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
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.
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.
150 Comments on AMD RDNA 5 a "Clean Sheet" Graphics Architecture, RDNA 4 Merely Corrects a Bug Over RDNA 3
1. If they do repeat the entire RDNA3 stack with fixes, the very top chips (if theoretically consider 1:1 RDNA3:RDNA4) will still be less powerful, than nVidia counterparts. And eventually would end up as lower class cards, eg 7900XT becoming 8800XT etc, which loses its premium value anyway. Unless, of course, AMD won't try to sell it at premium prices...
2. The allocation for TSMC waffers became pricey, and the foundry time, at least for higher nodes, is completely (?) reserved for Apple, and nVidia enterprise stuff. So while the waffers are fewer, it's more reasonable to make more of the smaller, more reliable dies from it.
This is just my take, which I've shared,and It may be wrong.
They want AMD to force Ngreedia in cutting prices just so they can buy cheaper Ngreedia gpus.
They never had any intentions in buying AMD gpus.
We don't know for sure whether chiplets are going to feature in future RDNA4 and RDNA5 products, but they've definitely pushed the core count beyond what monolithic CPUs could ever hope to achieve, and that's exactly what graphics cards need. The 4090 is fast because it has so many shader cores (as well as the bandwidth and power delivery to feed them) but more cores basically means more performance, which is why AMD are likely chasing the dream of splitting compute/shader arrays across multiple dies.
better RT performance alone is not worth it ( i dont play any games that has any decent RT/PT implementations. Re4 remake and LOUP1 and thats it oh wel & WD : Legion (GARBAGE) i do have GP Ultiamte and can play many MS games life FS 2020 and Forza etc but i dont really like those games (mostly use GP for Xbox 360 game like Skate/sakte 3 on Pc/& Ally
so this also has been proven that RDNA3 had bugs cuz RDNA4 is just a "BUGS FIX" Generation !!!! YAY! how about @amd .. You give me 30% of my $$$ back for RDNA3 Cards i bought or face a Class Action Lawsuit like with Bulldozer Lies////
Proves AMD KNEW RDNA3 HAD BUGGSif N48 is just a "RX 5700XT" basicly then it should only be $399 to 8700XT ? but have an 8700XTX with say more Ram (32GB Clamshell) idunno! or 8700XTX with 24Gbps GDDR6 but still 16GB but its Way faster an way higher clocked!! but has to have 16GB GDDR6 @minimum!!!
i already have a AMd refeernce 7900XT and it serves me fine... if cheap enough yeah id love to play with N44/48 but if it cost too much id just rather get anotehr ref 7900XT or even a 7900GRE (should have been the actuall 7800XT 16GB from the start)
i used to be an Big Time AMD Fan //// Im Not anmore!! they cant make RDNA APUs on AM4 :( only CRAP VEGA 8 :( and the AM5 Desktop APUs also suck ( its the best so far but? I want an 8Core 16 Threads APU with 40-60+ RDNA3.5 or RDAN4 Compute Units ( or 20-30 WGP's (Work Group Units) (Sounds better than Work Group Processors lol
It's like saying "I'm not gonna buy a 7600 because it's not faster than my 6650 XT, hmph!" Well, guess what. You're not the target audience anyway. ;)
And I don't want to spoil you all the secrets of the world but...
1. Santa Claus isn't real, and
2. No company has ever cared whether you're a fan or not, so you'd better not be, and instead, observe events from a distance and make educated buying decisions.
Speaking of RX 7900 XT. It is a big, hot and power hungry card. If AMD succeeds in making a new card within 5 or 10% the performance, and at 50-60% the power draw, that would be a huge selling point, and motivation for the users to prefer the latter.
In actual fact it is about monitor performance as well. The 7900XT is perfect for 4K 144Hz freesync panels. That was the apex but that spec has been updated with Ultrawide monitors with high refresh and pixel density rates and 4K 240Hz panels. I expect that the top end card will come with the same VRAM buffer of 20GB but the chip to be much faster to push those panels. I wonder if that would change the feel of racing Sims that run at high refresh rates already like AMS2? I know that on launch the narrative on the 7900XT was not good but with driver updates and new Games to play it is great and a card that will have you explore your library.
Also, look at the Steam hardware survey. There is no RX 7600 out there, which only proves how terrible offer it actually is.
People don't buy, because it became prohibitively expensive.
The users with RX 480 should look for RX 7800 XT, or RX 7700 XT in the worse case.
store.steampowered.com/hwsurvey/videocard/
The Steam hardware survey only shows the common mindset of "Nvidia=good, AMD=bad", so I wouldn't rely on it.
www.techpowerup.com/review/amd-rx-480/24.html
www.techpowerup.com/review/amd-radeon-rx-7600/32.html
When launched, the RX 480 8GB was only 38% behind the top dog then R9 Fury X, and GTX 1080 was 84% faster.
Today, RX 7600 is 100% behind RX 7900 XTX, and RTX 4090 is 136% faster.
Hardware power wise it was always mid-range at best, something around the level of a gtx 1060. And as far as that goes, the rx7600 is in the same spot as the rx480 was. Except that in performance-per-dollar, it shows an extremely unimpressive, low increase considering that 7 or 8 years have since passed.
In fact, the RX 6500 XT and the RX 480 launched at the same price and had almost the same performance, so eight years got you zero performance per dollar increase (and the 6500 was even missing some features the 480 had, in the video decoder as I recall).
Yes, the 6500XT was a turd that was also missing features, but it was never even supposed to be a complete desktop GPU and it likely hit the market as a desktop card simply because AMD had these chips destined to be secondary GPUs in laptops, but all the laptop manufacturers were either in lockdown, or had cancelled SKUs that these chips were destined for because of production and supply issues in China at the time. This is the same time when desktop GPUs were selling for 200% their MSRP due to scalping.
When the 6500XT was selling for $250, used RX 5700 with an MSRP of $379 were all over ebay for $1000, and selling so fast you had to be quick if you wanted one!
I have one, (missing video codec aside) its 4GB of GDDR6 performs on-par or a smidge better than an RX 580 8GB in most games.
IMO: Navi 24 was a missed opportunity... Gen4x4 mobile-derived GPU, and no one bothered to 'steal the limelight' with an M.2 GPU based on it. (Not even a Riser-Included Kit for SFF builds that the cards were clearly better-suited for)
They're power-optimised companion dies designed to hang off a 4x link to the laptop IGP in Cezanne and Rembrandt mobile APUs, which is why they're missing a bunch of video engine features and outputs - those are already in the IGP and there was no point duplicating the existing stuff. As a 6500M it's a remarkably decent offering at <50W for slim laptops. Nvidia don't really have anything in that range, you were basically going to buy a chungus plastic 3050 laptop, or pay a price premium to get a 3050/3060 in a thin-and-light. In theory there's a Geforce MX 570 based on Ampere but I've never seen one on sale, might be something that hasn't made it to the UK, France, Holland, or Singapore which are the only countries I buy from for work.Given the size of some Gen5 SSD heatsinks, I think a Navi24 M.2 GPU isn't entirely unreasonable.
For me what wrecks RT image quality is the crawling shadows from sample noise and the multiple frames of delay between something appearing in frame and the lighting and shadows looking anything close to "correct" as you'd see in a static screenshot.
Case in point, the crown-jewel of raytracing in 2024; CP2077 2.1 with RR:
How it's supposed to look, after 10-20 frames of temporal filtering without moving the camera view at all, vs how it actually looks in motion that's representative of a real gameplay experience.
Yeah, RT lighting is a dogshit splotchy mess in motion. Don't believe the "best-case-scenario" still screenshots or video where the camera is only moving forwards. Camera tilts and pans, as well as strafing sideways are all going to introduce these really really ugly, unacceptably low-quality temporal artifacts. Even people with RTX hardware seem to be conned by it because whenever they take their hand off the mouse to press the screenshot key, they stabilise the image for half a second which means the captured screenshot looks fine, even though they know it looks wrong and dirty whenever they're actually playing.
You know what, I was so focussed on RT lighting when I made those screenshots that I completely missed how bad DLSS looks in motion, too. Clearly it also needs a few frames to smooth out the image and in motion the ugliness of 720p really shines though (DLSS performance is the only way my GPU can get >60fps with path-tracing)
Here's hoping RDNA5 really focuses on image quality in motion and not just temporal nonsense that's only good for carefully curated screenshots and video that looks alright after youtube/twitch compression has taken its pound of flesh...
If you've OC'd yours I presume you've put the power limit slider all the way to the right and likely have the GPU using ~300W?