Amd navi: everything we know so far and what we expect

To put ourselves in a position on what AMD Navi is, we are going to do a quick retrospective of what happened at AMD since the Zen Project began.

The goal was to break with all of the above when it comes to how to manufacture and evolve chips and architectures over the years in search of one last saving hand that would allow them not only to save themselves from demise, but to come back and compete.

The need to not be able to individually develop the CPU and GPU divisions came together with a project that intertwined them and needed them to be developed under a common framework in which they would interconnect.

If Intel was a CPU company and Nvidia one of GPUs, AMD had the option of being a parent company that housed CPU and GPU divisions. This is what he did after acquiring ATI with a dire result.

But it also had the possibility of being a unique company that designed solutions combining both components, not simply integrating them in isolation as 'parts'. In this he is currently with the Zen project.

It would be unfair not to note that resources were allocated unequally for both 'legs'. AMD focused on the Zen Core (CPU) almost all its capacity, leaving the RTG (GPU) relegated to a second row with a much lower budget and number of engineers, not being the stated objectives that were wanted to be achieved.

It is also necessary to mention Sony's 'support' in the development of the project based on the company's 'request' for a semicustom solution to become the heart of its next game console (PS5).

And this order must remain in the reading buffer for the entire time that we read the article because Navi must stick to it, as it has been and we will check later.

Let's say Sony released the message:

Changes in the GPU Division Roadmap between 2017 and 2018

Within the Zen project, the CPU and GPU chips coexist in order to interconnect through the Infinity Fabric.

In the CPU division, everything seems to be going well in terms of reaching the milestones set out in its Roadmap, as well as the advances in (outsourced) manufacturing technology. The same does not happen in the graphics division (RTG) with Raja Koduri in front.

Raja Koduri wanted to prioritize 'her' RX Vega, but her boss assigned 2/3 RTG to NAVI.

2018: Incorporation David Wang and

We will not enter who moves the tab first, but the facts are that Raja goes to Intel to be master and master of a whole new division of GPU that he will have to create and develop there, and Lisa Su puts RTG engineering at D. Wang who in the past had worked at ATI, ArtX, SGI, Axil Workstations, LSI Logic as well as AMD itself.

Incidentally, there is also the incorporation of M. Rayfield who will define the strategy and business model and who had previously worked at Micron and Nvidia, highlighting his achievements in efficiency and the consolidation of agreements with manufacturers.

Just 1 year later Rayfield left AMD, leaving everything concerning RTG in the hands of Wang… and Su, which we remember, is above.

And there are changes to the Raja Roadmap.

On the Investor Day of the year 2017 the AMD Roadmap pointed to VEGA solutions of 14nm (2017) and 14nm + (2018).

Here were RX Vega 14+ provided by R. Koduri

What finally materialized were: The Raven Ridge APUs with Vega Graphics, Apple's Imac and Intel's Kaby Lake G.

Due in part to the memory prices of the moment, to which HBM2 was even more sensitive, VEGA was launched on a very limited basis, fortunately, as it offered a price / performance / efficiency ratio far removed from what AMD had projected.

Radeon Vega Frontier was later accompanied by RX Vega 56 and 64 (2017).

In addition, and to the general disappointment of the public, they launched the Polaris 20 refresh that had never appeared in the Roadmap and that consisted of a slight frequency optimization and cost reduction.

In mid-2018 there is no trace of 14nm + Vega products, so the only thing that seems to make sense is waiting for a new tweak in the form of Polaris 30 and assuming that the Koduri Roadmap was internally canceled without announcing anything publicly.

Here there are no more 14+ (nor R. Koduri), and the new Vega are delayed until 7nm.

This is reinforced by AMD's announcement that Vega 7nm production for the Data Center segment will begin in 2019, without specifying any launches for the gaming segment at that time.

This is very consistent with the new guidelines that appear in David Wang's presentation in the Next Horiz (7) on which emphasizes the priority of ' Continuous performance per watt gains', which he shares with the Zen core for cpu and redirecting the development and evolution of the RTG GPU chip towards the Data Center to later adapt the product and create the rest of the solutions based on it.

Any questions? A photo is worth a thousand words.

Looking again at the original Roadmap, in 2019 there should have been Navi-based products jumping to the 7nm node… if the predictions were met.

However, we have witnessed the addition of 7nm Vega, or what is the same: a delay of Navi at least temporarily, waiting for what happens next week at E3, location to which David Wang directed any relative question with Navi in ​​the past Computex.

If Navi is still not ready, even if it is announced (something else is when it launches), AMD would be forced to remain present in the market with revisions of Polaris (Polaris 30), manufactured in Globalfoundries on the 12nm node) and Vega 7 Otherwise, Polaris, Vega and Navi will coexist.

Taking into account what was seen a couple of days ago at Apple's WWDC 2019, specifically inside the new Mac Pro and its graphic hardware, it seems that they continue to develop Vega solutions including AMD Infinity Fabric Link GPU interconnect technology, which is the equivalent to Nvidia's NVLink.

Radeon PRO Vega 2 DUO and Infinity Fabric Link GPU interconnect.

Computex 2019 - NAVI and RDNA

Let's look at the little official and current information about Navi (after all the soap opera) that AMD offered at Computex because everything will finally take shape.

Radeon shall cover the following within AMD's portfolio:

• • Graphics & GamingGPU architecture for the future of GamingPC - Console - Cloud

The new generation of gaming will be covered by the RDNA architecture and the Navi family of graphics solutions.

The first point, Graphics & Gaming, specifies that we are in the gaming variant of the graphics solution, in case anyone needs to remember that we are in the AMD of Project Zen. Led by David Wang and Mike Rayfield.

Next Horizon reminder: Navi, yes, but within the strategy, not 'to your ball'. Reference to the previous year with Vega 7nm, first for datacenter and then for gaming with Radeon VII.

Next Horizon reminder: Data Center architecture that will later be adapted.

Here the 'reminder' moment at Computex.

The second point, GPU architecture for the future of Gaming, declares the need to build an architecture that allows the coexistence of different devices that can play together. Console and cloud pc.

Next Horizon Reminder: Develop GPU for Data Center and cover PC / Console / Cloud.

On the third point, PC - Console - Cloud, the truth is that in these months we have attended announcements that are quite related to this (apus for Xbox and PlayStation, Google Stadia, Samsung opting for Radeon), perhaps, ironically, from what we least know specifically is about the PC solutions.

NAVI is a new FAMILY of Radeon GPUs.

It has been built to drive the future (which begins now) of video games and it is a principle on which the coming generations of this new era will rest.

In the presentation there is a direct and specific mention (required?) To PS5, powered by Radeon Navi (and zen 2 yes). The king what is the king.

The technical characteristics that underpin Navi are:

• • Manufactured in the 7nm node of TSMC. New RDNA architecture (does not imply not being able to coexist with GCN). Lower consumption, higher frequency and performance. PCI 4.0 enabled.

Next Horizon reminder. PCIe 4.0 does not reach RX Radeon 'per se'.

AMD is very fond of its previous and long-standing GCN architecture and urges that it be considered to continue with future versions of Vega in other versions of workload and computing.

RDNA is the equivalent of 'zen' on cpu.

It is to redo architecture from scratch looking at the needs of the future that are currently considered. Pay attention to the following (especially point 1).

I transcribe what is read in the slide: 2 billion gamers demand

1. 1. Social Interaction ControlPerformanceResolutionStreamingEfficiency

And this is what there is. This is what AMD marks as premises and benchmarks in each step it will take with Navi and RDNA. Surely many will not like anything at all, but do not ask the elm for pears.

On the one hand we have the David Wang slides in which he details the objectives of evolution and development of GPU chips, firstly in the Data Center environment and which will later have to be useful in Gaming.

On the other we have the confirmation that Radeon focuses as needs to be covered in the next decade something that is not exclusively PC Gaming, but an ecosystem with devices with disparate gross forces in which thanks to it, the user must be able to participate.

RDNA will be the base on which the products of the next decade will be developed. These are its technical characteristics:

• • Completely new design of the Compute Unit Better efficiency Increase IPCC Multilevel hierarchical cache Reduce latency Greater bandwidth Lower consumption.

Another important characteristic of RDNA and Navi is: Streamlined graphics pipeline thanks to this allows it to be optimized for performance per cycle and high frequencies.

Performance improvements and consumption reduction RDNA "Navi" vs. VEGA

• • 25% performance per clock50% performance per Watt

As we see, the cadence of the previous jump is maintained, since between VEGA and Polaris the performance improvements and the reduction in consumption were the same.

RDNA is the new architecture. Navi will be the new products.

The nomenclature for Radeon RX PC solutions will jump to 5xxx (50 is for AMD's 50th anniversary in 2019).

They will be available in July (7/7 even on Sunday?) With a location for the world presentation on June 10 at E3.

The package of the Navi graphics is smaller than the GCN-based Polaris / Vega graphics, something that does not attract attention either because some are manufactured on 7nm and the last ones on 14nm.

Navi was Lisa Su's priority. Navi is not Radeon RX to the maximum and enfica only to the PC.

AMD announces the use of GDDR6 memory and PCIe 4.0 support on the RX 5700.

Well this is all there is. It doesn't seem like Nvidia's throne at the apex of PC gaming enthusiast is under threat.

In any case, what was announced by AMD seems to confirm a clear commitment to cover a broad coexistence market and progress through each generation, being more efficient and increasing performance, which would bring it closer to Nvidia step by step.

At the Computex presentation, a demo of a Navi GPU in which it competed against an RTX 2070 could be seen, something that without knowing the prices and segmentation says nothing.

On the other hand, Nvidia has threatened SUPER on Computex, which appear to be optimized versions (Ti?) Of their Geforce RTX, or at least of the versions that they think may be 'threatened' by the launch of Navi.

We recommend reading the best graphics cards on the market

Google is rumored to be moving tab with Stadia (which uses Radeon solutions), so perhaps Nvidia has to decide whether to introduce them a little before E3 or wait until next week.

Next week we will have definitive information on the models, technical specifications and prices.