AMD revealed plans to release an APU for servers during its Financial Analyst Day presentation back in May, and we now have a better picture of what that processor might look like. AMD has released a journal article about the product called "Achieving Exascale Capabilities through Heterogeneous Computing" in IEEE Micro, and according to Italian site Bits and Chips, this new chip is called the "Exascale Heterogenous Processor," or EHP for short.
Bits and Chips reproduced a fascinating high-level overview diagram of the EHP, as well. Looking at this diagram, one could be forgiven for thinking that someone mislaid a diagram of the Fiji GPU at first glance–there are a lot of similarities. The EHP die combines 16 Zen CPU cores (which Bits and Chips says can process up to 32 simultaneous threads) with an AMD "Arctic Islands" GPU core. The die sits on an interposer that connects the chip to some amount of HBM2 RAM.
In the article abstract, AMD says the EHP is meant to address the need for high-performance, power-efficient computing components without requiring integrators to design and purchase custom components. The company argues that its GPU technologies represent a natural replacement for these custom parts in highly parallel computing applications, and it also believes that an APU-like chip like the EHP is necessary to reach the full potential of GPU computing in these applications. It's not a strech to say such a chip could form the basis of a powerful consumer PC, as well.
|Nanoxia Project S case slides into home-theater setups||12|
|Nvidia previews Xavier SoC with Volta GPU for self-driving cars||13|
|be quiet! Silent Loop AIO liquid coolers hum along quietly||2|
|Microsoft catapults datacenter performance with FPGAs||38|
|Asus J3455M-E mobo sails out with Apollo Lake SoC aboard||18|
|AOC's Agon family of gaming monitors heads stateside||16|
|Google Data Saver improves mobile browsing on narrow pipes||10|
|Toshiba expands its budget SSD lineup with its OCZ TL100||13|
|Rumor: Nvidia and Apple may reunite for future Mac GPUs||29|