Intel Stratix 10 FPGA main die measures 560 mm2 and contains 17 billion transistors.
EMIB is used to connect the main die to 6 transceiver dies.
link
Thank you for the info, also this article should put some rest on the Intel claimed density vs real products effective one: 17 B transistors / 560 mm2 =30M transistors/mm2, so actually twice what their core line reaches on the same 14 nm process... yeah density depends on product and design more so than the node used.
Of course this FGPA runs at 1 GHz rather than 4.5 GHz like core cpus so the HP process is less dense for a reason... yet it seems still low when Zen sits at 24 M transistors/mm2 and can reach up to 4 GHz, 3.5 more effortlessly.
Is the heat/power issue really so great that you have to lower density over 50% to reach 500 MHz-1 GHz above?
I think intel certainly needs to bring 6 and 8 cores to the mainstream. The issue with intel IMO is they simply don't know how to charge a reasonable price anymore. I'm sure Skylake-x will be great, but if it still costs $1000 for an 8 core chip, no one will buy the thing except for a few enthusiasts. 95% of people wanting more than four cores will spend their money elsewhere.
8 cores I don't know but 6 should have come a year sooner in my opinion. Intel didn't grant us that in 2016, with all the time after Skylake released in 2015 for designing a six core mainstream rather than only updating the process with Kabylake, so here they are in this messy situation against 6-8 cores Zen that are very competitive.
If Kabylake would have upped the game a with 4.2 GHz+turbo six cores at the same price it would have been really hard to decide: six faster cores (clocks+IPC) or eight for a little more money (1700X) or slightly less (1700)?
Coffelake will bring exactly this kind of questions but a year too late, let's see if 14nm++ makes it worth all the wait. Also who knows if Skylake-x six cores turns out to be better? It should be on 14nm+ btw...
IMHO it's starting to get all too mixed up with two core architectures on two different platforms. I'm probably waiting a bit still before upgrading to either Zen2 or Icelake when they're both out and compared in 2018, by that time we should have a clearer idea on what's better overall for a PC: more cores? More cache? Best single thread performance?
So does that mean we can expect desktop Icelake in Sept. 2018?
I'm expecting that to happen, just like Skylake released in time after the 14nm initial mess. On that note, 10nm+ shouldn't be much worse than 14nm++ and still be better than 14nm+ according to the slides, so clocks shouldn't fall too much from Coffelake. The process might allow for speeds as high as Kabylake can reach today... do we know if Icelake brings any core count increase then?
At ~half the size from Coffelake (149 mm2) the die might be just 80 mm2 or so for an enhanced six core, that's way to little. Triple the L3 cache like in the leaked Geekbench entry and it's still below 90 mm2... I'm not seeing mainstream dies that small till 7-5nm nodes.