The next generation of Radeon: computational learning without restraint

Vega architecture moves to 7 nm with And highly impressive capabilities - for servers and professionals in the first stage

Cores 2 may have stolen most of the limelight because of its potential and significance as a basis for the long future, but one can not ignore the architecture For graphics processors, which will gain new life thanks to the transition to a highly advanced 7 nanometer manufacturing process - for the world of computational learning first, and perhaps even exclusively.


AMD officially introduced its Radeon Instinct MI60 and Radeon Instinct MI50 accelerators to compete with leading Tesla products Thanks to additional material elements that give them the ability to operate at the level Maximum in double-floating point calculations of 64 bit (ie half the full processing power of standard 32 bit floating calculations, compared with one quarter of that in previous generation accelerators), and thanks to the support of 8 bit and 4 bit integers that are very effective for the base Deep.

A wide range of computational configurations in one unit - to support all areas of the HPC organizational and scientific world

The new VEGA cores typically contain the same 64 performance clusters with 4,096 units in total, or over 13 billion (!) Transistors with a total area of ​​only 331 square meters for unprecedented density, with a promise of possible improvement of up to 25 In performance with the same power consumption compared to the Vega 10 core in the production process of , Apparently referring to the ability to increase the practical operation frequencies by approximately 25 under the same power consumption and heat emission.

Get updates from everyone in TelgramGet updates from us all at TelgramJoin the channel now


The 20 VEGA 7 cores will be attached to the Radeon Instinct models to four HBM2 Plus chips with a total volume of up to 32GB, which will deliver a gigantic effective bandwidth of 1,000 / gigabyte per second - combined with the PCI-Express connector 4.0 dual-speed system that will connect the accelerators to the main processing units and a set of dedicated Infinity Fabric connections with a bandwidth of one 100 / Gb / s that will allow up to four cards to communicate directly and work together, .

Impressive technologies that we would be very happy to receive in other AMD processing products

Under a power envelope of 300W, for both the MI60 and the MI50, which will contain a slightly reduced amount of cores and half of the HBM2 memory with an array Passive and relatively compact, the new accelerators are designed to deliver Maximum 15TFLOPS, or 15 trillion floating-point calculations every second, plus 7.4TFLOPS in double precision floating point calculations, almost 30TFLOPS in floating point calculations in half of the standard precision - and up to 108 trillion calculations per second in the basic INT4 configuration. Quite amazing numbers for the same processing core that has been accompanying us for over a year actually.


A leap over Vega 10 and tough competition against the Volta NVIDIA's latest updates

Despite the fact that this is a relatively large chip, for advanced lithography in the early stages of maturity, the red chip developer promises that the first customers will receive the Radeon Instinct MI60 before the end of the current quarter, ie before the end of the 2018 - Which can help to bite out any market share from prima facie.

The improvement in performance is relatively moderate to the improvement of density in the new lithography - but it is certainly much more predictable than the pre-developed core for the new age

Would you like to see the renewed Vega in the 7 nm coming to the gaming market, despite the fact that, Have not provided any statement or clue in the meantime? Or is it better for a developer to concentrate efforts on polishing the cores? Of the new generation for home users (even if it means another six months or more without new announcements)? Tell us in comments.

Should we also have a chance in the field of domestic games and processing?