Quick Notes: Navi Refresh and RDNA2 Both In 2020, According to AMD
by Ryan Smith on January 28, 2020 9:00 PM EST
As part of today’s FY2019 earnings call, AMD CEO Dr. Lisa Su had a few words to say about AMD’s future GPU plans – an unexpected nugget of information since we weren’t expecting AMD to reveal anything further at this time.
In short, for this year AMD is planning on both Navi product refreshes as well as parts based on the forthcoming RDNA 2 GPU architecture. To quote Lisa Su:
In 2019, we launched our new architecture in GPUs, it's the RDNA architecture, and that was the Navi based products. You should expect that those will be refreshed in 2020 - and we'll have a next generation RDNA architecture that will be part of our 2020 lineup. So we're pretty excited about that, and we'll talk more about that at our financial analyst day. On the data centre GPU side, you should also expect that we'll have some new products in the second half of this year.
All told, it looks like AMD is setting themselves up for a Vega-like release process, launching new silicon to replace their oldest existing silicon, and minting new products based on existing and/or modestly revised silicon for other parts of their product stack. This would be very similar to what AMD did in 2017, where the company launched Vega at the high-end, and refreshed the rest of their lineup with the Polaris based Radeon RX 500 series.
AMD's GPU Roadmap As Of July 2019
But as always, the devil is in the details. And for that, we’ll have to stay tuned for AMD’s financial analyst day in March.
Source: AMD FY2019 Earnings Call
Facebook
Twitter
RSS
89 Comments
View All Comments
Yojimbo - Saturday, February 1, 2020 - link
I believe RT cores do not have SIMD/SIMT compute elements. Whether any of the shader cores are used in the computation I do not know, but I do expect there to be special compute cores. As far as tensor cores, I think most likely they are rewiring of the existing shaders, as I said. However, it's not correct to say the cores "do not exist". That's like saying neutrons don't exist because they are made of quarks, or trees don't exist because they have chloroplasts and chloroplasts can exist on their own.SaberKOG91 - Saturday, February 1, 2020 - link
They only added silicon for BVH accelerators to groups of shaders because the traversal is not efficiently computed otherwise. The rest of RT math is all floating point and can be represented as SIMD vector operations. Once you have scheduled an RT program against an SM, the rest of the resources are unavailable for further computation. This is why we see such a heavy drop in performance with RT enabled.All of Nvidia's marketing around simultaneous operations relies on different SMs running different kinds of shader programs. The only actual operations that you can run in parallel within a group of 8 shaders are int32 and fp32 operations. Once those shaders are allocated as Tensor Cores, it appears you can't use them for int32 either. You definitely can't do tensor ops and RT ops at the same time within an SM and I'm pretty sure the int32 resources are used for RT ops.
I meant that Tensor Cores don't exist as distinctly separate hardware like the marketing spin would have us believe. Of course they are physically represented as changes to the structure within an SM.
neblogai - Wednesday, January 29, 2020 - link
AMD might do a simple refresh of Navi10 cards. Right now, 5600XT is on the heels of 5700. So, they may just do something like 5700XT refresh that is Navi10 but with a boost from 16Gbps memory and higher GPU clocks, and 5700 refresh with the same VRAM but higher GPU clocks.Hul8 - Wednesday, January 29, 2020 - link
Or they could do a "cost down", "performance down" version of 5700/XT:Use the existing Navi 10 GPU, but pair it with cheaper memory and lower-end power delivery, maybe cooling, to generate "6600" series cards. They'd not only be cheaper in order to hit that x600 series price range, but would also offer less headroom to overclock to levels similar to the possible "6700 non-XT".
The 6600 non-XT could even be based on 5600 XT, so the memory bandwidth would differentiate 6600 and 6600 XT more.
haukionkannel - Wednesday, January 29, 2020 - link
This is what I was Also thinking.Low end will continue to be produced witouth raytrasing aka rdna1 and highend will have raytrasing aka rdna2. They may Also overlap!
You can get either get faster $300 gpu without raytrasing or slover $300 gpu with raytrasing.
Spunjji - Wednesday, January 29, 2020 - link
I'm of the opinion that's how Nvidia should have done it too, instead of dragging RTX all the way down to the 2070 and 2060.Alistair - Tuesday, January 28, 2020 - link
I don't mind waiting 1 or 2 years between releases, but only if it is a full product stack. We still have 1080 ti performance from AMD 3 years later. 3.5 years to catch up is a bit sad, kind of a sign of the times, GPUs are just not improving fast enough anymore (nVidia also).Cellar Door - Wednesday, January 29, 2020 - link
So don't wait but the current flagship and stop complaining.Alistair - Wednesday, January 29, 2020 - link
bought a 1080 4 years ago, the 5700 xt is hardly an upgrade, i've been waiting a long time, might as well complain after 4 years... hope the 5800 xt is coming soon and next second halfSpunjji - Wednesday, January 29, 2020 - link
I'm confused about why it's a bad thing that your high-end card still provides high-end performance 4 years later. It's not like games have suddenly become more demanding and we're all drowning in low frame rates.The 5700XT is a good upgrade for the people who are still running Maxwell-generation GPUs. It's simply not meant for you.