Rumor | Leaked slides reveal AMD's 2-year roadmap, 'Raven Ridge' delayed till 2018
Another roadmap has leaked and this time, its about AMD. The slides, which were obtained by Informatica Cera, reveal some interesting (and to an extent, obvious) facts about AMD's plans for the next year and after. Before we move onto the details, please do note that the authenticity of the slides has not been completely verified so do take this information with a pinch of salt.
A 'masterpiece' contender
First up, Matisse and Picasso — and no, we aren't talking about great artists here. These are supposedly codenames for AMD CPUs and APUs set to land in 2019. Matisse will be the desktop CPU and will likely be based on the Zen 2 architecture. AMD said that they have enough headroom for squeezing in more IPC and clock speeds. We could see more IPCs in Zen 2, which will help sustain the pressure on Intel. There are no revisions of the current Zen architecture planned for 2018. However, Pinnacle Ridge will succeed the current Summit Ridge lineup and offer a performance boost.
On the APU side, we see that Raven Ridge is pushed to 2018 (bummer!). We have reported earlier that Ryzen 5 'Raven Ridge' is offering up to 48% higher multi-core performance compared to the 'Bristol Ridge' A12-9800 APU. What we didn't know at that time was the number of stream processors (SPs) that the on-die Vega GPU would contain. We now get to know from the slide that the on-die Vega GPU in 'Raven Ridge' could have up to 11 compute units (CUs), which means up to 704 SPs (each CU comprises of 64 SPs). A possible error on the slide is the mention of 'Bristol Ridge' APUs featuring Polaris GPUs. This could mean two things — either the slide is not accurate or, Polaris (GCN 4th gen) and GCN 1.2 (Tonga/Fiji, 3rd gen) have been used interchangeably, given that there is not much difference between the 3rd and 4th generations of GCN (Graphics Core Next). 'Raven Ridge' would be continue well into 2019, refreshed as Picasso with a performance boost.
Vega 20 — Fastha and more powerful
We also get to see AMD's plans for the newly launched Vega GPU architecture. The current Vega 10 GPU is part of the Radeon Vega 64 and 56, and also the Vega Frontier Edition. While Vega 10 is based on a 14nm process, Vega 20 will be based on a 7nm process. Vega GPUs manufactured in 2018 will, however, feature Global Foundries' new 12nm LP process. AMD will also be releasing the next version of its EPYC server-centric CPU codenamed, Rome: both Vega 20 and EPYC-Rome will support the PCIe Gen4 bus. It is also interesting to note that AMD will be iterating upon its ROCm (Radeon Open Compute) platform in Q1 2018. ROCm is an open-source framework for Hyperscale-class GPU computing. Going by the slide, Vega 20 seems capable of taking on Higher Order Reasoning computing tasks made possible by ROCm 2.1.
Fight to the finish
Finally, we have a slide showing off Ryzen 5 Pro Mobile's performance in comparison with an Intel Kaby Lake Core i5. We aren't exactly sure which Core i5 model is being referred to here, but the Ryzen 5 APU seems to ace the Core i5 in multi-threaded and GPU intensive tests, while still running at comparable idle power. At least theoretically, it could probably be safe to assume that the GPU performance of the on-die Vega will easily trounce even Intel Iris Pro graphics but it would be difficult to ascertain overall performance unless we get our hands on a Ryzen 5 APU notebook.
Going by these slides, we might have to wait a bit more longer before we can put it through our tests. Although Ryzen helped AMD make a good comeback in the desktop market, it is kind of disappointing to see the 'Raven Ridge' pushed back even further especially when Intel is putting its might behind its 8th generation CPUs.