From AMD stating that the Naples 2p platform only has 64 PCI-E lanes per proc not 128 like the 1p, so each zepplin SOC is using 16 lanes of phy , my assumption is that xGMI will be PCI-E 3.x based in terms of encoding etc ( AMD have said they can do infinty fabric over PCI-E as well) because of distance, i guess it could be 802.3 based but then there wouldn't be any inter proc issues like neilz hinted at.
In-game texture sampling looks awful.
But its 64 lanes between each Naples socket not 16.
Yes and each Zepplin SOC has a possible 32 phy lanes so 128 in total. That is one of the big interesting point for Zen. On 1P very large amounts of all flash NVME attached storage is very nice, combinations of GPU's and storage is very nice etc. But you go to 2P and you loose 16x4x2 or 128 lanes out of a possible 256 lanes.
You have to remember its more then just CPU->memory. that bandwdith is for all PCI-E attached devices, the ethernet controllers, encryption units, CPU and memory. Intels omni path is ~100Gb/s.
= 12.5GB/s vs 15.7GB/s for 16* PCIE3 so...
yes but omni path goes significantly further and thats for just 1 link you can have more then one. I dont know what QPI does these days.
http://www.tomshardware.com/news/amd-ryzen-chipset-drivers-windows-10-x64,34254.html
id is closer to AMD at the moment because AMD and Bethesda have a large marketing agreement.
AFAIK Kaby-lake + Z270 chipset provide 40 gen3 pci-e lanes while Ryzen + X370 chipset provide 20 lanes, which are insufficient for a x16 videocard + two x4 M.2 drives like the Samsung 960 Pro. And things get worse with APUs compared to Ryzen CPUs.
I'm not sure consoles were the reason as much as a need for lower CPU overhead. Busy areas in Fallout 4 were rough and modding doesn't help the situation. DX12 should still be an option, but Vulkan would have wider support.
That's still sufficient for a mainstream platform. The video card probably isn't using all that bandwidth and that would still rely on maxing the M.2 drives and GPU simultaneously. APUs may be interesting, so I wouldn't call them worse yet. There's the potential they consume zero lanes or are attached to chips with extra lanes lacking the ability to get off package. A situation could exist where an APU on AM4 has a mid range GPU and all the lanes. For more lanes there should be larger sockets.
Id-software has always been lobbying OpenGL, and Vulkan is the successor. AMD has traditionally had pretty bad OpenGL drivers, but Vulkan changes this completely. Vulkan design is based on Mantle, so it suits AMDs hardware very well. Consoles obviously are important in bringing AAA developers closer to AMD. Vulkan is the platform that allows Id-software to bring AMD console features to PC, such as async compute and cross lane ops. This collaboration benefits both parties.
