Is it an APU, or is it a “desktop” CPU and GPU on one board? CPU specs are close to the 7600x but downlocked. And with dedicated vram I’d assume the GPU is it’s own separate thing.
GPU looks like it’s probably a tweaked RX 7400 based on the specs.
This seems to blur the lines between desktop and mobile APU’s, but I would bet that’s it’s closer to a higher clocked mobile chip, than it is to desktop. The only reason I think this is the case is due to the similarity spec wise with the Max 385, and that it’s semi-custom.
If it was just a 7600x CPU + 7600 GPU I think they would have just said so.
It could be separate CPU+GPU, but I think it might be possible that it is built more like a SOC, where the GPU is just given its own dedicated VRAM.
Looking at the hardware of say a PS5, it has 16 GB of GDRR6, the same as the Steam Machine’s VRAM.
If everything is soldered anyway, there is no reason to have separate chips for CPU+GPU, especially if that hardware already exists like the AMD Ryzen AI Max line.
Well I’m probably wrong then, framework said they couldn’t get good performance and maintain signal integrity with upgradable memory for the Ryzen Max cpus, so this is likely discrete Cpu and GPU. Probably all soldered in the same mainboard though.
Well I’m probably wrong then, framework said they couldn’t get good performance and maintain signal integrity with upgradable memory for the Ryzen Max cpus
On the other hand, Framework is run by far right sympathizers and are a few billion short of what Valve’s R&D might have access too.
If everything is soldered anyway, there is no reason to have separate chips for CPU+GPU, especially if that hardware already exists like the AMD Ryzen AI Max line.
Cost is a factor because just as with Steam Deck the two SKUs will only differ in storage space, not in performance. Using last gen RDNA3 is 100% a cost driven choice.
There was the story recently that AMD demanded a very high minimum order (10 million or so?) for semi-custom versions of the lasest Ryzen and RDNA iterations for some Xbox handheld which is unlikely that handheld would sell.
By going this route, Valve avoided this. Surely there is spare manufacturing capacity for RDNA3 by now.
I would have thought unified memory would pay off, otherwise you spend your time shuffling stuff between system memory and vram. Isn’t the deck unified memory?
What you lose shuffling between CPU and GPU you gain by not having your GPU and CPU sharing the same bandwidth.
Apple gets away with it by having an ungodly massive memory bus. I don’t think valve is getting a 512 bit memory bus on what’s probably a RX 7400/Ryzen 7600 tier CPU. Both of those combined would be like half that?
Apple gets away with it by having an ungodly massive memory bus.
It’s kind of impressive how effective Apple’s marketing team was towards developers when they started that push towards ARM PCs. A lot of people can remember that having shared memory benefits from not having to copy memory between the CPU and GPU, but barely any of them remember that the only reason it’s feasible is because Apple gave their devices insanely high memory bandwidth.
On the opposite end of the spectrum, look no further than the original Nintendo Switch. With an incredible 64-bit memory bus and 1600MHz memory clock speed, it was already being bottlenecked by its memory bandwidth 2 years into its lifespan. And that’s counting first/second-party titles like the Link’s Awakening remaster, not even shitty ports of games made for other consoles.
Is it an APU, or is it a “desktop” CPU and GPU on one board? CPU specs are close to the 7600x but downlocked. And with dedicated vram I’d assume the GPU is it’s own separate thing.
GPU looks like it’s probably a tweaked RX 7400 based on the specs.
To keep the package small, they might still have laptop type discreet GPU, just integrated on the same board.
This seems to blur the lines between desktop and mobile APU’s, but I would bet that’s it’s closer to a higher clocked mobile chip, than it is to desktop. The only reason I think this is the case is due to the similarity spec wise with the Max 385, and that it’s semi-custom.
If it was just a 7600x CPU + 7600 GPU I think they would have just said so. It could be separate CPU+GPU, but I think it might be possible that it is built more like a SOC, where the GPU is just given its own dedicated VRAM.
Looking at the hardware of say a PS5, it has 16 GB of GDRR6, the same as the Steam Machine’s VRAM.
If everything is soldered anyway, there is no reason to have separate chips for CPU+GPU, especially if that hardware already exists like the AMD Ryzen AI Max line.
According to Dave2D’s review, RAM is upgradeable, and GPU has dedicated VRAM.
Well I’m probably wrong then, framework said they couldn’t get good performance and maintain signal integrity with upgradable memory for the Ryzen Max cpus, so this is likely discrete Cpu and GPU. Probably all soldered in the same mainboard though.
On the other hand, Framework is run by far right sympathizers and are a few billion short of what Valve’s R&D might have access too.
Cost is a factor because just as with Steam Deck the two SKUs will only differ in storage space, not in performance. Using last gen RDNA3 is 100% a cost driven choice.
There was the story recently that AMD demanded a very high minimum order (10 million or so?) for semi-custom versions of the lasest Ryzen and RDNA iterations for some Xbox handheld which is unlikely that handheld would sell.
By going this route, Valve avoided this. Surely there is spare manufacturing capacity for RDNA3 by now.
I would have thought unified memory would pay off, otherwise you spend your time shuffling stuff between system memory and vram. Isn’t the deck unified memory?
What you lose shuffling between CPU and GPU you gain by not having your GPU and CPU sharing the same bandwidth.
Apple gets away with it by having an ungodly massive memory bus. I don’t think valve is getting a 512 bit memory bus on what’s probably a RX 7400/Ryzen 7600 tier CPU. Both of those combined would be like half that?
It’s kind of impressive how effective Apple’s marketing team was towards developers when they started that push towards ARM PCs. A lot of people can remember that having shared memory benefits from not having to copy memory between the CPU and GPU, but barely any of them remember that the only reason it’s feasible is because Apple gave their devices insanely high memory bandwidth.
On the opposite end of the spectrum, look no further than the original Nintendo Switch. With an incredible 64-bit memory bus and 1600MHz memory clock speed, it was already being bottlenecked by its memory bandwidth 2 years into its lifespan. And that’s counting first/second-party titles like the Link’s Awakening remaster, not even shitty ports of games made for other consoles.