No AMD Radeon "Navi" Before October: Report

[btarunr]

AMD "Navi" is the company's next-generation graphics architecture succeeding "Vega" and will leverage the 7 nm silicon fabrication process. It was originally slated to launch mid-2019, with probable unveiling on the sidelines of Computex (early-June). Cowcotland reports that AMD has delayed its plans to launch "Navi" all the way to October (Q4-2019). The delay probably has something to do with AMD's 7 nm foundry allocation for the year.

AMD is now fully reliant on TSMC to execute its 7 nm product roadmap, which includes its entire 2nd generation EPYC and 3rd generation Ryzen processors based on the "Zen 2" architecture, and to a smaller extent, GPUs based on its 2nd generation "Vega" architecture, such as the recently launched Radeon VII. We expect the first "Navi" discrete GPU to be a lean, fast-moving product that succeeds "Polaris 30." In addition to 7 nm, it could incorporate faster SIMD units, higher clock-speeds, and a relatively cost-effective memory solution, such as GDDR6.



View at TechPowerUp Main Site

 

[xkm1948]

With Navi still being GCN it is probably wont be much of a challenger to current nvidia anyway.

 

[Space Lynx]

With Navi still being GCN it is probably wont be much of a challenger to current nvidia anyway.

or even if it is a challenger it will use a lot more energy. so yeah. eh.

 

[Fluffmeister]

At least eight months away... Ouch.

 

[Basard]

With Navi still being GCN it is probably wont be much of a challenger to current nvidia anyway.

Something's gotta change. They've been running with the same GPU for years now, pretty much unchanged since Fury, and it's never been that great. They should have caught the hint after Fury's release. Now that I look, it was summer of 2015.... yeesh.

 

[FordGT90Concept]

With Navi still being GCN it is probably wont be much of a challenger to current nvidia anyway.

We don't know if Navi is GCN or isn't. All we know for sure is that it was designed for PS5, just as Polaris was designed for Xbox One X (has the full 40 compute units instead of the gimped, but higher clocked, 36 of the desktop cards).

Also, GCN has been evolving with each generation. Vega, for example, significantly changed the compute units (longer pipes allowing for higher clocks).

GCN is still a jack-of-all-trades and master-of-none architecture. I'm not convinced AMD wants to fork Radeon architectures from Instinct and Pro cards. They effectively need two engineering staffs to do that.

What GCN is really lacking is tiled rasterization. If Navi brings that, it will more align it with GeForce performance in gaming.

 

[notb]

At least 2 better reasons than "7nm allocation":
1) AMD said they're working on RTRT implementation, so it doesn't make much sense to release new cards without it. Maybe they need more time for polishing their solution.
2) 7nm is rubbish at the moment, so they'll allocate all of it to the bin and wait until TSMC improves this tech.

 

[FordGT90Concept]

1) AMD said they're working on RTRT implementation, so it doesn't make much sense to release new cards without it. Maybe they need more time for polishing their solution.

Too late in the development cycle to rush in tensor cores. AMD has also already tested internally that GCN can do DXR.

 

[IceShroom]

We don't know if Navi is GCN or isn't. All we know for sure is that it was designed for PS5, just as Polaris was designed for Xbox One X.

Also, GCN has been evolving with each generation. Vega, for example, significantly changed the compute units (longer pipes allowing for higher clocks).

Custom hardware don't work that way.
AMD creats an architecture, then AMD's partner decide what to include or not. AMD designed polaris then Microsoft decided what feature to include in their custom SoC.
And if we go by your logic, then it can be said that Nvidia designed Kepler and up for smartphone and Switch.

 

[64K]

I am eagerly awaiting what AMD will bring with 7nm Navi. I'm hoping for competition with the 2080 Ti for a reasonable price but I know that could be a stretch. In any case all AMD has to do is offer competition for Turing entry level through upper midrange level for a reasonable price and that's the vast majority of the market right there. Very few buy in at the 2080 Ti level anyway. It's just way, way out of reach financially.

 

[Aldain]

why is this not tagged as a rumor???

 

[bug]

At CES AMD announced their plans all the way to Q3. And Navi wasn't in there.
Some looked at that and concluded they want to sneak Navi in, but I was pretty sure it meant Navi is still far from launch.

Still, it matters little if Navi launches two-three months early or late. What matters is if it can take the fight to Nvidia. So far, it looks like it can't. To make things worse, in Q4'19 means it will go up against 7nm Turing (or whatever it will be called).

 

[FordGT90Concept]

Custom hardware don't work that way.
AMD creats an architecture, then AMD's partner decide what to include or not. AMD designed polaris then Microsoft decided what feature to include in their custom SoC.
And if we go by your logic, then it can be said that Nvidia designed Kepler and up for smartphone and Switch.

From cited article:

The heart of the Xbox One X is a GPU that's roughly based on AMD’s GCN 4 (Polaris) architecture. It offers 40 compute units, 2560 stream processors, and 32 ROPs. For comparison, an AMD Radeon RX 480 offers 36 CUs, so the Xbox One X offers 11% more compute hardware than the RX 480. Compared to the PlayStation 4 Pro, the Xbox One X offers about 43% more shader throughput.

There’s of course more custom blocks here as well. A console designed for 4K and HDR still needs to work with SDR 1080p displays, and the Xbox display controller can supersample down from 4K to 1080p, or even 1440p, as needed. There’s media blocks for HEVC as well, to handle the 4K video requirements for Blu-Ray and streaming, and the Xbox Game Capture can also capture at 4K now.

These are all things the desktop Polaris card can do too (and most other GCN cards). AMD can scale the number of compute units depending on application but the foundations are identical.

This one is exclusive to Polaris and newer:

Everything Xbox One X can do Polaris desktop cards can do too. It's really a chicken or the egg argument. AMD puts engineering resources into features both types of customers like.

To make things worse, in Q4'19 means it will go up against 7nm Turing (or whatever it will be called).

Ampere? Unless Ampere is replacing Volta.

 

[notb]

Too late in the development cycle to rush in tensor cores.

Adding tensor cores shouldn't be a problem. It's a separate module. Even if Navi wasn't MCM, this could literally be put in a separate chip (granted the interconnector is fast enough).
Nvidia sells a tensor accelerator and it can work in tandem with a traditional GPGPU device.

Also, how do you know they haven't started earlier? They most likely knew a lot about Nvidia's RTRT hardware all along. They may have been surprised by performance, so their solution needed more work to be competitive.

AMD has also already tested internally that GCN can do DXR.

By principle you can run RTRT (DXR) even on a x86 CPU. Actually CPUs are quite competitive in this (ray tracing is a painfully sequential mathematical problem).
Using normal GPGPU you may be able to run a 500x500px 30fps render (the "preview" window in 3D software). Purpose built hardware is orders of magnitude faster.

 

[bug]

Adding tensor cores shouldn't be a problem. It's a separate module. Even if Navi wasn't MCM, this could literally be put in a separate chip (granted the interconnector is fast enough).
Nvidia sells a tensor accelerator and it can work in tandem with a traditional GPGPU device.

Also, how do you know they haven't started earlier? They most likely knew a lot about Nvidia's RTRT hardware all along. They may have been surprised by performance, so their solution needed more work to be competitive.

By principle you can run RTRT (DXR) even on a x86 CPU. Actually CPUs are quite competitive in this (ray tracing is a painfully sequential mathematical problem).
Using normal GPGPU you may be able to run a 500x500px 30fps render (the "preview" window in 3D software). Purpose built hardware is orders of magnitude faster.

If anything, it's the performance of Turing that was known to AMD (they only had to look at Volta and extrapolate). The surprise was taking what was obviously too large a chip, putting it in what should have been a Quadro card and sell it as a very expensive consumer product.
Luckily for us Nvidia took a hit based on Turing sales, so hopefully we won't see that move ever again. But this is way OT.

 

[FordGT90Concept]

Adding tensor cores shouldn't be a problem. It's a separate module. Even if Navi wasn't MCM, this could literally be put in a separate chip (granted the interconnector is fast enough).
Nvidia sells a tensor accelerator and it can work in tandem with a traditional GPGPU device.

Also, how do you know they haven't started earlier? They most likely knew a lot about Nvidia's RTRT hardware all along. They may have been surprised by performance, so their solution needed more work to be competitive.

By principle you can run RTRT (DXR) even on a x86 CPU. Actually CPUs are quite competitive in this (ray tracing is a painfully sequential mathematical problem).
Using normal GPGPU you may be able to run a 500x500px 30fps render (the "preview" window in 3D software). Purpose built hardware is orders of magnitude faster.

Look how many games actually use DXR. Then look at the primary customer Navi is for: Sony who doesn't even use DirectX in their consoles. I just don't see tensor cores nor DXR as being a priority for AMD, especially not something to derail product timelines for. As the OP says, I think the delay is because of 7nm issues moreso than Navi itself. The fact there's limited availability of Radeon VII cards also hints at 7nm issues.

I wouldn't be surprised if Microsoft is insisting on DXR for their next Xbox which might be Arcturus architecture (follows Navi).

 

[unikin]

WFT, first abysmal Radeon VII release and now no Navi till Q4? What's wrong with AMD? NVidia is opening bottles right now. AMD left me no choice but to buy RTX 2060. I'm not waiting another year to get my hands on Navi. What a joke of a company. AMD has successfully destroyed ATI's legacy of good and affordable GPUs.

 

[bug]

Look how many games actually use DXR. Then look at the primary customer Navi is for: Sony who doesn't even use DirectX in their consoles. I just don't see tensor cores nor DXR as being a priority for AMD, especially not something to derail product timelines for. As the OP says, I think the delay is because of 7nm issues moreso than Navi itself. The fact there's limited availability of Radeon VII cards also hints at 7nm issues.

DXR is exclusive to DirectX, but RTRT isn't
https://devblogs.nvidia.com/vulkan-raytracing/

That said, I have no idea whether Navi will have RTRT capabilities or not.
Also, TSMC has run into prod issues before and had to scrape their entire 22nm node. I imagine 7nm is harder than that, even with the lessons learned, so yeah, prod issues combined with GloFo's downshifting on 7nm is what probably cause the delay.

WFT, first abysmal Radeon VII release and now no Navi till Q4? What's wrong with AMD? NVidia is opening bottles right now.

Nvidia has been opening bottles ever since Vega (if not earlier). They may have run out of bottles to open by now.

 

[FordGT90Concept]

DXR is exclusive to DirectX, but RTRT isn't
https://devblogs.nvidia.com/vulkan-raytracing/

AMD launched their RTRT extension for Vulkan almost a year ago. There's Radeon Rays (OpenCL solution) before that too.

 

[64K]

WFT, first abysmal Radeon VII release and now no Navi till Q4? What's wrong with AMD? NVidia is opening bottles right now.

You have to understand AMD's position financially to fully understand why they lag behind:

2017 Annual Financial Reports

Intel 71 billion dollars revenue and 21 billion dollars profit
Nvidia 9.7 billion dollars revenue and 3 billion dollars profit
AMD 5.3 billion dollars revenue and 43 million dollars profit (this includes both CPU and GPU businesses)

The problem AMD has is pretty obvious and frankly it's amazing that they were even able to bring Ryzen to market as it is.

 

[IceShroom]

From cited article:

These are all things the desktop Polaris card can do too (and most other GCN cards). AMD can scale the number of compute units depending on application but the foundations are identical.
View attachment 116039
This one is exclusive to Polaris and newer:
View attachment 116040
Everything Xbox One X can do Polaris desktop cards can do too. It's really a chicken or the egg argument. AMD puts engineering resources into features both types of customers like.


Ampere? Unless Ampere is replacing Volta.

Cause those are feature of Polaris and Microsoft decided to implement those on their console.
DXR was introduced for GCN 2 and up and a feature of AMD.
And new GPU architecture AMD must support new video codec and thats why every polaris and up has HEVC support.

 

[Space Lynx]

At least 2 better reasons than "7nm allocation":
1) AMD said they're working on RTRT implementation, so it doesn't make much sense to release new cards without it. Maybe they need more time for polishing their solution.
2) 7nm is rubbish at the moment, so they'll allocate all of it to the bin and wait until TSMC improves this tech.

something tells me 7nm factories are about to start mass production on playstation 5 chips as well. sells of all gpu's are low, sells of cpu's are becoming saturated. apple phones are not selling that fast anymore, TSMC has more spare time than ever before, and sony is probably being more aggressive since it will be a fully backwards compatible ps5. i for one am buying a ps5 on launch day... i hope sony surprises everyone with a spring 2020 release.

 

[Aldain]

You have to understand AMD's position financially to fully understand why they lag behind:

2017 Annual Financial Reports

Intel 71 billion dollars revenue and 21 billion dollars profit
Nvidia 9.7 billion dollars revenue and 3 billion dollars profit
AMD 5.3 billion dollars revenue and 43 million dollars profit (this includes both CPU and GPU businesses)

The problem AMD has is pretty obvious and frankly it's amazing that they were even able to bring Ryzen to market as it is.

2017??? dude you kinda "internet explorer here"...

again.. why is this topic not tagged as a rumor???

 

[64K]

2017??? dude you kinda "internet explorer here"...

2017 is the latest Yearly Financial Report. 2018 isn't in yet.

I hear it constantly that AMD is lazy or incompetent and that used to be true but not since Lisa Su became CEO. They dug themselves into this financial pit by being that way in the past but once again to fully understand why they lag behind you have to look at their present financials. They are a very, very small company compared to Intel and they are small compared to Nvidia as well.

 

[unikin]

You have to understand AMD's position financially to fully understand why they lag behind:

2017 Annual Financial Reports

Intel 71 billion dollars revenue and 21 billion dollars profit
Nvidia 9.7 billion dollars revenue and 3 billion dollars profit
AMD 5.3 billion dollars revenue and 43 million dollars profit (this includes both CPU and GPU businesses)

The problem AMD has is pretty obvious and frankly it's amazing that they were even able to bring Ryzen to market as it is.

I understand where AMD stands financially, BUT they should not lie to their own fans. I have been postponing GPU buy based on AMD's GPU roadmap for more than a year and when bloody Navi should at least get release date AMD goes silent and offers us abysmal overpriced and nowhere to be found R7 - or to be truthful quality failed and crippled Instinct MI150 instead. WTF, what kind of PR is this? They're doing marvelous job if they want to lose remaining 28 % share of PC gaming market. I always bought AMD for my gaming rig just to support the underdog and partially because of ATI nostalgia (I was forced to go green for my workstations because of Adobe) but I'm not willing to go red anymore. They should admit it if they don't have competitive architecture, close the doors and go back to their drawing boards until they have it, like they did with Ryzen. Saying to fans that they will compete and then offer us bullshit after bullshit is the worst thing they can do. Someone should get fired in their marketing department. Their last good release was Polaris 30 months back for god sake.