Monday, May 13th 2024

AMD RDNA 5 a "Clean Sheet" Graphics Architecture, RDNA 4 Merely Corrects a Bug Over RDNA 3

AMD's future RDNA 5 graphics architecture will bear a "clean sheet" design, and may probably not even have the RDNA branding, says WJM47196, a source of AMD leaks on ChipHell. Two generations ahead of the current RDNA 3 architecture powering the Radeon RX 7000 series discrete GPUs, RDNA 5 could see AMD reimagine the GPU and its key components, much in the same way RDNA did over the former "Vega" architecture, bringing in a significant performance/watt jump, which AMD could build upon with its successful RDNA 2 powered Radeon RX 6000 series.

Performance per Watt is the biggest metric on which a generation of GPUs can be assessed, and analysts believe that RDNA 3 missed the mark with generational gains in performance/watt despite the switch to the advanced 5 nm EUV process from the 7 nm DUV. AMD's decision to disaggregate the GPU, with some of its components being built on the older 6 nm node may have also impacted the performance/watt curve. The leaker also makes a sensational claim that "Navi 31" was originally supposed to feature 192 MB of Infinity Cache, which would have meant 32 MB segments of it per memory cache die (MCD). The company instead went with 16 MB per MCD, or just 96 MB per GPU, which only get reduced as AMD segmented the RX 7900 XT and RX 7900 GRE by disabling one or two MCDs.
The upcoming RDNA 4 architecture will correct some of the glaring component level problems causing the performance/Watt curve to waver on RDNA 3; and the top RDNA 4 part could end up with performance comparable to the current RX 7900 series, while being from a segment lower, and a smaller GPU overall. In case you missed it, AMD will not make a big GPU that succeeds the "Navi 31" and "Navi 21" for the RDNA 4 generation, but rather focus on the performance segment, offering more bang for the buck well under the $800-mark, so it could claw back some market share from NVIDIA in the performance- mid-range, and mainstream product segments. While it remains to be seen if RDNA 5 will get AMD back into the enthusiast segment, it is expected to bring a significant gain in performance due to the re-architected design.

One rumored aspect of RDNA 4 that even this source agrees with, is that AMD is working to significantly improve its performance with ray tracing workloads, by redesigning its hardware. While RDNA 3 builds on the Ray Accelerator component AMD introduced with RDNA 2, with certain optimizations yielding a 50% generational improvement in ray testing and intersection performance; RDNA 4 could see AMD put more of the ray tracing workload through fixed-function accelerators, unburdening the shader engines. This significant improvement in ray tracing performance, performance/watt improvements at an architectural level, and the switch to a newer foundry node such as 4 nm or 3 nm, is how AMD ends up with a new generation on its hands.

AMD is expected to unveil RDNA 4 this year, and if we're lucky, we might see a teaser at the 2024 Computex, next month.
Sources: wjm47196 (ChipHell), VideoCardz
Add your own comment

169 Comments on AMD RDNA 5 a "Clean Sheet" Graphics Architecture, RDNA 4 Merely Corrects a Bug Over RDNA 3

#151
ARF
The DGPU roadmap has been leaked:



RDNA 5 GFX13 3nm GDDR7 very late 2026 or early 2027. AMD abandons the ultra high-end segment, namely what RX 6950 XT and RX 7900 XTX are, and the entry level, namely where RX 6400 is.

For me, it is a really bad roadmap.
Posted on Reply
#152
TumbleGeorge
ARFThe DGPU roadmap has been leaked:



RDNA 5 GFX13 3nm GDDR7 very late 2026 or early 2027. AMD abandons the ultra high-end segment, namely what RX 6950 XT and RX 7900 XTX are, and the entry level, namely where RX 6400 is.

For me, it is a really bad roadmap.
Rather, there are not enough details yet, and that is why the field from the end of 2026 and beyond is like this.
Posted on Reply
#153
AusWolf
TumbleGeorgeRather, there are not enough details yet, and that is why the field from the end of 2026 and beyond is like this.
Considering that no chip is named (Navi 5 is a product family, not a chip), I think this is the case. How much of the vertical segment is covered by "Navi 5" is inconsequential.
Posted on Reply
#154
ARF
If the trend continues (Navi 2 = 4 chips, Navi 3 = 3 chips, Navi 4 = 2 chips, Navi 5 -> ... ), one can expect Navi 5 to be one single chiplet that will serve and cover all market tiers.
Remember the Raja Koduri's original promise, there was one thing called "scalability".

Navi 58 - 4 chiplets.
Navi 56 - 3 chiplets.
Navi 54 - 2 chiplets.
Navi 53 - 1 chiplet.

Posted on Reply
#155
TumbleGeorge
Ah, I don't remember that as a promise. At best, shared intentions. But how does this continue to matter today, since the gentleman has long been working for another company?
Posted on Reply
#156
AusWolf
ARFIf the trend continues (Navi 2 = 4 chips, Navi 3 = 3 chips, Navi 4 = 2 chips, Navi 5 -> ... )...
That's not a trend, just a coincidence. Remember, Navi 1 was also 2 chips. Sure it would be good for AMD to achieve total scalability with chiplets for Navi 5, but whether it'll happen or not, we'll see.
Posted on Reply
#157
ARF
AusWolfThat's not a trend, just a coincidence
Definitely not a coincidence. Can't be a coincidence, but rather a carefully executed plan to decrease the projects costs, and in the future - maybe exit the market segment altogether. We will see.
Depends on whether they will be able to create an extremely fast interconnect between the chiplets.
Posted on Reply
#158
AusWolf
ARFDefinitely not a coincidence. Can't be a coincidence, but rather a carefully executed plan to decrease the projects costs, and in the future - maybe exit the market segment altogether. We will see.
Depends on whether they will be able to create an extremely fast interconnect between the chiplets.
Decreasing project costs sounds like a sensible plan, but it's not a necessary step to leave the business.
Posted on Reply
#159
ARF
AusWolfDecreasing project costs sounds like a sensible plan, but it's not a necessary step to leave the business.
Today, AMD has so much money to afford high quality projects in the graphics department, that its intention to not compete with the black-leather jacketed man is really bizarre.
It is actually becoming worse by the year. AMD is virtually absent in the OEM market, including laptops.
Posted on Reply
#160
AusWolf
ARFToday, AMD has so much money to afford high quality projects in the graphics department, that its intention to not compete with the black-leather jacketed man is really bizarre.
If you think that not competing on the high-end is the same thing as not competing at all because only x90 halo cards matter, then I can't entertain this conversation any further.
Posted on Reply
#161
ARF
AusWolfIf you think that not competing on the high-end is the same thing as not competing at all because only x90 halo cards matter
That is not the point. If it turns out that the chiplet designs don't work at all, then all market segments will suffer, and will take the hit.
We will see if they will invent the breakthrough technology needed to make GPU chiplets work without very significant performance loss.
Posted on Reply
#162
AusWolf
ARFThat is not the point. If it turns out that the chiplet designs don't work at all, then all market segments will suffer, and will take the hit.
We will see if they will invent the breakthrough technology needed to make GPU chiplets work without very significant performance loss.
So far, the only problem of the chiplet design is idle / low load (video playback) power consumption, and the fact that they can't split the main components into chiplets without adding latency that a GPU architecture can't bear. Rumour says that Navi 4 will be monolithic, but we'll see if they manage to get around the above with Navi 5.
Posted on Reply
#163
ARF
AusWolfSo far, the only problem of the chiplet design is idle / low load (video playback) power consumption, and the fact that they can't split the main components into chiplets without adding latency that a GPU architecture can't bear.
Idle / videoplayback high power consumption is because of engineering incompetence, not because of the chiplets. AMD has done it for generations, actually I remember that I upgraded Radeon HD 4890 1GB to Radeon HD 6870 1GB (gaining what, measly 30% more performance), simply because the newer card had the memory clocks fixed. :banghead:
Posted on Reply
#164
AusWolf
ARFIdle / videoplayback high power consumption is because of engineering incompetence, not because of the chiplets. AMD has done it for generations, actually I remember that I upgraded Radeon HD 4890 1GB to Radeon HD 6870 1GB (gaining what, measly 30% more performance), simply because the newer card had the memory clocks fixed. :banghead:
I'm not so sure considering that idle power consumption of their CPUs has been high ever since chiplets are a thing, and hasn't been fixed.
Posted on Reply
#165
ARF
AusWolfI'm not so sure considering that idle power consumption of their CPUs has been high ever since chiplets are a thing, and hasn't been fixed.
For some reason they want to maintain the clocks extremely high. If an intel CPU can downclock to 900 MHz, or even lower to 450 MHz, the AMD CPUs don't fall under 3.9 GHz.
That's a nasty design choice. Have you heard why they want this to happen?
Posted on Reply
#166
AusWolf
ARFFor some reason they want to maintain the clocks extremely high. If an intel CPU can downclock to 900 MHz, or even lower to 450 MHz, the AMD CPUs don't fall under 3.9 GHz.
That's a nasty design choice. Have you heard why they want this to happen?
CPU core clocks aren't the problem. Even at high clocks, they eat peanuts. The IO die is the problem.
Posted on Reply
#167
ARF
AusWolfCPU core clocks aren't the problem. Even at high clocks, they eat peanuts. The IO die is the problem.
This means they have a problem with the north bridge. They have to learn how to design one.
Posted on Reply
#168
fb020997
Space LynxMost likely I will keep my 7900 XT until RDNA5, I have plenty of backlog and don't give a damn about Ray tracing or Physx or any of the nonsense that has come and gone before.
I’m waiting for at least 2.5-3 times the performance on raster and at least 5 years before I change my 7900XT, as I did from my Vega 64 to 7900XT upgrade (mid ‘23).
Posted on Reply
#169
AusWolf
fb020997I’m waiting for at least 2.5-3 times the performance on raster and at least 5 years before I change my 7900XT, as I did from my Vega 64 to 7900XT upgrade (mid ‘23).
The way it should be. :)

I don't understand people complaining about not having massive uplifts across each generation. Are they really so desperate to spend money? :D
Posted on Reply
Add your own comment
Nov 23rd, 2024 11:33 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts