Monday, February 20th 2023
AMD RDNA4 Architecture to Build on Features Relevant to Gaming Performance, Doesn't Want to be Baited into an AI Feature Competition with NVIDIA
AMD's next-generation RDNA4 graphics architecture will retain a design-focus on gaming performance, without being drawn into an AI feature-set competition with rival NVIDIA. David Wang, SVP Radeon Technologies Group; and Rick Bergman, EVP of Computing and Graphics Business at AMD; gave an interview to Japanese tech publication 4Gamers, in which they dropped the first hints on the direction which the company's next-generation graphics architecture will take.
While acknowledging NVIDIA's movement in the GPU-accelerated AI space, AMD said that it didn't believe that image processing and performance-upscaling is the best use of the AI-compute resources of the GPU, and that the client segment still hasn't found extensive use of GPU-accelerated AI (or for that matter, even CPU-based AI acceleration). AMD's own image processing tech, FSR, doesn't leverage AI acceleration. Wang said that with the company introducing AI acceleration hardware with its RDNA3 architecture, he hopes that AI is leveraged in improving gameplay—such as procedural world generation, NPCs, bot AI, etc; to add the next level of complexity; rather than spending the hardware resources on image-processing.
AMD also stressed on the need to make the GPU more independent of the CPU in graphics rendering. The company took several steps in this direction over the past many generations, with the most recent being the multi-draw indirect accelerator (MDIA) component introduced with RDNA3. Using this, software can dispatch multiple instanced draw commands that can be issued on the GPU, greatly reducing the CPU-level overhead. RDNA3 is up to 2.3x more efficient at this than RDNA2. Expect more innovations along these lines with RDNA4.
AMD understandably didn't talk anything about the "when," "what," and "how" of RDNA4 as its latest RDNA3 architecture is just off the ground, and awaiting a product ramp through 2023 into the various market-segments spanning from iGPUs to mobile GPUs, and mainstream desktop GPUs. RDNA3 is currently powering the Radeon RX 7900 series high-end graphics cards, and the company's latest 5 nm "Phoenix Point" Ryzen 7000-series mobile processor iGPUs. You can catch the 4Gamer interview in the source link below.
Sources:
4Gamers.net, HotHardware
While acknowledging NVIDIA's movement in the GPU-accelerated AI space, AMD said that it didn't believe that image processing and performance-upscaling is the best use of the AI-compute resources of the GPU, and that the client segment still hasn't found extensive use of GPU-accelerated AI (or for that matter, even CPU-based AI acceleration). AMD's own image processing tech, FSR, doesn't leverage AI acceleration. Wang said that with the company introducing AI acceleration hardware with its RDNA3 architecture, he hopes that AI is leveraged in improving gameplay—such as procedural world generation, NPCs, bot AI, etc; to add the next level of complexity; rather than spending the hardware resources on image-processing.
AMD understandably didn't talk anything about the "when," "what," and "how" of RDNA4 as its latest RDNA3 architecture is just off the ground, and awaiting a product ramp through 2023 into the various market-segments spanning from iGPUs to mobile GPUs, and mainstream desktop GPUs. RDNA3 is currently powering the Radeon RX 7900 series high-end graphics cards, and the company's latest 5 nm "Phoenix Point" Ryzen 7000-series mobile processor iGPUs. You can catch the 4Gamer interview in the source link below.
221 Comments on AMD RDNA4 Architecture to Build on Features Relevant to Gaming Performance, Doesn't Want to be Baited into an AI Feature Competition with NVIDIA
And FYI overclock3d.net/news/gpu_displays/amd_s_gpu_market_share_dropped_to_an_all-time_low_in_q3_2022_-_nvidia_dominates/1
Also, there are games with RT where 7900xtx is faster than a 4070Ti like Forspoken, The Calisto protocol both at 4k but I'm guessing these games are not good to evaluate RT performance right?
Also do you want to know why the 4070ti is drawing less power in ray tracing games and the 7900XT doesn't ? It's because it's clearly bottlenecked by something else, likely memory bandwidth, what a great well balanced architecture Nvidia designed lol.Yeah bro maximum power draw is useless but I am sure everyone is picking a 4070ti over a 7900XT because of the video playback power draw, duh, now that's important stuff right there. Do you know what the power draw is when you use notepad ? I reckon that's even more important.
Clutching at straws much ?Duh, obviously.
Most people don't really care about RT at all. I am using RTX but I would take more raster perf any day over RT perf which is a joke in most games unless you buy the absolute most expensive GPU, or you will be looking at crappy framerate with RT on anyway
www.techpowerup.com/review/asus-geforce-rtx-4070-ti-tuf/32.html
The low bandwidth of 4070 Ti shows the higher the solution go, 504 GB/s is pretty low for a 799 dollar GPU and 12GB VRAM is also low in 2023.
It does not consume 30% more, haha. It consumes 12% more. 284 vs 320 watts in gaming load. And when you actually overclock those cards, 7900XT will perform even better, because Nvidia always maxes out their cards and chips so overclocking will net you a few percent performance at most.
7900 XT will age better for sure, you will see in 1-2 years. In some games, it's on par with 4080, just like 7900XTX beats 4090 in some games. In RT, no, but RASTER yes. Once again, pretty much no-one cares about ray tracing at this point. It's a gimmick and I will take 100% higher fps any day over RT.
Why are you even talking about power consumption when you have a 4090 which spikes at 500+ watts, it has terrible performance per watt and laughable performance per dollar.
Actually 4090 is the worst GPU ever in terms of perf per dollar.
www.techpowerup.com/review/asus-geforce-rtx-4070-ti-tuf/33.html
:roll:
And 4080 Ti and 4090 Ti will probably release very soon making 4090 irellevant, just like last time.
We need to be carefull about how we use that tool, (who's becoming more than a tool) a few generation down the line, we might just end up with a society addicted to get instant results, and less interested to learn stuff. Studies shows that gen Z are really not that tech literate...because they don't need to understand how something actually work to use it, it's been simplified so much.
So in that sense I like AMD statement, we don't need to use A.I for every little thing. It's a wonderfull thing for sure, but overusing it might also have bad consequences.
AMD said that it didn't believe that image processing and performance-upscaling is the best use of the AI-compute resources of the GPU
Also AMD:
AMD FidelityFX™ Super Resolution (FSR) uses cutting-edge upscaling technologies to help boost your framerates in select titles1 and deliver high-quality, high-resolution gaming experiences, without having to upgrade to a new graphics card.
www.amd.com/en/technologies/fidelityfx-super-resolutioni assume they mean they no longer what us to have high quality without having to upgrade to a new gpu. From a sales perspective it makes sense.
3080 10GB is already in trouble for 3440x1440 users
Nvidia gimping makes people upgrade faster, smart tactic
You are using 2060 6GB which barely does 1080p today, 2060 "Super" came out for a reason, now with 8GB VRAM :laugh:
You also bough into Intel 6C/6T is enough I see, sadly 6C/6T chips are choking only a few years after, 6C/12T is bare minimum for proper gaming, just like 8GB VRAM is bare minimum for 1440p and up
AMD generally gives you better longevity than both Nvidia and Intel, wake up and stop being milked so hard
The 4070ti has less memory bandwidth but a lot more L2 cache, L2 is going to be faster than the L3 AMD has on it's GPUs but then they also increased the memory bandwidth this generation as well. In other words bandwidth absolutely does matter, that's why they had to increase the L2 cache in the first place however more cache is not a complete substitute for VRAM bandwidth, the 4070ti is slower at 4K than than it is at lower resolutions, the only explanation for that is in fact the lack of memory bandwidth and possibly memory capacity as well depending on the game.
So you proved me wrong by agreeing with me that the XT draws a lot more power. Great, and yes that's usually the case, you prove me wrong every single time by admitting that everything I said is absolutely the case. Gj, keep it upYeah, that 6c12t that amd launched in 2023 for 350 gives you great longevity over the 14c Intel offers. LOL
www.hpcwire.com/2022/03/08/amd-xilinx-takes-aim-at-nvidia-with-improved-vck5000-inferencing-card/
What is being said is that they don't want to build this into the GPUs and force ordinary users to pay a lot more for something that can be adapted to run in regular shaders.
Actually if you'd use your head for a second you'd realize that what you're saying is complete nonsense anyway, a GPU is typically always facing 100% utilization, it makes no sense that a GPU with let's say 300W TDP limit would ever average out at 200W with 400W maximum readings, it just wouldn't happen. As usual your complete lack of understanding of how these things work prohibits you from ever making a coherent point.
But as I keep saying none of that matters, you're just wrong, it doesn't use 30% more power. Do you not know how to read or are you purposely ignoring this ?Completely delusional.
Your welcome.
I'm not sure why you even brought up CPU's but launch prices are pointless, only people that always buy the latest thing care about launch prices. The 7600X has 6 performance cores, and now sells for less than $250, Intel is still charging over $300 for 6 performance cores, also you get less upgrades from an Intel board.
If you have a 300W average, a 300W limit and a 400W power maximum then that means the contribution of that 400W maximum figure to the average is basically none whatsoever and the power limit is doing it's job as it's supposed to. That's why your example is dumb and nonsensical, this isn't a matter of opinion, you just don't know math.
There isn't a single card on those charts which has a disparity between average and maximum that big, nonetheless going by maximum is still preferable, for instance it's useful in choosing a power supply. You think it's useless because you simply don't know what you're talking about.
He said "with the company introducing AI acceleration hardware with its RDNA3 architecture, he hopes that AI is leveraged in improving gameplay—such as procedural world generation, NPCs, bot AI, etc; to add the next level of complexity; rather than spending the hardware resources on image-processing."
Isn't it something that's up to the game and CPU entirely ?
Does it mean that they will now go towards a more bruteforce approach ? like full raster performance instead of going down the AI-path like Nvidia ?
I'm confused
Or even better. Downclock it 2000Mhz and then measure. But when they check how fast can it render then obviously no limits but then they do not bring the power consumption up since it is irrelevant. :laugh:
- Strawman: fabricating AMD's L3 cache as the same thing as Nvidia's L2.
- Loaded question: claiming that I somehow "hailed" a performance enhancement to try to make me appear to either blindly support it or shy away thus somehow "retracting" the statement about Nvidia's L2 cache.
- Black-or-white: implying that my criticisms of Nvidia automatically make me blindly agree with allow any of AMD's actions.
- Ambiguity: ignoring that AMD's L3 cache is cheaper to implement versus Nvidia's L2 cache using direct die space.
In short: AMD's implementations are more generic and better suited for long-term performance whereas Nvidia's is a knee-jerk reaction to do everything it can to have the "best of the best" in order to work off of the mindless sports mentality of "team green" versus "team red".The thing also is on the gamers side: You don't need huge amount of raw "AI compute power". You just run the trained model on the actual data. You don't train it live. The learning pass require way more processing power.
The thing is Nvidia do not have a specialized compute/AI architecture like AMD. (They have a brand, but it use the same architecture as gamers GPU most of the time).
In the end, who know what strategy is the best. Nvidia with less monolithic chips and architecture but large space of die used for AI that is wasted most of the time for gamers or AMD with 2 architectures and a specialized gamers architecture more tailored for the current workload.
We have also to consider that there is ASIC in the loops that destroy GPU application.
In the end, i do not know. I am pretty sure those big boss of tech company sees things we do not see. But i don't think that AMD move is bad. They just say our gamers cards won't compete on RAW AI power vs Nvidia. They don't say their CNDA arch won't compete or that they won't have ai acceleration on RDNA GPU.
Just that it won't be a focus.
Just report it they say, I did, it didn't work now I'm here.
As for AI I fully align with AMD.
Use it for in game adversarial intelligence because using it to invent extra frames is arse IMHO especially given NPC and in game enemies are total shite.
Instead of an intelligent team to fight or a clever boss we get refurbished boss fights that are just about learning different attack patterns, Great.
Companies don't get to decide what is gimmick, consumers will do that. Looks like AMD is following Nokia lead