Charlie D over at SemiAccurate also writes that Infinity Fabric was spawned from SeaMicro's Freedom Fabric.
"AMD is also promoting its Infinity Fabric as more than just an intra-package interconnect solution but also as a way to communicate externally with other sockets and even with GPUs. Coincidentally AMD's RX Vega GPUs will also use the Infinity Fabric both internally and externally. Given the importance of the Infinity Fabric to AMD's server efforts it's beginning to seem like AMD's acquisition of SeaMicro may have borne more fruit than we first thought."
AMD's ARM core should be pretty much finalized. I get that it is much easier to compete with just Intel/Nvidia than with everybody else - good choice there - but there is a bit of a golden opportunity to release ARM parts on the same platform(socket) as the x86 parts in both server and desktop. ARM will start to penetrate the market and AMD might as well be a leader there. They had limited resources now but next year or on 7nm they really should invest in this.
In semi-custom folks need to see more wins and AMD has talked about an ARM based win. At the end of the day it's a business segment that was somewhat neglected at this event.
Vega was 2016, late 2016, maybe it ships a little in late 2016 and now it's mid 2017 ,if that. It's highly likely that there was no big Polaris last year because Vega was supposed to arrive. In my view, AMD needs to achieve near perfect scaling with multiple GPU die(chiplets) by the end of the decade, likely with Navi but curious where Vega is on this path so the rumored dual Vega will be the one to watch.
The big wins are not just Epyc , in PC there is demand for more cores in both desktop and laptop. Intel has been focused on quad cores in desktop and dual cores in laptop. AMD doubles that, in both desktop and mobile- Raven Ridge is likely mostly quad cores as AMD has the CCX complex so every die will have 4 cores and they might as well use them. Zen does well up to 1V and 3.2-3.5GHz, hopefully AMD gets the uncore power consumption to solid levels and Raven Ridge will be quite smashing. In gaming , Vega is bound to be better than Intel's integrated GPU and having 4 cores to spread the load helps further. Going bellow 12-15W might be difficult for Raven Ridge so maybe AMD won't address that segment with Zen this year. Today AMD has minimal revenue share in PC CPU/APU ,maybe 3-4% and the market has a few years left in it - all goes to hell once commercial starts to replace PCs with glasses- my bet is that by 2025 PC sales will be close to zero, assuming somewhat compelling AR/MR products arrive by 2020.
In auto there was that Tesla moment and AMD hinted at some activity in this segment. Mid term there is room to disrupt in auto as folks focus on autonomy but forget to focus on thr system. Cars are far too complex. If you re-architect the car and simplify it (no more a few km of wires), you reduce development time, manufacturing costs and even operating costs. Someone that can solve this problem and autonomy at the same time, would have a solid chance to be a leader in the segment.
Good point on the Treadripper price. It has twice the cores/threads of the Ryzen 7, which tops out at $499. So Threadripper is more like in the $800-900 price range.
AMD is out of the ARM server business for now - the company is only building x86 servers for now. Automotive is more opportunistic and not a big push. Semi-custom business was covered by the game consoles and it's not a business they can forecast.
In my opinion, the execution hasn't been bad. I think the Radeon team is trying to reposition Vega after NVIDIA dropped the 1080Ti, TitanXp, and Volta. Building graphics cards with HBM2 technology is a challenge.
The obvious big win for AMD will be Epyc. There's a lot of built-up demand for some real compeition to Intel's Xeon.
AMD's Forrest Norrod was asked that question in an informal discussion and he said Infinity Fabric does not use the SeaMicro Freedom Fabric. Instead Infinity Fabric is an outgrowth of coherent HyperTransport.
Threadripper can't cost over 1000$, there is no market for it and they need a sustainable ecosystem around the platform - motherboards, CPU coolers. They can have a SKU or two above 1000$ but AMD needs to have a 16 cores at 799$ so customers can fit CPU +mobo in 1000$. Will be interesting to see how they price 12 cores and less, if they have less than 12. No point in pricing it higher than server parts either - you can get a low clocks dual socket 8 cores Xeon E5-2620 v4 at 411$ today in retail.
Anyway this was the weirdest presentation ever. AMD and its maniacal focus on shooting itself in the face. Lots of positive news, especially with Naples but then they went ahead and provided an impossibly low financial outlook.
They made the claim that in PC Ryzen expands the SAM from 9 billion to 28 billion, that Vega in PC/Workstation expands it from 2.25 billion to 5 billion (2016 numbers) and that server is at 0% today so another big opportunity. All good until AMD decided to project double digit revenue growth. If you want to be cautious just don't provide any long term revenue outlook or folks will focus on the negative and fail to see the positives.
They forgot to mention ARM server, semi-custom, cars and address long term plans for machine learning.
AMD still has severe execution problems. Vega lost an entire cycle with it's delays, Zen wasn't quite ready on time. AMD's RX 560 hit retail this week under the radar despite being a substantial upgrade over the RX 460 - that's just really bad business.
As for Epyc, Zen has higher throughput than Intel's core ,except with 256-bit vector code and it's pretty efficient so no reason not to sell well.