When will APU's get so good that they outdo GPU's generally? Say for games, rendering and stuff. They seem more compact and battery efficient too, given they don't have to power an entire GPU. T. clueless but wants to know more about APU's
>>108546884AMD's 780m igpu on rdna3 architecture gave you about the same performance as an rx580 gaming wise on dual channel lpddr5x 7500MT/s, their current 8060s igpu which is rdna3.5 a refresh with some of the features from rdna4 but not all is about on par with a laptop rtx4060~4070 depending on the task but those are all being fed with quad channel lpddr5x 8000MT/s.Seeing how greedy they've been about memory channels I don't expect we're going to see APUs replacing midrange GPUs on desktops any time soon because that would cut into HEDT sales allowing more than 2 memory channels on consumer parts.
soon, but search engine says 5600G iGPU is like third of RX480 GPU performance
>>1085490695600G uses a Vega iGPU, while it was much better than intel's built in graphics at the time it's quite weak compared to dedicated gpus, especially now days. It's very far behind their current iGPUs.
>>108546884>When will APU's get so good that they outdo GPU's generally? Say for games, rendering and stuff.never>They seem more compact and battery efficient too, given they don't have to power an entire GPU. yeah, that's why they suckthere's literally no point to them beyond driving the display with the gpu off
>>108549831Idk, xe3 is around 1080ti opencl perf, console GPUs are also Apus normally, Nvidia seems like they want to make Intel / Nvidia igpu monsters (check out how Intel's chiplets have almost modularity to them and Nvidia has now a stake at Intel ). Outside of hardcore gaming we are at most 2 - 3 years from Apus being good enough for everybody it seems, 100 - 150k opencl geekbench just has to be enough to drive most games at least raster, path tracing could be mandated though like rt to drive gpu sales
>>108546884what's an apu? like an indian gpu or something?
>>108551473>console GPUs are also Apus normallyThat's for better DRM and not making AMD share IP with nvidia/intel only.Having The GPU and CPU on the same chip is more expensive. There is literally no upside in PCs other than>>108549831>driving the display with the gpu off
>>108546884>When will APU's get so good that they outdo GPU's generally?Probably with DDR6. High end DDR5 gives you 80-90GB/s bandwidth, that's a lot but not enough to compete with discrete cards. If DDR6 doubles that then we have enough to outdo low-midrange cards completely. Or you can go the soldered RAM route with Strix Halo and Medusa Halo. MH GPU tiles will be re-used as discrete GPUs, so technically at that point APUs reached parity with x60/x70 tier GPUs on account of both using the exact same chips.I think the bigger problem is that AMD doesn't have any decent GPUs so even when they do a good APU, it'll still be several tiers below Nvidia.>>108549122>>108546930I had a 5600G, RX580, and currently have a 8600G. 5600G was pretty weak compared to the RX580, but the 8600G feels around par with it. Probably a bit slower in benchmarks but in games it feels about the same. The up side is that the 8600G alone uses so little power. Using a 7600/9600 is +15W even in idle, and the RX580 is another +15W in idle. With the 8600G I'm down to something like 20-25W idle. Not as good as Intel (which can go so low that ATX PSUs can't keep up and you need a Pico PSU), but given how retarded strong the iGPU is, it's a much better deal.really wish AMD did not fuck up the Gorgon Point rollout so much, they don't even have a 880m chip out, let alone a 890m one.
>>108546884Look up intel arc b390. It's pretty good but it's really expensive right now.
>>108549831>yeah, that's why they suckOnly because all current APUs are repurposed mobile chips. Except for shit like Strix Halo, where you have a full GPU die + zen chiplets. They work that crap out so it fits in the socket, and they'll be not too much more expensive and can fit giant size GPUs there.The only problem remaining is that DDR5 is simply not fast enough for it, it's why Strix Halo uses soldered quad channel ram. But with DDR6 they'll be able to reach as much as 200GB/s with just two ram sticks which will be enough for anything south of a x070 in performance.Plus you can put as much as 128GB or more there, which is a big bonus for AI garbage, so I don't see why AMD wouldn't go in that direction.Imagine a 40CU igpu + 12c/24t Zen 7 cores + x3d cache all in one socket with DDR6. Would be a dream SFF machine, you could get it down to the size of a steam cube, but be 4x more powerful and have way the fuck more connectivity.
>>108555129The die size would be bigger than your mother.
>>108557968No it wouldn't. The Halo APUs use chiplets, and Medusa Halo is going to have a die they will also use for the x060 and x070 discrete GPUs. It'll be literally a dGPU die + zen 6 chiplets.With zen 7 they are moving to a new socket so there's a chance they'll change things so it's large enough to put the halo APUs to desktop, with DDR6 you have fast enough RAM.
>>108551657There's a small apustaja that's running the GPU essentially
>>108555129Honestly this is what amd should do. Just make the cpu and gpu the same massive chip. Undercut the fuck out of nvidias xx60 cards by having an integrated solution that is as good as a xx60 card.
>>108546884Either Medusa Halo with RDNA5 next gen or Titan Lake with Nvidia igpu the next next gen.
>>108560312>Just make the cpu and gpu the same massive chip.Can't be done, AMD can't afford it, just look at their GPUs. They've been trying to push chiplets with the 7900 for the same reason, they are unable to fab giant chips. It's difficult enough for Nvidia but they charge 10k$ per chip in return.It's why Medusa Halo will share GPU dies with the RDNA x060 and x070 graphics cards (technically the other way around, they'll put Medusa Halo iGPUs into discrete cards with LPDDR5).
