So by looking up GDDR5 and HBM bandwidth then this "The product operates with an I/O data rate of 16Gbps(Gigabits per second) per pin, which is the industry’s fastest. With a forthcoming high-end graphics card of 384-bit I/Os, this DRAM processes up to 768GB(Gigabytes) of graphics data per second." means something like 96 GB/s per module? (8 GB so 8 * 96 = 768) HBM1 (to compare.) being 128 GB/s but the Fury being classed as 512 GB/s due to 4 GB each using 128 so 128 * 4 = 512, I guess? https://en.wikipedia.org/wiki/GDDR5_SDRAM https://en.wikipedia.org/wiki/GDDR6_SDRAM https://en.wikipedia.org/wiki/High_Bandwidth_Memory Eh what do I know, faster and better than GDDR5 in any case. 😀 (And probably cheaper than the upcoming HBM3 or what revision that was at now.) EDIT: What's the TitanX(p) clocked at again, 12 GB/s with GDDR5X no? A nice little increase then for GDDR6 and then tweaked further by increasing the clock speed I suppose. But what do I know, just thinking out loud trying to figure out the actual memory speed rather than the marketing text above using the product they tested and it's memory speed total. :P

Enjoy the ride until 2025.

So Navi will come with GDDR6 i assume

Awesome! Yah I kinda figured GDDR6 would be the next step for most GPUs. HBM2 is nice but it's damn expensive especially if you put more than 8GB on a card. With enthusiast class cards hitting 12GB it's very likely the next jump will be up to 14GB or 16GB. With those larger memory sizes we'd need to see a jump in bus width too most likely. Probably 448-bit or 512-bit memory bus would be ideal for a card with that much memory. We haven't seen memory bus width like that since the GTX280 cards. lol 448-bit * 16000Mhz / 8 = 896GB/sec 512-bit * 16000Mhz / 8 = 1024GB/sec