The team behind OptiScaler has successfully integrated support for AMD FSR 4 in a new update, allowing it to be used in any DX11/12 game. Meanwhile, a new rumor claims that an upcoming AMD UNDA/RDNA 5-based GPU will feature a memory bus similar to that of NVIDIA’s RTX 4090. No one knows for sure why Team Red hasn’t furthered its own support for FSR 4 with its RX 9000 series GPUs, but fortunately, the developers of OptiScaler continue to have your back. Any game (over 300 estimated) that supports at least AMD FSR 2, NVIDIA DLSS 2, or Intel XeSS can have OptiScaler seamlessly now run AMD FSR 4 instead of the older and less efficient upscaling libraries. Tom’s Hardware notes that users will still need to configure settings per game and perhaps tweak things as needed for best performance, so don’t expect a simple on/off toggle to be the end of the road here, but still, for those willing to put in the effort, there are significant gains to be had.
Known hardware information leaker Kepler_L2 has been speculating about AMD’s next series of GPUs. While some focus has been on what type of APU Microsoft will be using for its next Xbox, there has also been discussion on how AMD would top that for PC users. It’s been rumored that the APU could feature a graphics solution with 64 CUs with a 384-bit memory bus. Kepler has added that AMD will likely roll out its UDNA/RDNA 5 series in upgrading as follows: “64CU/256bit+32CU/128-bit to 96CU/384-bit+64CU/256-bit+32CU/128-bit.”
Kepler added, “With the Xbox(?) APU leak having 384-bit bus and what looks like 80 CU (very similar layout to Navi21), I don’t see RX 10K topping out at 64 CU,” and that the Xbox custom chip could perform very similarly to the NVIDIA RTX 5080/4090 GPUs. It was also claimed that the flagship PC AMD GPU with 96 CUs and a 384-bit memory bus could be configured similarly to Navi31 and that “I do expect them to change the advertised core count with RDNA5, since VOPD will go from best-case scenario to average-case-unless-something-weird-happened scenario.”
Discussion (16 replies)
Join Discussion →This is ofc speculation by Kepler, but he makes a fantastic point
In the next gen a console (xbox?) could have 2 different chiplets (one for cpu & one for gpu) fused together
What he is asking is could AMD be re-using the console gcd chiplet itself as its top RDNA 5 card (navi 51?)
I am of course curious how this will work... but I'm also not expecting a lot here. AMD needs to step up and show for consumer they are a prime time competitor and not the UHF holdout.
Well, if the RTX5070 performs like a RTX4090, clearly the Xbox custom APU could to it too, right?... right? :rolleyes: :rolleyes:
The xbox could be revealed/launched next november
By then we would have 1 year of the 5080 super being on the market. If the xbox comes close to that performance then the msrp of the xbox could track very close to the street price of the 5080 / 5080 super at that point in time (next year end)
So, AMD couldn't make a 9000 series card to compete with the RTX4090, but suddenly they can build an APU that can match it?
We're talking production cycle timelines and development timelines here. You think they made the 9070 chips and said.. "You know what... Job's done... everyone take a year or two off to chill."
No before the 9070 was ever out they were moving on and creating what was next. They clearly reached a determination about the capabilities of the current gen and went a different route again. But we're talking about something coming in 2026... most likely, they can learn from each other.
Now do I think they will have a 5080 equivelant graphics performance next year as part of an APU? Ehhhhh maybe. I'm more just saying that it isn't outside of the realm of possibility.
Now if MS and Sony said... make us an APU that makes us compete with X goal... and if you do it we will make you our prime CPU and GPU supplier for the next series of Data Centers we are making or upgrading for AI processing and cloud services... then AMD would double the eff down and get it done. They would be foolish not to. That would help alleviate the risk of loss on development if they had guaranteed sales.
But nobody wants to do that... because the risk is very high.
If I'm not mistaken, so far, there hasn't been a console that beats the prior gen top of the line GPU. Not saying it can't happen, but odds are against it.
Even taking the current 9070XT and "shrinking" it to fit an APU would be a monumental task.
It really depends on if AMD has some... 'new' method. Time will tell.
At this point I'm assuming the biggest improvemnts will come from "fake" frames and "AI" crap and not as much from real advancements in silicon.
I'm kinda in this camp, except for a different reason == how many 12VHPWR lines is the APU gonna require? No way that's happening in a console form factor where people will complain about it needing external power bricks and sounding like a jet plane.
Physics is a bitch, and I don't think there will have been that many node advances between the 4090 and whenever this releases.
I honestly believe that AMD has had issues with scaling GPU performance with the chiplet design. Physics as Brian_B said are holding them back. Traces between chips on a PCB can only move data so efficiently and so quickly. SO they have been held back.
Going with a complete SOC a la the Apple M series systems is clearly a solution here. Look at what they are doing with the little Ryzen Max or whatever they are called SOC's for the small form factor AI boxes.
I'd love to see someone try gaming on one of those! :) I don't think it will be great. But I DO think that will be the path forward for a new GPU solution and for consoles. (GPU solution being a slice of that in effect but scaled way up.)
And yea... I just kind of imagined a GPU for consumers with 128 gig of 'vram'. lol... not gonna happen but it popped in my head.
Console's will never compete with modern high end GPU's. They've always used software trickery to get them to run at certain resolutions and frame rates. They can't pack that much power in to a small package without it causing issues thermally or power wise.
If you undervolt a 5080 how low can the power consumption go ?
don't know, never tried it. But performance will fall off a cliff as well
Actually, depending on how low you go, performance difference is minimal, that's why undervolting is becoming very popular. But even if you undervolt power will rise automatically depending on load. So power draw will still be high, just not htat high.