7770 has 10 CU while Durango is supposed to have 12, and durango has to feed 8 cpu cores too
so your saying the bandwidth speed is not enough? i thought durango had 2 cores reserved for the os.
7770 has 10 CU while Durango is supposed to have 12, and durango has to feed 8 cpu cores too
7770 has 10 CU while Durango is supposed to have 12, and durango has to feed 8 cpu cores too
Yeah, let's ignore the eDRAM and pretend this thing is working solely from DDR3...7770 has 10 CU while Durango is supposed to have 12, and durango has to feed 8 cpu cores too
7770 has 10 CU while Durango is supposed to have 12, and durango has to feed 8 cpu cores too
But Durango also 102GB/s ESRAM to assist with GPU available bandwidth. So it's very likely it'll outperform a 7770 in every case.
I thought it was 68GB's
I know it were 8 cycles with the VLIW architectures, but is this documented or measured somewhere for GCN? I remember that GCN presentation where AMD claimed vector instructions of the same wavefront could be issued back to back on one SIMD opposed to the interleaving of wavefronts necessary for VLIW (which implies a 4 cycle latency for GCN). This of course does not apply to certain scheduling hazards, for instance when scalar and vector ALU access scalar regs directly after each other. But this has to be dealt with in the code.It can only issue instructions the same thread it started with after 8 cycles.
Durango's RAM bus is 68GB/s the ESRAM is 102GB/s
Think of it like a PC with 68 GB/s main memory and 102 GB/s VRAM.I thought it was 68GB's
I thought it was 32mb of edram are they the same thing?
I thought it was 32mb of edram are they the same thing?
Think of it like a PC with 68 GB/s main memory and 102 GB/s VRAM.
Be nice to see how those move engines are set up.You have 32MB of ESRAM to which the bandwidth is 102GBs and you have the 8GB DDR3 memory pool to which the bandwidth is 68GBs. The move engines suppossely coordinates the data movement between both pools of memory.
Doesn't really matter where the operations are performed as the BW consumed is the same whether writing FB to DDR3 or ESRAM. Okay, there are particulars about the setup which'll tell us how much BW overhead there is which is important for the overall picture, but the actual BW available to Durango's GPU is certainly a lot more than 68 GB/s!Well, according to ERP developers are encouraged to use DDR3 as the framebuffer...I am more of the theory that ESRAM is a big L3 cache for GPGPU ops.
Doesn't really matter where the operations are performed as the BW consumed is the same whether writing FB to DDR3 or ESRAM. Okay, there are particulars about the setup which'll tell us how much BW overhead there is which is important for the overall picture, but the actual BW available to Durango's GPU is certainly a lot more than 68 GB/s!
That's a pretty nice analogy.Think of it like a PC with 68 GB/s main memory and 102 GB/s VRAM.