Bofferbrauer2 said:
EricHiggin said:
Ryzen mobile 4000 is also still entirely Vega I believe, just significantly optimized apparently, and 8CU's max. RDNA is quite a bit more power efficient than GCN, or was, so 45w is higher than it should be if this was the same chip using RDNA. The die size should also be smaller if it was using RDNA should it not?
I'm starting to believe the rumors of SNY helping to fund RDNA. Possibly even MS as well. Why would AMD still be using Vega and GCN for their APU's? Could be manpower and time, but it wouldn't surprise me if part of the deal is that Navi RDNA has to launch on console APU's first, before desktop or laptop APU's. This would allow AMD to launch the RDNA 5000 Series discrete GPU's last year and this year, but they would have to hold off on RDNA APU's until next year, after the consoles launch holiday 2020.
*Smart shift tech for AMD APU's should also help. Being able to shift and focus resources from CPU to GPU and vice versa should allow for even better overall performance from the hardware that ends up in the consoles.
|
Well, RDNA is just an evolution of GCN and based on the GCN instruction set. It just got heavily optimized and overhauled at some places, especially code scheduling, most notably on the Wavefront: The Wavefront shrank from 64 to 32 threads (64 threads is still supported) for better utilization and can now add new instructions every cycle to it instead every 4 cycles, and 2 CU are grouped together into a Work Group Processor, but it's base is still GCN. I thus believe that the Vega in the 4000 APUs are actually closer to Navi, just don't go all the way and are thus still called Vega.
And the reason why the amount of CU got dropped to 8 could also be in part simply because the bandwidth limitations are choking bigger chips anyway, so only putting 8 of then onto the chip makes them significantly smaller without hampering the graphics performance much.
|
JRPGfan said:
EricHiggin said:
Ryzen mobile 4000 is also still entirely Vega I believe, just significantly optimized apparently, and 8CU's max. RDNA is quite a bit more power efficient than GCN, or was, so 45w is higher than it should be if this was the same chip using RDNA. The die size should also be smaller if it was using RDNA should it not?
I'm starting to believe the rumors of SNY helping to fund RDNA. Possibly even MS as well. Why would AMD still be using Vega and GCN for their APU's? Could be manpower and time, but it wouldn't surprise me if part of the deal is that Navi RDNA has to launch on console APU's first, before desktop or laptop APU's. This would allow AMD to launch the RDNA 5000 Series discrete GPU's last year and this year, but they would have to hold off on RDNA APU's until next year, after the consoles launch holiday 2020.
*Smart shift tech for AMD APU's should also help. Being able to shift and focus resources from CPU to GPU and vice versa should allow for even better overall performance from the hardware that ends up in the consoles.
|
These APUs (4800H or 4800U) are still useing Vega, but its a much improved version, useing some of their findings from RDNA.
Supposedly these vega parts give +59% better performance than old Vega cores.
Which is why you see them say 8CU, will give higher performance (~30%) than last gens 11CU's.
|
Can't help but wonder how much Vega has actually changed then. Why not give it a different name? In terms of marketing, having a new GCN arch that's much more capable than Vega should be more enticing to buyers than simply saying we improved Vega quite a bit.
Didn't think of bandwidth starvation. Good point. AMD surely can sell as many APU's and chiplets as they can make, so the smaller they are the better. The only buyers who lose are any who want a no compromise 1080p/60 APU, but it seems like AMD doesn't want to cross that line, at least not yet.
