AMD CEO: Next-Generation Zen and rDNA Core Focus is Architecture, Not Process Technology – Will Transition to 5nm at the Appropriate Time
During its recent earnings call, AMD's CEO, Lisa Su, stated that upcoming Zen and rDNA cores would focus more on architecture than process technology. This statement comes soon after AMD's first full quarter of 7nm product availability which includes their EPYC, Ryzen, and Radeon lines that led to their highest quarterly revenue since 2005.
AMD's Next-Gen Zen & rDNA Cores To Focus on Architecture Over Process Technology As It's The Highest Leverage For Their Product Lineup Going Forward
Replying to a question raised by a reporter about process technology and architecture being a core part of the current 7nm lineup, AMD's CEO stated the following:
So, Timothy, the way I would answer that question is, we’ve made a set of choices, and the set of choices include process technology, they include architecture, our chiplet architecture, they include sort of our overall system architecture. And I think we’ve made it set of good choices.
Going forward, we are not relying on process technology as the main driver. We think process technology is necessary. It’s necessary to be sort of at the leading edge of process technology. And so, today, 7-nanometer is a great node, and we’re getting a lot of benefit from it. We will transition to the 5-nanometer node at the appropriate time and get great benefit from that as well. But we’re doing a lot in architecture. And I would say, that the architecture is where we believe the highest leverage is for our product portfolio going forward.
The AMD Zen and rDNA cores are based on TSMC's bleeding-edge 7nm process node and are the first high-performance parts to feature the new process technology. Given those high-end products rely on new process nodes for several years, AMD's main focus in the coming years would be to maximize the performance benefit from the existing 7nm technology, offering major performance uplifts primarily on the architecture side. This means that the upcoming Zen and rDNA family would utilize the 7nm node for years to come but this doesn't remove 5nm from the roadmap.
According to AMD's CEO, they would definitely transition to 5nm but only at an appropriate time. We know for a fact that AMD is working on their next-generation Zen and rDNA cores. AMD has confirmed future CPU cores up till Zen 5 and the next-generation rDNA architecture has also been confirmed for 2020. The thing about these cores is that while AMD won't be transitioning to sub-7nm nodes yet, they will utilize the optimized 7nm+ technology to give them a boost in terms of efficiency. Also, AMD doesn't feel the need to hurry in process technology as they are already way ahead of their competitor (Intel) who is still relying on 14nm in 2019 for their high-performance product lineups and expected to move to 10nm in 2020.
Just like Zen+ used an intermediate 12nm FinFET node to deliver on some decent efficiency upgrade on Ryzen and Radeon products, the next-gen Zen 3 and rDNA 2 cores would be based upon the 7nm+ process node which would leverage their overall efficiency. However, AMD at the same time states that the optimized node won't play a major role in terms of performance uplift as they would rely on the architecture to do that job.
AMD's recently updated roadmap showed that Zen 3, the architecture powering 'Milan' CPUs, will be arriving in 2020. The Zen 3 core would be based on the 7nm+ process node which would be put up against the 10nm Ice Lake-SP and 14nm++ Cooper Lake Xeon processors. In terms of efficiency, AMD has highlighted that their processors would offer much better performance per watt & just by looking at the slide, we can also note that even EPYC 'Rome' processors are designed to compete favorably against Intel's 2020 Xeon products. This is something that AMD had hinted back since 2018 when Rome was still being designed.
AMD's CTO, Mark Papermaster also revealed that Zen 3 is built on the foundations of Zen 2 and will primarily leverage efficiency along with an uplift to overall performance.
“TSMC may have been measuring a basic device like a ring oscillator — our claims are for a real product,”
“Moore’s Law is slowing down, semiconductor nodes are more expensive, and we’re not getting the frequency lift we used to get,” he said in a talk during the launch, calling the 7-nm migration “
Looking ahead, a 7-nm-plus node using extreme ultraviolet lithography (EUV) will “primarily leverage efficiency with some modest device performance opportunities,”
AMD already has some of the most advanced packaging solutions in the industry, with their Zen chiplet design on Ryzen and EPYC CPUs. They are only expected to further refine it in the coming generations to major architectural changes. During a presentation at the HPC-AI Advisory Council in the UK, AMD unveiled some major changes coming with their Zen 3 based EPYC Milan CPUs that include cache redesigns and the possibility of having 8 cores inside a CCX. It will be interesting to see how this pans out, you can read more here.
AMD Zen 3 7nm+ Chip Architecture - Zen 2 Refined For Next-Gen Servers and Desktops
The 7nm+ Zen 3 core design has been completed and we can see production commence sometime in 1H of 2020. While Zen 2 was the first processor architecture to be based on the 7nm node, Zen 3 would be based on an evolutionary 7nm+ node which allows for 20% more transistors than Zen 2's 7nm process. The 7nm+ process node also delivers 10% better efficiency. It is possible that we may not get as significant of a core bump as we got with Zen 2 processors but nevertheless, the added performance uplift would put more pressure on Intel in the server and desktop space.
One of the flagship products of the Zen 3 core architecture would be the 3rd Gen EPYC line known as Milan. The EPYC Milan series of processors would be deployed in the new Perlmutter super-computer, designed by CRAY. Early specifications for a singular node reveal that the Milan CPU has 64 cores and 128 threads and AVX2 SIMD (256-bit). There's also support for 8-channel DDR ram & support for greater than 256 GB per node.
Aside from Zen 3, Zen 4 and Zen 5 have also been confirmed. The 3rd Generation EPYC lineup which was officially confirmed as 'Genoa' during the 2nd Generation EPYC 'Rome' launch event. Zen 4 is currently in-design and will have a launch scheduled around 2021-2022. It will be using the sub-7nm process node and definitely be a proper upgrade over the Zen 2 / Zen 3 design. As with every Zen generation, the Ryzen and Ryzen Threadripper lineups would also be available, battling out and dominating the desktop/HEDT landscape with sheer amounts of performance & disruptive value proposition.
AMD CPU Roadmap (2017-2022)
|Architecture||Zen (1)||Zen (1) / Zen+||Zen (2) / Zen+||Zen (3) / Zen 2||Zen (3) / Zen 3 (+)||Zen (4) / Zen 3 (+)||Zen (4)||Zen (4) / Zen (5)|
|Process Node||14nm||14nm / 12nm||7nm||7nm||7nm||5nm / 6nm||5nm||5nm / 3nm|
|Server||EPYC 'Naples'||EPYC 'Naples'||EPYC 'Rome'||EPYC 'Rome'||EPYC 'Milan'||EPYC 'Genoa'||TBD||TBD|
|Max Server Cores / Threads||32/64||32/64||64/128||64/128||64/128||TBD||TBD||TBD|
|High End Desktop||Ryzen Threadripper 1000 Series (White Haven)||Ryzen Threadripper 2000 Series (Coflax)||Ryzen Threadripper 3000 Series (Castle Peak)||Ryzen Threadripper 3000 Series (Castle Peak)||Ryzen Threadripper 5000 Series (Chagall)||Ryzen Threadripper 6000 Series||Ryzen Threadripper 7000 Series||Ryzen Threadripper 8000 Series|
|Ryzen Family||Ryzen 1000 Series||Ryzen 2000 Series||Ryzen 3000 Series||Ryzen 4000/5000 Series||Ryzen 5000 Series||Ryzen 6000 Series||Ryzen 7000 Series||Ryzen 8000 Series|
|Max HEDT Cores / Threads||16/32||32/64||64/128||64/128||64/128||TBD||TBD||TBD|
|Mainstream Desktop||Ryzen 1000 Series (Summit Ridge)||Ryzen 2000 Series (Pinnacle Ridge)||Ryzen 3000 Series (Matisse)||Ryzen 5000 Series (Vermeer)||Ryzen 5000/6000 Series (Warhol)||Ryzen 6000/7000 Series (Raphael)||TBD||TBD|
|Max Mainstream Cores / Threads||8/16||8/16||16/32||16/32||16/32||16/32||TBD||TBD|
|Budget APU||N/A||Ryzen 2000 Series (Raven Ridge)||Ryzen 3000 Series (Picasso Zen+)||Ryzen 4000 Series (Renoir Zen 2)||Ryzen 5000 Series (Cezanne Zen 3)||Ryzen 6000 Series (Rembrandt Zen 3+)||Ryzen 7000 Series (Phoenix Zen 4)||Ryzen 8000 (Strix Point Zen 5)|
AMD RDNA 2 7nm+ GPU Architecture - High-End Graphics Lineup With Ray Tracing Support
Moving on to the GPU side, AMD has also revealed that the RNDA 2 GPU architecture is currently in-design and scheduled for launch in 2020. Given that the Zen 3 design has completed and RDNA 2 is still in design, we can say that the CPU launch would be much ahead of the 7nm+ GPUs. We can see a possible CPU launch in mid of 2020 with GPUs arriving in late 2020.
There isn't much we know about RDNA 2 aside from rumors but what AMD has officially talked about is Ray Tracing which will be available on their next-gen GPU lineup. With their next-gen RDNA architecture, AMD plans to have hardware-level integration on their GPUs to support real-time ray tracing in games. This would put them on par with NVIDIA’s RTX technology which implemented ray-tracing support through hardware-level integration last year with their GeForce RTX 20 series cards.
AMD also wants to push RDNA 2 towards the higher-end spectrum of the market. While the first generation RDNA GPUs perform great in the $300-$500 segments, we would likely see a range of enthusiast-grade designs with RDNA 2 based Radeon RX series graphics cards. These would take the fight to NVIDIA's RTX 2080 SUPER / RTX 2080 Ti but NVIDIA isn't the company that would just silently sit through a competitor's launch. Plans of NVIDIA's 7nm GPUs are underway and it is likely we would see a grand launch in 2020 for their next-generation graphics architecture, presumably known as 'Ampere'.
It should also be pointed out that high-end Navi GPUs might retain High-Bandwidth memory design like the current flagship. While AMD is featuring GDDR6 memory on their mainstream RDNA based cards, it is likely that the company would go ahead with the newer HBM2E VRAM.
The HBM2E DRAM comes in 8-Hi stack configuration and utilizes 16 Gb memory dies, stacked together and clocked at 3.2 Gbps. This would result in a total bandwidth of 410 GB/s on a single and 920 GB/s with two HBM2E stacks which is just insane. To top it all, the DRAM has a 1024-bit wide bus interface which is the same as current HBM2 DRAM. Samsung says that their HBM2E solution, when stacked in 4-way configuration, can offer up to 64 GB memory at 1.64 TB/s of bandwidth. Such products would only be suitable for servers/HPC workloads but a high-end graphics product for enthusiasts can feature up to 32 GB memory with just two stacks which is twice as much memory as the Radeon VII.
AMD GPU Architectures Comparison:
|Wccftech||AMD Radeon RX 400 Series||AMD Radeon RX 500 Series||AMD Radeon RX Vega Series||AMD Radeon RX 5700 Series||AMD Radeon RX 5800 Series?|
|Process Node||Global Foundries 14nm||Global Foundries 14nm||Global Foundries 14nm / TSMC 7nm||TSMC 7nm||TSMC 7nm+|
|GPUs||Polaris 10, Polaris 11||Polaris 20, Polaris 21, Polaris 22, Polaris 30||Vega 10, Vega 20||Navi 10, Navi 11||Navi 20, Navi 21, Navi 23|