Today we have seen a lot of action surrounding the soon to be released GTX 1660 Ti, which at one point many considered a fantasy created by strange minds and not an upcoming product at all. Doubt has been removed with the leak of details and pictures of packaging, spotted by WCCFTech and others.
Thanks to the packaging we know the card will have 6 GB GDDR6 VRAM, DirectX 12 support, ANSEL support and Turing Shaders, though no mention of Ray Tracing appears. The back of the card features DVI-D, HDMI, Display Port and the Virtual Link connector which was missing from some custom RTX series cards. Check out the link for more models from third party vendors.
"Featuring the same Turing GPU architecture, the new GeForce GTX graphics cards will exclude Ray Tracing but feature faster shading performance through the enhanced GPU design while utilizing the 12nm process node."
Here is some more Tech News from around the web:
- AMD's 7nm Navi GPUs reportedly delayed until October @ The Inquirer
- Apple supply chain: TSMC to remain sole iPhone chip supplier @ DigiTimes
- Apple sued because two-factor authentication is inconvenient @ The Inquirer
- LibreOffice 6.2 is here: Running up a Tab at the NotebookBar? You can turn it all off if you want @ The Register
- New Part Day: Mapping With RealSense Cameras For $200 @ Hackaday
- Leaky child-tracking smartwatch maker hits back at bad PR @ The Register
- Amazon launches its own cheesy teleshopping channel @ The Inquirer
- 10Gtek X550-T1 10G Ethernet Converged Network Adapter Review @ NikKTech
So it’s 12nm but I’m not yet
So it’s 12nm but I’m not yet convinced that the shader cores are Turing just on that process node alone. And what is the ROP count on this SKU.
Now on to your Tech Talk headlines:
“• AMD’s 7nm Navi GPUs reportedly delayed until October @ The Inquirer”
The release grape vine states 2H of 2019 for Navi and that could be anytime between July 1 to Dec 31 2019. And one would expect that AMD would at least want a date before Oct 2019 to be ready for back to school as well as for the Thanksgiving through XMAS holiday shopping season.
When is the Next Big Trade/Industry event where AMD may just be providing more information. And at least AMD has its Radeon VII out there and hopefully there can be a lower binned Vega 20 based variant than the Radeon VII with at least 52-56 working nCUs that’s more similar to the Vega 56 price segement.
My take on AMD’s use of 4, 4GB HBM2 stacks on Radeon VII is that no one in the HBM2 market was using any 2GB capacity HBM2 stacks and the Server/HPC/AI Pro GPU accelerator market was wanting 8GB or higher Per HBM2 stack capacities anyways and will soon be moving to the higher than 8GB HBM2 capacity that’s allowed for in that new revised JEDEC HBM2 standard.
I’ll bet that any 2GB HBM2 supplies would have had to been a custom order from AMD and that the cost per 2GB HBM2 stack would not have been that much different from the volume pricing on the 4GB HBM2 stacks.
Come on AMD whip up a Vega 20 based Radeon Pro DUO for some dual Vega 20 DIEs(Each with 16GB HBM2) on a single PCIe card action! And wire up them 2 GPU puppies up Via xGMI(Infinity fabric) links across the PCIe card’s PCB instead of using PCIe Links for the inter-GPU DIE connection Fabric.
I’m sure that there will-be/have-been some Vega 20 dies coming off of TMSC’s 7nm diffusion lines with even less than 56 down to 50 and below working nCUs. So even a Dual 50 nCU Vega 20 binned DIE SKU with 100 nCUs and 100-128 working ROPs will have some damn good Pixel fill rates even down-clocked a bit relative to a single GPU die based GPU Card. That would be one beast even with the HBM2 clocked a bit lower across 2 GPU DIEs at 8 total HBM2 stacks across 2 GPU DIEs for a total of 32GB of available HBM2 based VRAM, even at 800-900GB/s HBM2 bandwidth per GPU DIE. AMD could probably ask about $1200+ for the part and make plenty of Prosumer Sales.