AMD Vega: The new generation of graphics that bet on the power thanks to HBM2

AMD sought efficiency with Polaris and graphics as the remarkable Radeon RX 480, but the new generation of graphics cards of this company wants to put the difficult things to Nvidia in the field where it traditionally dominated: performance.

To achieve this comes AMD Vega, a new architecture that is focused on power and not so much on efficiency, and that benefits from the adoption of HBM2 (High Bandwidth Memory) its new standard for graphics. This could be AMD’s most important architectural leap in years, and we already have the first details of this release.

AMD Vega
Image Source: Google Image

New approach, now it’s high-end

One of the secrets to the good long-term behavior of the Polaris architecture was the use of the so-called GCN (Graphics Core Next), a series of microarchitectures that favors parallelism and will once again be a protagonist in AMD Vega. Polaris, as indicated in AnandTech, was an optimized version of the GCN 1.2 architecture (“GCN 3”) for AMD’s 14 nm FinFET manufacturing processes.

These graphs of the Polaris family performed well in the field of energy efficiency, but were not oriented to conquer the high range. Here AMD’s message was perhaps too optimistic and created false expectations, but the disappointment caused by Polaris – if it really existed – should be completely mitigated with AMD Vega.

You may also like to read another article on iMindSoft: Nvidia GeForce GTX 10 for laptops: The same power for games and virtual reality

Welcome to the new Next Compute Unit (NCU)

The new generation of computing units (Vega NCU) is another of the pillars of this new architecture in which up to 512 operations of 8 bits per clock cycle can be executed (either 256 operations of 16 bits, or 128 operations of 32 Bits per clock cycle), which makes it possible to optimize these operations for higher clock frequencies and to obtain higher IPCs (Instructions Per Clock).

The new computing unit is able to manage a pair of FP16 operations on a single FP32 ALU, something that can have a notable impact on the execution of less precise operations, which can be processed in half cycles if they comply with The necessary conditions (basically, that they are almost identical).

The impact of this type of improvement will not always be applicable, and if you are thinking about games you may find yourself disappointed, because these less precise operations or work with integers – another type of operation that gains ground in Vega – has an impact Direct in the capacity of the GPU in segments like the one of the neural networks and its algorithms of inference.

With the new NCUs it is possible to optimize that work, basically making it possible to obtain more performance in a smaller number of cycles, something that explained a long time in WCCFTech and that is another of the fundamental factors in that new landmark of the performance that puts to AMD In a new dimension in the segment of graphics cards. Gamers should be happy, because this leap from AMD encourages more than ever a sector that is still very relevant.

AMD Vega
Image Source: Google Image

HBM2, Vega Performance Key

AMD promises that with Vega we will have the most scalable memory architecture in the world in terms of graphics processors, but also the presence of a cache with an even wider bandwidth thanks, among other things, to HBM2, the technology it offers Twice as wide as the original version of this technology.

The capacity of these chips also increases, and while in HBM1 there was a limit of 1 GB per chip (Fiji could only have 4GB of VRAM) in Vega we find numbers that can be much higher, and although it is probable that in principle we see Cards with 8 GB of HBM2 memory, could reach 32 GB in certain applications, for example for models for intensive computing and artificial intelligence.

The introduction of a new, much more scalable memory architecture and components as a cache controller allows large amounts of data to be managed to be apparently much more efficient, something that is important again in intensive workloads but perhaps also in games. Here AMD showed data regarding ‘The Witcher III’ and ‘Fallout 4’, where it reserves much more memory than is actually used (almost double), suggesting that there will also be tangible benefits in games.

When will the first graphics cards based on AMD Vega arrive?

This is the first time that AMD offers detailed information about its new graphic architecture, and although as they say in AnandTech there is still “a lot of fabric to cut”, it is clear that the changes are really remarkable and the advantages should be too.

Obviously we can not do that until we have the new AMD graphics with us. The firm has not given concrete dates for the launch of these products, but AMD has confirmed the existence of two Vega chips that will be responsible for showing us what this new generation is capable of.

The tests and leaks that have appeared appear to put Vega ahead of the GTX 1080, and it talks about gross yields of 12 or 12.5 TFLOPs, clearly above the 9 TFLOPs of the graphs of NVIDIA although, With consumptions also superior. AMD is expected to present the first graphs of the Vega family during this first half of the year, so we’ll be watching.

You May also like: Guide to Give Your Company the Proper Branding.

Leave a Comment