AMD Roadmap Leak: Major Platform, Graphics Changes Coming in Zen 4


This web site may perhaps receive affiliate commissions from the backlinks on this site. Phrases of use.


A freshly leaked roadmap statements to show the long term of AMD CPUs, with Zen four highlighted as a significant inflection issue for the organization. As is always the case, viewers must retain in brain that we’re speaking about rumors, not actuality.

It is been quite a though since I did an AMD roadmap investigation, so we require to chew by means of a bit of what is coming down the pipe in 2021-2022 before we talk about Zen 4. The top rated row of codenames refers to AMD’s desktop CPU lineup. The 2nd row is AMD’s desktop and cellular APU chips. The third column — Vermeer, Cezanne, Lucienne, Van Gogh, and Chagall — are the CPUs AMD is top with in 2021.

Lucienne (3rd row, third column) is a refresh of 2020’s Renoir mobile system, but with slightly superior clocks and a lot more threads. Van Gogh is a bit of a puzzle. It looks like it derives from the operate AMD has completed on the Xbox Collection X and PlayStation 5, with a Zen 2 CPU combined with an RDNA2 GPU, but it reportedly targets a 5-18W energy envelope.

There is proof from AMD that Van Gogh exists, but it occupies an odd place in AMD’s product lineup relative to Cezanne and Lucienne. This slide references LPDDR5, but some new leaked proof indicates the chip could support quad-channel DDR5. Concerning the two, LPDDR5 seems considerably more likely. DDR5 is not even in-marketplace but, and a conventional desktop quad-channel memory interface on a laptop chip would burn a ton of ability. At last, there is Chagall, AMD’s Threadripper platform.

A single critical observe: This document aligns Vermeer, which released in 2020, with Cezanne, which introduced in 2021. This document is a products family roadmap, not a temporal roadmap, which is to say: It reveals which products and solutions will be involved in a given generation or Ryzen loved ones, not accurately when those merchandise will start.

Our assumption is that Warhol will debut on desktop in 2021, but that Rembrandt, Barcelo, and Dragon Crest will all be 2022 merchandise. The 6nm claim is a minimal surprising and may well not be precise. According to TSMC, the profit of 6nm over 7nm is that it presents up to 18 per cent enhanced transistor density, with no communicated benefits in functionality or ability. AMD and Intel never normally try to improve for density, due to the damaging impact it can have on clock velocity.

Rembrandt provides RDNA2 to cell and desktop APUs in 2022, but the Barcelo refresh guarantees Vega survives into another 12 months. This slide implies that AMD will launch AM5 and DDR5 help on cellular very first, and that Zen 3 will straddle both equally sockets, with the desktop taste limited to DDR4 and the mobile version supplying DDR5. AMD may possibly be jumping for DDR5 and LPDDR5 aid to get edge of the electricity discounts the two standards provide around regular DDR4.

If these predictions are correct, it indicates Rembrandt will be AMD’s most state-of-the-art system for a lot of 2022. We suspect AMD would skip releasing desktop APUs centered on Rembrandt, presented that the cellular chips are supposedly employing a different CPU socket. It appears to be as even though the x86 producer has some thing else in thoughts.

What Zen four, Ryzen 7000 May possibly Provide

Zen four presumably launches for desktop in late 2022 or early 2023. It’ll be AMD’s first desktop platform on 5nm, its 1st chip to attribute PCIe five., and it’ll be the initial time AMD has supplied a graphics answer on each CPU.

AMD is the organization that first popularized the concept of a CPU and GPU sharing the same piece of silicon, all the way back again to its acquisition of ATI back again in 2006. It’s ironic, hence, that Intel has done a better work of producing baseline graphics ability accessible across its complete product or service line. Starting with Zen four, this modifications, and RDNA2 turns into offered across the complete product stack. That is a even bigger transform than it could possibly look.

AMD may possibly have adopted chiplets for its desktop CPUs, but its APUs are firmly monolithic, together with the 8-core Cezanne/Ryzen 5000 APUs available in mobile. There is very good cause for this. Inspite of the advertising and marketing all around them, chiplets are not a unilateral favourable. AMD pays a penalty in terms of die place, latency, and power intake when compared to a monolithic chip. In the desktop room, these negatives are tiny — especially thinking about that AMD can discipline a 12-main and a 16-main desktop CPU at a lower for every-chip value than it would normally pay back. In laptops, however, AMD has picked to stick with a more common style. AMD’s consoles, likewise, are monolithic architectures.

Integrating a GPU into a chiplet CPU layout presents certain challenges. AMD could theoretically make a monolithic GPU chiplet to sit along with the I/O die, but the fabric specifications from this variety of arrangement would be formidable. It may possibly work for 1 CPU chiplet connected to 1 GPU chiplet, but we doubt AMD could wire this solution into Epyc.

Alternately, AMD could establish a GPU into just about every Ryzen chiplet, and distribute workloads throughout multiple chiplets although managing the overall array of chiplets as a contiguous graphics card. The GPU cores in each individual chiplet would presumably join to the CPU cores by means of the L3 or most likely an L4 cache. In the diagram underneath, that would equate to a GPU cluster currently being tucked in between the CCX and the “Infinity Fabric” block — separate from the CCX functionally, but still connected through the LLC.


We be expecting AMD will redesign Infinity Material for the Zen 4 die shrink and quite possibly include a new I/O die, assuming a person does not tip up for Zen three . Inserting a theoretical GPU die into the CCX topology previously mentioned does not modify the selection of Infinity Material inbound links desired across the complete chip, on the other hand, due to the fact the GPU CUs would be completely integrated into each and every chiplet.

Every Zen four chiplet would contain, say, 128-256 GPU compute cores (this number is entirely theoretical and could be larger). A 12-core or 16-core Ryzen 7000 would present commensurately extra GPU cores. A 64-main Epyc with 256 cores for every chiplet would supply a highest of two,048 GPU cores throughout the whole CPU.

AMD has been reticent to examine AI in considerably depth, but executives have explained to ExtremeTech that they aren’t blind to the fast advancements or extended-time period prospective of the business. As opposed to most of its friends, like Qualcomm, Intel, and Apple, AMD has not introduced a very low-ability NPU to market or centered on adopting SIMD instruction sets that precisely boost AI performance. If AMD needs to contend in this space — and it states it does — incorporating a CU cluster into just about every Zen four chiplet would supply a assured accelerator unit. Distributing a workload throughout a number of chiplets could also lessen very hot location development.

I suspect — again, if this leak is precise — that this is why the RDNA2 block for Zen four is eco-friendly on desktop but purple on mobile. All of AMD’s regular, monolithic patterns, for both equally Vega and RDNA2, are indicated in red. We see one solitary, solitary inexperienced block. It demonstrates up at the exact time other rumors suggest AMD will start RDNA2 baked into every single Ryzen chip.

From the commencing, AMD’s huge theme with Ryzen has been re-use, with the identical chiplet layout scaling from minimal-stop desktop to large-conclude server. A Zen 4 chiplet with integrated RDNA2 hardware (or CDNA2, if AMD went that route with Threadripper/Epyc) would give extra processing horsepower for AI calculations by leveraging AMD’s present IP somewhat than necessitating a from-scratch answer.

AMD will be facing off towards fresh levels of competition from Intel’s Alder Lake and Raptor Lake processors from late 2021-2023. But aside from the new hybrid cores coming with Alder, we really do not know considerably about these chips, other than the actuality that Intel expects substantial energy price savings thanks to the adoption of hybrid cores.

Now Go through:

Leave a comment

Your email address will not be published.