It’s been a pretty interesting couple of weeks in the GPU industry (okay, that’s a subtle understatement), with a whole slew of news from both Nvidia and AMD on upcoming graphics cards and architectures. But Vega news is particularly prevalent at the moment, and another source has popped up citing that we’ll not only see the high end Vega 10 graphics card debut before the years end, but also Vega 11 in the early new year.
Firstly, let’s discuss Vega 10. The latest rumors indeed seem to tally with older information that we’ll be seeing Vega 10 using a later and improve version of the GCN architecture, which will see several hardware changes to improve the performance and efficiency of the graphics card. Supposedly, this 9th generation AMD Radeon graphics architecture will sport significantly changed SIMD units, with each SIMD now capable of processing variable length wavefronts.
Vega 10 also adds the ability for the GCN architecture to ensure peak stream processor occupancy and reduce access times to both cache and the main memory (which in the case of Vega 10, is HBM2). If you’re unsure what this means – it basically means that the thousands of processors inside the GPU (the SIMD units) will be better kept at higher levels of ‘occupancy’ – in other words, they’ll be better used rather than possibly having certain units idle during complex scenes due to poor load balancing or latency as data is being shuffled between the caches or GPU’s memory.
Polaris is the ‘4th’ generation GCN. We can Presume Vega therefore will be Generation 5.
Supposedly there’ll also be even more changes in the memory compression of the GPU. Color compression helps losslessly ‘compress’ data as it’s sent around the graphics card, and by doing so reduces the size in both memory and means it can be sent faster.
Supposedly, Vega 10 will come in a multitude of different configurations – some for the compute / professional market (think servers or deep learning) and of course a variety of products for gaming. Naturally the highest end GPU will feature 4096 shaders and 16GB of HBM2 RAM. It will (if these leaks are accurate) mean that Vega 10 has a max TDP of 230 W, and this means a 6 and 8 pin PCIe power connector are sufficient.
So what about Vega 11? Well, we’ve touched on Vega 11 before – and these new leaks seem to clarify that it takes the Polaris architecture and puts it on steroids. Essentially the GPU’s will run faster than Polaris 10 cards while sucking up less energy, and does this with a mixture of an updated graphics architecture and higher clock speeds. As you might be aware, the bog standard RX 480 runs at about 1266 Mhz, and so bumping this speed up to say just 1400Mhz with a tweaked architecture could breathe life into AMD’s the mid range cards. 
The only thing which is a bit weird about these leaks is that Vega 11 will also feature HBM2, but at only half the number of stacks and therefore just 8GB. Interestingly, this is primarily for notebooks and to replace Polaris 10 on the desktop.
Other sources report that we’ll be seeing a dual Vega 10 card released for the professional market, and that AMD’s team is currently hard at work on the graphics card.
As usual, take these leaks and rumors with a grain of salt until we see the actual hardware be formally announced.
