• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Linux Driver Update Hints at Upcoming AMD RDNA 3.5 GPU in "Strix Point" APU

AleksandarK

News Editor
Staff member
Joined
Aug 19, 2017
Messages
2,999 (1.07/day)
In recent developments, Linux's open-source graphics ecosystem is making significant strides to accommodate AMD's upcoming RDNA3.5 architecture, also known as RDNA3+ or GFX11.5. Mesa 23.3, a library in the Linux graphics software stack, is now being updated for RDNA3.5, marking a substantial milestone. This upcoming update is particularly tailored for the impending Ryzen 8000 "Strix Point" APU series, which will incorporate the Navi 3.5 architecture. While AMD has maintained secrecy regarding specific enhancements accompanying this refresh, we expect decent performance improvements. This includes the anticipation that the Ryzen 8000 APUs will feature an increased number of Compute Units (CUs), where the current highest number is 12 CUs, and the increase could bump that figure to 16 CUs. The official announcement of the Ryzen 8000 series is expected in early 2024 when we will learn more about its GPU configuration and performance.



View at TechPowerUp Main Site | Source
 
IMO, these iGPUs are rhe most exciting thing going on in consumer graphics....it has the potential to make the financial barriers to PC gaming the lowest they've ever been, and my prediction is thwt these APUs/iGPUs will represent a bigger threat to Nvidia's consumer dGPU Dominance in certain market segments than any dGPU AMD has made in the past decade
 
I wonder if we'll get those APUs on desktop? Or we'd have to wait for someone like minisforum to make a mobo with them?
 
These RDNA generations seem to have even worse generational gap than Intel Lakes since Skylake. At least they add cache and increase frequency.
 
I wonder if we'll get those APUs on desktop? Or we'd have to wait for someone like minisforum to make a mobo with them?
A part of me hopes so. A HTPC upgrade might be imminent. Although, another part of me hopes not, because I don't want to spend more money on PC parts in the next couple of years. :laugh:
 
IMO, these iGPUs are rhe most exciting thing going on in consumer graphics....it has the potential to make the financial barriers to PC gaming the lowest they've ever been, and my prediction is thwt these APUs/iGPUs will represent a bigger threat to Nvidia's consumer dGPU Dominance in certain market segments than any dGPU AMD has made in the past decade
It doesn't seem like Strix Point will be like that, it will be more like Rembrandt -> Phoenix, so a decent improvement but nothing too special.

I would be more excited for a Strix Halo(if it exists at all) which is the big APU that would actually have the potential to take a good bite out of the dGPU market with it being 12 Zen 5 cores and 40 CUs and a 256-bit memory bus.
 
IMO, these iGPUs are rhe most exciting thing going on in consumer graphics....it has the potential to make the financial barriers to PC gaming the lowest they've ever been, and my prediction is thwt these APUs/iGPUs will represent a bigger threat to Nvidia's consumer dGPU Dominance in certain market segments than any dGPU AMD has made in the past decade

I agree that the potential is there. I think >/=1536sp/3nm clocks will be a very interesting scenario...At that point we're conceivably talking about something similar to a PS5.

That said, I'm kinda curious who gets there first, AMD or Intel. Still, whomever does it (~100W GPU + 65W CPU), if not both, will probably be successful.
 
I would be more excited for a Strix Halo(if it exists at all) which is the big APU that would actually have the potential to take a good bite out of the dGPU market with it being 12 Zen 5 cores and 40 CUs and a 256-bit memory bus.
I wonder whether this kind of APU will achieve the very initial idea that CPU part does integer and GPU part does floating point.
 
I wonder whether this kind of APU will achieve the very initial idea that CPU part does integer and GPU part does floating point.
No, that idea was abandoned years ago because, shocker, the GPU is needed to do GPU things, and GPUs SUCK at serialized tasks like CPUs are built for. AMD's GPGPU HSA attempts never produced results, at least for consumers


It doesn't seem like Strix Point will be like that, it will be more like Rembrandt -> Phoenix, so a decent improvement but nothing too special.

I would be more excited for a Strix Halo(if it exists at all) which is the big APU that would actually have the potential to take a good bite out of the dGPU market with it being 12 Zen 5 cores and 40 CUs and a 256-bit memory bus.
If strix halo actually comes out it will be very exciting. 256 bit 8533 mhz memory bus, 40 CUs, and zen 5 in one package in a 13-14 inch would be absolutely incredible.

I agree that the potential is there. I think >/=1536sp/3nm clocks will be a very interesting scenario...At that point we're conceivably talking about something similar to a PS5.

That said, I'm kinda curious who gets there first, AMD or Intel. Still, whomever does it (~100W GPU + 65W CPU), if not both, will probably be successful.
It'll be AMD, intel has been stuck on 96EU for years now, they MIGHT get 128 on meteor lake. Eventually. The 192EU chip is DOA as of now.
 
It'll be AMD, intel has been stuck on 96EU for years now, they MIGHT get 128 on meteor lake. Eventually. The 192EU chip is DOA as of now.
Yep. Intel is too busy cramming as many (E or P, doesn't matter) cores into their CPUs as they can while keeping the power consumption below a kilowatt. This might change when they switch to chiplets, we'll see.

If strix halo actually comes out it will be very exciting. 256 bit 8533 mhz memory bus, 40 CUs, and zen 5 in one package in a 13-14 inch would be absolutely incredible.
I don't think it exists. The 6 nm Navi 24 is 107 mm2 big with 1024 shader cores. There's no way AMD can shove that into a CPU even on 5 or 4 nm (I hope I'm wrong, though).
 
Moving from 12 CUs to 16 CUs? Isn't the iGPU memory bottlenecked? Will these have triple-channel RAM or something to alleviate the memory bottleneck?
Back to the 1366 days! where triple channel was amazing~~
 
Moving from 12 CUs to 16 CUs? Isn't the iGPU memory bottlenecked? Will these have triple-channel RAM or something to alleviate the memory bottleneck?
They're moving from 7500 to 8533 LPDDR5X, but bandwidth will continue to be an issue. They are also giving it a 32MB infinity cache. Maybe theyll gift us with a x3d variant.
 
They're moving from 7500 to 8533 LPDDR5X, but bandwidth will continue to be an issue. They are also giving it a 32MB infinity cache. Maybe theyll gift us with a x3d variant.
I forgot all about the Infinite Cache. This would be the third generation APU since the cache was introduced with RDNA2, so it's about time.
 
The problem with AMD APUs is that they are very late to the party. They've being released in year after the main desktop and mobile counterparts. But the availability is nowhere great even past a years from announcement/release.

And this is huge problem, since these are logically the best and most reasonable hardware people want to use as general purpose/daily driver. The demand is huge. And AMD have to fix their supply ASAP. The APU's sale alone, could if not "revive", but at least bring some cash the RTG department. IMO
 
The problem with AMD APUs is that they are very late to the party. They've being released in year after the main desktop and mobile counterparts. But the availability is nowhere great even past a years from announcement/release.

And this is huge problem, since these are logically the best and most reasonable hardware people want to use as general purpose/daily driver. The demand is huge. And AMD have to fix their supply ASAP. The APU's sale alone, could if not "revive", but at least bring some cash the RTG department. IMO
I think part of the problem is that every year AMD improves both their supply and the demand for their products. I think mobile is the biggest product line to suffer from this, because where desktop and server share dies, mobile does not. Every chip made for mobile can only be sold to mobile. And they're more expensive than the desktop chips and I suspect the buyers are the least likely market to notice the difference between a Ryzen 7040 mobile chip and a 7030.
 
I wonder whether this kind of APU will achieve the very initial idea that CPU part does integer and GPU part does floating point.
No, the latency to just launch jobs/threads on GPU is already too high for that. Not to mention that for modern AMD APUs, for the CPU to communicate with the GPU or vice-versa, it needs to pass through the main memory, they don't have a direct communication for the data(that would require a cache that is shared for both).
If strix halo actually comes out it will be very exciting. 256 bit 8533 mhz memory bus, 40 CUs, and zen 5 in one package in a 13-14 inch would be absolutely incredible.
I don't believe it would be for 13-14 inches, as it would likely be 100~200W in TDP. I think it would be great for console-like miniPCs and such. And would probably be better to use DDR5 rather than LPDDR and give the option for people to overclock memory, so they can have more memory bandwidth.

It would probably be great and really kill anything below 6500 XT for sure, with it being able to likely compete with 6600 XT class of GPUs, which if competitive in price, it could really kill those. What would be the point of getting a 4060 when you can get a whole APU with the same-ish GPU performance for only somewhat higher price?
 
I think part of the problem is that every year AMD improves both their supply and the demand for their products. I think mobile is the biggest product line to suffer from this, because where desktop and server share dies, mobile does not. Every chip made for mobile can only be sold to mobile. And they're more expensive than the desktop chips and I suspect the buyers are the least likely market to notice the difference between a Ryzen 7040 mobile chip and a 7030.
Apparently with Zen 6, APU's and CPU's will be the same unified design. So their will be no "CPU" in effect, everything will be an APU, basically they are doing an Intel.
 
Apparently with Zen 6, APU's and CPU's will be the same unified design. So their will be no "CPU" in effect, everything will be an APU, basically they are doing an Intel.
Heck, even Zen 4 is an APU now.
 
Back
Top