21–30 of 41 posts
How come integrated graphics is in the CPU, rather than being part of the chipset? For an actual single-chip SoC, i suppose it has to be, but even my Ryzen 5 7600X has graphics. I would have thought resources on the CPU would be at a premium, so you'd put them all towards compute. Particularly since integrated graphics doesn't need to be that powerful.
How come integrated graphics is in the CPU, rather than being part of the chipset? For an actual single-chip SoC, i suppose it has to be, but even my Ryzen 5 7600X has graphics. I would have thought resources on the CPU would be at a premium, so you'd put them all towards compute. Particularly since integrated graphics doesn't need to be that powerful.
How come integrated graphics is in the CPU, rather than being part of the chipset? For an actual single-chip SoC, i suppose it has to be, but even my Ryzen 5 7600X has graphics. I would have thought resources on the CPU would be at a premium, so you'd put them all towards compute. Particularly since integrated graphics doesn't need to be that powerful.
Today's northbridge (aka: Memory controller) is on CPUs. GPUs need a powerful memory controller. And the most powerful memory controller between CPU and Southbridge/Chipset is the memory controller on the CPU itself.
How come integrated graphics is in the CPU, rather than being part of the chipset? For an actual single-chip SoC, i suppose it has to be, but even my Ryzen 5 7600X has graphics. I would have thought resources on the CPU would be at a premium, so you'd put them all towards compute. Particularly since integrated graphics doesn't need to be that powerful.
More generally there isn't really a place for low performance integrated graphis any more and southbridge style chips are made with old old cheap processes that cripped with poor memory access would probably not run any modern desktop well. A second option for the memory would be to put a small amount of local memory on the mobo along with the chipset, which again would be slow and still costly while losing the normal iGPU advantage of unified gpu & cpu memory access to th same data (UMA).
Valve, could we please get a Ryzen 390-ish Steamdeck platform refresh? Or maybe if Intel wants game devs to test on their GPUs they'd cut a small deal to make an Intel Xe refresh.
My guess is that Valve will wait at least 1 more generation, since Zen 6 will be on N2p which will be the first nanosheet generation (with promised 30% power reductions compared to the n3e that zen5 uses). Valve clearly is more interested in a console model with new releases only when there is a massive performance uplift compared to the prior generation, so I think we are at least 1 gen away. Specifically, I think t…
Maybe Zen 6. AMD already provides the Z-Series specifically for handhelds. The Steamdeck uses currently the regular Zen2 with a RNDA2 from Zen3+. Best would be passive cooling, which would probably run well with native Linux ports like Counter-Strike 2. But I'm worried that they need to use fans again.
How come integrated graphics is in the CPU, rather than being part of the chipset? For an actual single-chip SoC, i suppose it has to be, but even my Ryzen 5 7600X has graphics. I would have thought resources on the CPU would be at a premium, so you'd put them all towards compute. Particularly since integrated graphics doesn't need to be that powerful.
Strix Halo is the powerful-GPU product version. More generally there isn't really a place for low performance integrated graphis any more and southbridge style chips are made with old old cheap processes that cripped with poor memory access would probably not run any modern desktop well. A second option for the memory would be to put a small amount of local memory on the mobo along with the chipset, which again would…
I think Strix Halo is up to 44 CUs? Which is more than a 7600xt but less than a 7700xt.
So a bit on the low-end in the scheme of dGPUs. But Strix Halo might be the most powerful iGPU ever made.
Earlier quoted context omitted.
What’s AMD's strategy in not having consumer chip support for ROCm? It's puzzling. No way to get critical mass of development interest if the bar to entry is high.
I think that part of the issue is the split between CDNA for data centers [1] and RDNA for consumer products [2] with AMD only having the money to focus on the bigger data center market. There are rumors that both architectures will be merged into UDNA in the future, which will hopefully improve ROCm support, but for now it's lacking [1] https://www.amd.com/en/technologies/cdna.html [2] https://www.amd.com/en/technol…
It's not rumor. It came straight from an executive: https://www.tomshardware.com/pc-components/cpus/amd-announce...
I think I’d buy something with Strix Halo or Strix Point if there was official ROCm support. As of 6.4.1 from earlier this month there’s still not, as I understand it. I’d be delighted to be corrected on this matter.
There is (unofficial) ROCm support for Strix Halo with ROCm 6.4.1. But like Llama.cpp and such were seg faulting but ROCR-based OpenCL was working and other workloads. ROCm GPU Compute Performance With AMD Ryzen AI MAX+ "Strix Halo": https://www.phoronix.com/review/amd-strix-halo-rocm-benchmar...
Earlier quoted context omitted.
Strix Halo is the powerful-GPU product version. More generally there isn't really a place for low performance integrated graphis any more and southbridge style chips are made with old old cheap processes that cripped with poor memory access would probably not run any modern desktop well. A second option for the memory would be to put a small amount of local memory on the mobo along with the chipset, which again would…
Powerful for iGPU. I think Strix Halo is up to 44 CUs? Which is more than a 7600xt but less than a 7700xt. So a bit on the low-end in the scheme of dGPUs. But Strix Halo might be the most powerful iGPU ever made.
edit: there's a new marketing claim from AMD that it beats M4 in some configuraton by 2.6x: https://www.amd.com/en/developer/resources/technical-article... .. but it's a small memory mode of M4 Pro, I wonder if there's independently benchmarked numbers of M4 Max vs the 395 out there.
I think I’d buy something with Strix Halo or Strix Point if there was official ROCm support. As of 6.4.1 from earlier this month there’s still not, as I understand it. I’d be delighted to be corrected on this matter.