is only serviceable, until then.

Interesting, thanks. AMD gpus - at least, the 7900 series looks pretty decent for Davinci Resolve, Premiere Pro etc. But, in Blender - that guy didn't utilize HIP-RT for Blender - I don't like that many of these reviewers don't even mention it exists - either say it crashed or didn't work or that they didn't use that setting - the default for AMD is HIP - but, of course results will suck compared to Nvidia and Optix. I dunno about the other programs - many seemed to either only work with Nvidia or are really fine tuned for Nvidia cards. But, AMD gpus did okay (imho) with video editing.
 
RDNA 5 top card could be 3x-4x faster than 7900 XTX & priced up to $2000


View: https://youtube.com/watch?si=YBwwCAG0-kw07H4-&v=B5jxm1rmMH4



Forget FPS benchmarks.... with such raw processing claims just bench the raw data processing to see what is actually going on with the processing performance.
Resistance increases TDP and electrical energy costs but doesn't necessarily improve real time human user performance.

I'm ok with 1080p gaming on a 26" display, but it doesn't really matter because UT3 servers are empty.

Bigger fish to fry.
 
I do hope that they can return to the level of competitiveness they had during the RX 6000 series era, but the rumors of them pulling out of the high end next generation are disconcerting...

It seems like Radeon only has a hot patch every 3 to 4 hardware generations...
 
Last edited:
I've seen some rumors about 3x performance using exotic gpu chiplet build in RDNA5. I don't believe them.

AMD would need to master gpu chiplet manufacturing process in RDNA5 when they have decided to skip it entirely in RDNA4 either for profitability or practicality reasons.
 
I do hope that they can return to the level of competitiveness they had during the RX 6000 series era, but the rumors of them pulling out of the high end next generation are disconcerting...

It seems like Radeon only has a hot patch every 3 to 4 hardware generations...

Yeah prior to the 6900XT I can't remember the last time AMD was going toe to toe with Nvidia's flagship. Was hoping they would be able to do the same thing with RDNA3 but here they are back to competing with Nvidia's 2nd best instead and next gen looks to be either the same thing where they can only compete up to the 2nd best from Nvidia, or even worst where they can only compete with maybe the 3rd/4th best in Nvidia's lineup. Maybe RDNA5 can get things back to the 6900XT days but meh I'm even doubting that.
 
Yeah prior to the 6900XT I can't remember the last time AMD was going toe to toe with Nvidia's flagship. Was hoping they would be able to do the same thing with RDNA3 but here they are back to competing with Nvidia's 2nd best instead and next gen looks to be either the same thing where they can only compete up to the 2nd best from Nvidia, or even worst where they can only compete with maybe the 3rd/4th best in Nvidia's lineup. Maybe RDNA5 can get things back to the 6900XT days but meh I'm even doubting that.
Pretty much everything tells you "no". The days of "leaps and bounds" on rasterization are over, Nvidia has said it, and I still feel it's something Nividia could effectively "leave behind" if they wanted to (let AMD and Intel fight over that market).

AI tricks as "the future", that's different. If Nvidia wants to try to compete somewhat in that space, maybe there's still a desire in the consumer space there. But honestly, good AI isn't on a singular person's desktop, not matter what Nvidia GPU they have. With that said, the concept of a "many user" approach could be interesting. While it may have zero to do with gaming, it might still end up impacting gaming, and of course many other things.

IMHO, I expect greatly reduced offerings from Nvidia in the consumer GPU space.

Intel is the obvious wild card. I get it (world)... "desktop suck, me only want laptop"... etc...etc... But if "desktop" does matter, I still Intel crushing things there by offering a CPU + discrete GPU combo for less than an AMD GPU.

RDNA5 is likely more about fending off Intel than anything Nvidia, especially if Nvidia decides to reduce their interest in that space.
 
Rumours/speculations started that top RDNA 5 card will be professional only & not for gamers.(that is the cards could be priced so high compared to their gaming value that only professionals can afford them) Need to watch this space !!!

There is still a long way to go. Anything can happen !!
 
Last edited:
Does this mean everyone has given up on RDNA4 before it has even come out?
 
Does this mean everyone has given up on RDNA4 before it has even come out?
Rumor on RDNA 4 is that it will have 2 dies. A monolithic 7800xt on 4nm & a monolithic 7600 xt on 4nm

Performance likely to be in range of 4070 ti/4070 ti super & 4060 ti respectively

(This thread is on RDNA 5. Speculation is that it could be multi-GCD but maybe it has failed like crossfire & hence AMD beginning to pivot towards professional for the Multi-GCD cards 🤔 🤔🤔)
 
Rumours/speculations started that top RDNA 5 card will be professional only & not for gamers.(that is the cards could be priced so high compared to their gaming value that only professionals can afford them) Need to watch this space !!!

There is still a long way to go. Anything can happen !!

Would it be according to the rumors because it will launch not far after the RDNA 4 and launching bigger RDNA 4 could be good enough to participate in the the xx80 type of market like the xt-xtx do right now, until smaller RDNA 5 launch later on ?
 
Would it be according to the rumors because it will launch not far after the RDNA 4 and launching bigger RDNA 4 could be good enough to participate in the the xx80 type of market like the xt-xtx do right now, until smaller RDNA 5 launch later on ?
RDNA 4 (Navi 48) would have same die size as 4070 ti (monolithic 7800xt on 4nm). So performance unlikely to be higher than a 4070 ti super.

If we assume that above card launches in july this year (& 7800xt still selling for $450) then RDNA 4 max price could be $600

That leaves $700 to $1000 market open for a over-clocked Navi 31 refresh with better memory/cache etc. (No rumors of such a revision, yet. Purely my guess/speculation)

RDNA 5 would take 2 years. Assuming the top card for monolithic GCD would be $1200 & other cards dropping below but launching only in H1 2026

(Basically trailing Blackwell competition by 1 year)
 
Rumours/speculations started that top RDNA 5 card will be professional only & not for gamers.(that is the cards could be priced so high compared to their gaming value that only professionals can afford them) Need to watch this space !!!

There is still a long way to go. Anything can happen !!
Agreed. We need to talk about this just like we did about RDNA 3 pre-launch /s
 
RDNA 5 would take 2 years. Assuming the top card for monolithic GCD would be $1200 & other cards dropping below but launching only in H1 2026
I thought the fast RDNA 4 was because it would take too long and what was planned to be RDNA 4 initially will launch as RDNA 5 not far in its back eventually.

RDNA 4 (Navi 48) would have same die size as 4070 ti (monolithic 7800xt on 4nm). So performance unlikely to be higher than a 4070 ti super.
Yes, but could they not make it bigger if wanted, is the current rumor expecting that nothing with a better performance than a 7900xtx that is not a ridiculously high-priced option (do they mean more than a current 4090 by that ?) will exist before Q1 2026 from AMD ?

Leaving the old 4090-and possibly the new 5070-5080 cards all alone with no competition all this time ? When a smaller RDNA 5 (or bigger rdna 4) could compete with them ?
 
Last edited:
Leaving the old 4090-and possibly the new 5070-5080 cards all alone with no competition all this time ? When a smaller RDNA 5 (or bigger rdna 4) could compete with them ?
True. The rumours seem to be missing some detail

Apparently the 2 RDNA 4 chips (navi 44 & navi 48) are meant primarily for laptops. But the catch is that navi 48 has replaced navi 43 which was the initial design. So what is happening there? No idea!!

I believe no chance of RDNA 5 coming before 2 years as it would/should be on 3nm

Only thing, I can think of that fills the gap is a revision of 7900xt & 7900xtx to get 7950xt & 7950xtx.
 
I believe no chance of RDNA 5 coming before 2 years as it would/should be on 3nm
3nm product launched fall of 2023 (Iphone first, than M3 laptop-iMac), Nvidia 3nm product line should launch this year with the Blackwell pro-line and possibly late 2024 for the gaming line or early 2025.

Amd 3nm product launching in 2025 should be quite possible, it will be a full 2 years mature in the consumer field process by then, TSMC 2nm non-Apple product should start to be announced in 2025 and Apple 2nm product should launch in 2025.

is a revision of 7900xt & 7900xtx
The xtx is already a full Navi 31 GCD and MCD wise I think, maybe they can boost clock and watts quite a bit too.

Is there a source for that rumored product timeline ? Maybe I am just not visualizing it correctly.
 
Last edited:
Is there a source for that rumored product timeline ? Maybe I am just not visualizing it correctly.
I think rumours exist for RDNA 5 timeline (end 2025) & RDNA 4 (H2 2024. As early as jun/july )

No rumours of revision of 7900xt/7900xtx yet. I am just thinking it would be needed to fill the gap till RDNA 5 launches in 2 years
 
Back
Top