Intel is fucked now.

File: Trip.gif (2MB, 300x169px) Image search: [Google]

2MB, 300x169px

>>61026947
Indeed. IFPT Processor factor is easily going to shadow Intel's current shitshow, even with submerge layer threading and triple-channel quarts linking.
Stop buying processors for the next 4 months, it's pointless.

>>61026947
Is EPIC just a combo of two KIKE RIPPERS?

>>61027025
We've come full circle a second time.
Original dual cores were two single-cores glued together.
Original quad cores were two dual-cores glued together.
Now we're here. 16-32 cores? That's just 2-4 octo-cores glued together.

But hey, it works.

File: intel ceo JUST.jpg (119KB, 1000x1000px) Image search: [Google]

119KB, 1000x1000px

>>61026947
QUAD DAMAGE.

>>61027041
considering that the memefabric offers near perfect scalability, i'd say it works pretty damn well

File: Wafer_die's_yield_model_(10-20-40mm)_-_Version_2_-_EN.png (163KB, 3000x1000px) Image search: [Google]

163KB, 3000x1000px

>>61027041
It should be cheaper to manufacture. Smaller chips = better yields
and memefabric works well as >>61027184 posted

>>61027184
>>61027285
quick rundown on memefabric?
is it ayylmao tech?

>>61027334
It's an improved Hypertransport which is something AMD has been using for awhile.

>>61027334
its basically a communication bus/interface between cores which allows for amd to use multiple dies to build a cpu instead of fitting everything on the same die thus increasing yields per wafer
aymed tech that works really well (see intels ring bus for example, it sucks btw)

>>61027346
THE CONTRACT KELLER

>>61027334
It's a data and control fabric which is said to be a bundled set of buses which fall under HyperTransport other engines (including GPU DMA); essentially it provides high bandwidth interconnect for all SoC components but AMD adds a link per CCX and it can scale with memory bandwidth for a maximum of 512GB/s. It is likely the scraps of AMD's HSA infrastructure and has been put to extremely good use. It vastly improves the scalability of interconnected processors and makes any NUMA aware software basically scale core for core. IF also controls SenseMI for a lot of power saving features e.g. XFR.

Ryzen is capable of scaling to more than 60GB/s with 3600MT/s RAM that's available OOTB, and sees inter-CCX latencies as low as 110ns (which is not great), but it is said that Threadripper/EPYC clocks IF faster than 1:1 memory clock. The only thing we need to worry about is Vega.