Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]

[Deleted member 13524]

If I didn't hear it wrong, she did mention that Navi would be coming later this year.

Did she actually say Navi by name? I remember her mentioning "future 7nm GPUs" to release in 2019 though.
I'm just asking this because not mentioning Navi could be an indication of AMD not using the architecture name in the final product names, like they did with Vega (and started doing with Radeon VII).

Or surely they would have been all over this? Does this kill any chance of RTRT in next gen consoles?

On the contrary.
They just released an expensive high-end GPU without RT. Mentioning that their next-gen GPU architecture has RT hardware would put an Osborne effect on their new graphics card.

So regardless of whether they have RT hardware on Navi or not, they shouldn't have talked about it at CES.
They may even have RT hardware in Navi, release the cards and omit the functionality until the PS5/XBTwo release, just like they omitted the TruAudio DSPs in the Radeon HD7790 (released March 2013) until the PS4 released 8 months later.

 

[Kaotik]

Did she actually say Navi by name? I remember her mentioning "future 7nm GPUs" to release in 2019 though.
I'm just asking this because not mentioning Navi could be an indication of AMD not using the architecture name in the final product names, like they did with Vega (and started doing with Radeon VII).

She said future GPUs in 2019 at some point, but in the very end of the whole show she did mention Navi by name

 

[mpg1]

anybody figure how much area a 7nm 8 core zen chiplet takes up on the die?

 

[anexanhume]

anybody figure how much area a 7nm 8 core zen chiplet takes up on the die?

Between 70 and 80 mm^2, based on measurements of package pictures.

I thought this image was interesting too, since the issue of 4-core CCXs seems unresolved. Credit to Reddit.

AMD is working on it. She didn't mention it on stage though. She says they'll have more to announce on RT later in the year. Citation below. Take of it as you will, but this pretty much reinforces for me that we'll be seeing hybrid ray tracing on next gen consoles. I'm leaning on MS here as we saw them on stage and large parts of DX12 is made with GCN architecture in mind, but i hope PS5 has it too. Machine Learning reinforces the entire azure business model. The whole industry moving towards machine learning for animation, AI up-res, etc, are all profitable for MS. Storage, processing, and data collection is their main business. The idea that next generation consoles can run AI models very quickly means there's opportunity for them to look at their current 360BC program, and for X2, run AI-up-resolution on textures and resolution. Third party companies will focus on content creation and leveraging AI to do all the modelling work so that studios only need to buy the service, much cheaper than constantly hiring tons and tons of animators and artists. All of it being processed of course on Azure, and a whole industry of content creation on Azure is where the big money for them will be.

We already know that AMD is a big part of Azure. We know the next Scarlett SoCs have some Azure links in them. We only need to bring this picture together and see how MS profits or maximizes the cost efficiency from all of this, and stop looking at it from a 'trying to beat' their competitors. They're trying to take home money and lots of it, hard to see MS pass up such an opportunity.

https://www.pcworld.com/article/3332205/amd/amd-ceo-lisa-su-interview-ryzen-raytracing-radeon.html

“I think ray tracing is an important technology, and it’s something we’re working on as well, both from a hardware and software standpoint,” Su said. “The most important thing, and that’s why we talk so much about the development community, is technology for technology’s sake is okay, but technology done together with partners who are fully engaged is really important.”

Nvidia has received some criticism from enthusiasts concerning the price of its RTX cards and the relative of lack of game support at present. Su indicated that building a development ecosystem was important.

Later, Su expanded on her thought. “I don’t think we should say that we are ‘waiting,’” Su said, in response to this reporter’s question. “We are deep in development, and that development is concurrent between hardware and software.”

“The consumer doesn’t see a lot of benefit today because the other parts of the ecosystem are not ready,” Su added. [iroboto: DirectML is not ready for instance]

Good summary.

This paper and patent shows they’ve been considering HW accelerated RT for at least 4 years.

https://pdfs.semanticscholar.org/26ef/909381d93060f626231fe7560a5636a947cd.pdf

https://patentimages.storage.googleapis.com/9a/4e/87/9e66d9a430c575/US20130328876A1.pdf

 

[AlNom]

I'm guessing that if they haven't pimped a native oct-core, then it's just a 2x4 design again. Plus they probably had enough to modify between the fab process and other enchantments in the architecture before mucking around with L3 cache.

Would be worth considering/remembering the goal of the chiclet design is for yields, and 2x4 would probably just end up being simpler for creating a whole line-up of SKUs on desktop/mobile just as we've already seen.

Cynical-me also says chiclets and duct tape gum. :V

 

[mpg1]

Ok this math could be way off but:

If we take the Original Xbox One (363mm^2) and Xbox One X (359mm^2) as a reference of what launch next-gen die size may be.

And subtract 75mm^2 for the cpu. We are left with 285mm^2 for the GPU.

Vega VII is 331mm^2 for a 64CU(60+4 disabled). So about 5.17mm^2 for each CU.

285/5.17 gives about 55 CU's. You could round that up to 56 I guess.

56CU's would probably give 52 active CU's?

So basically for a chip size the same as launch Xbox One and Xbox One X would give a 8 core Zen + 52 active CU's.

If they manage better clocks ie 1.3GHz vs Xbox One X's 1.17GHZ.... it would give 8.652 Tflops.

Not overly impressive but it doesn't take into account significant improvements from zen, gpu architecture and memory and also assumes they stick with about a 360mm^2 chip.

 

[anexanhume]

Ok this math could be way off but:

If we take the Original Xbox One (363mm^2) and Xbox One X (359mm^2) as a reference of what launch next-gen die size may be.

And subtract 75mm^2 for the cpu. We are left with 285mm^2 for the GPU.

Vega VII is 331mm^2 for a 64CU(60+4 disabled). So about 5.17mm^2 for each CU.

285/5.17 gives about 55 CU's. You could round that up to 56 I guess.

56CU's would probably give 52 active CU's?

So basically for a chip size the same as launch Xbox One and Xbox One X would give a 8 core Zen + 52 active CU's.

If they manage better clocks ie 1.3GHz vs Xbox One X's 1.17GHZ.... it would give 8.652 Tflops.

Not overly impressive but it doesn't take into account significant improvements from zen, gpu architecture and memory and also assumes they stick with about a 360mm^2 chip.

If Navi doesn’t have FP64, INT8/INT4, that will cut down on size. Its memory controller is also likely to be smaller than a 4 stack HBM2 controller. Finally, we have no reason to assume it will be monolithic, meaning the GPU could be it’s own chiplet.

Outside of architectural changes, there’s also likely something to be said for a native 7nm design vs. a shrunk Vega core.

 

[function]

Ok this math could be way off but:

If we take the Original Xbox One (363mm^2) and Xbox One X (359mm^2) as a reference of what launch next-gen die size may be.

And subtract 75mm^2 for the cpu. We are left with 285mm^2 for the GPU.

Vega VII is 331mm^2 for a 64CU(60+4 disabled). So about 5.17mm^2 for each CU.

285/5.17 gives about 55 CU's. You could round that up to 56 I guess.

56CU's would probably give 52 active CU's?

So basically for a chip size the same as launch Xbox One and Xbox One X would give a 8 core Zen + 52 active CU's.

If they manage better clocks ie 1.3GHz vs Xbox One X's 1.17GHZ.... it would give 8.652 Tflops.

Not overly impressive but it doesn't take into account significant improvements from zen, gpu architecture and memory and also assumes they stick with about a 360mm^2 chip.

Or ... they use a CPU chiplet so they can have the latest possible architecture revision (and the most ideal perf/$$ bin a few months before launch) and go with a Vega 20 busting 360 mm^2 die for the GPU/NB alone.

And then you make an up-specced Ace Combat VI port for it. And it becomes the best system ever.

But seriously, a CPU chiplet makes a lot of sense for a launch system, especially if you go for a two tier setup. You pass the cost of the top binned chips onto the luxury buyer, and go for a slightly slower but much cheaper chip for the base system. And you don't need to lock down your CPU design with a branch from the mainline architecture that's two years old by the time the system launches. CPUs are almost certainly not as modular and adaptable - and suitable for parallel development - as GPUs.

 

[AlNom]

Vega VII is 331mm^2 for a 64CU(60+4 disabled). So about 5.17mm^2 for each CU.

Gotta have ROPs, L2 cache, front-end, I/O, media blocks etc.

 

[function]

Gotta have ROPs, L2 cache, front-end, I/O, media blocks etc.

Don't forget shape audio and *cough* kinect sram blocks ....

 

[anexanhume]

Gotta have ROPs, L2 cache, front-end, I/O, media blocks etc.

Speaking seriously, I wonder if Navi will up the cache like Turing has. I’ve seen some patent traffic about them compressing writes to the cache, so they may be aiming to increase efficiency that way.

 

[vipa899]

They may even have RT hardware in Navi, release the cards and omit the functionality until the PS5/XBTwo release,

That would mean they gave up the pc GPU battle, lets hope thats not the case.

 

[Deleted member 13524]

That would mean they gave up the pc GPU battle, lets hope thats not the case.

They "gave up the PC GPU battle" if they don't support, at launch, a feature that is currently implemented in one (1) title with a huge performance hit and very questionable IQ enhancements?

Isn't that a bit dramatic?

 

[Tkumpathenurpahl]

They "gave up the PC GPU battle" if they don't support, at launch, a feature that is currently implemented in one (1) title with a huge performance hit and very questionable IQ enhancements?

Isn't that a bit dramatic?

Yes, but not by Internet standards. There were no Hitler or rape references.

 

[mrcorbo]

More food for thought: AMD's GPU development history

AMD's Top-end Single GPU Video Card
2012 - 7970 - 28nm - 3.79 TF
2013 - R9 290X - 28nm - 5.63 TF
2015 - Fury X - 28nm - 8.6 TF
2017 - Vega 64 (Air) @ Boost Clock - 14nm - 12.67 TF
2019 - Radeon VII @ Boost Clock - 7nm - 13.82 TF

Consoles based on AMD GPU
2013 - PS4 - 28nm -1.8 TF
2016 - PS4 Pro - 14nm - 4.2 TF
2017 - Xbox One X - 14nm - 6 TF

A console based on a SoC containing an AMD GPU has never had more than 1/2 the total TF capability of the top-end GPU from AMD at any time during this period. Keep that in mind when setting your expectations for the performance of next-gen consoles.

 

[anexanhume]

More food for thought: AMD's GPU development history

AMD's Top-end Single GPU Video Card
2012 - 7970 - 28nm - 3.79 TF
2013 - R9 290X - 28nm - 5.63 TF
2015 - Fury X - 28nm - 8.6 TF
2017 - Vega 64 (Air) @ Boost Clock - 14nm - 12.67 TF
2019 - Radeon VII @ Boost Clock - 7nm - 13.82 TF

Consoles based on AMD GPU
2013 - PS4 - 28nm -1.8 TF
2016 - PS4 Pro - 14nm - 4.2 TF
2017 - Xbox One X - 14nm - 6 TF

A console based on a SoC containing an AMD GPU has never had more than 1/2 the total TF capability of the top-end GPU from AMD at any time during this period. Keep that in mind when setting your expectations for the performance of next-gen consoles.

And we don’t have any info on Navi yet. In these generations, could you halve the power and only give up 25% performance?

On top of that. We doubled PS4 in three years and tripled it in 4 years. We can’t double X1X in three years with the advantage of moving from 16nm to 7nm where we get the same performance for half the power?

 

[iroboto]

More food for thought: AMD's GPU development history

AMD's Top-end Single GPU Video Card
2012 - 7970 - 28nm - 3.79 TF
2013 - R9 290X - 28nm - 5.63 TF
2015 - Fury X - 28nm - 8.6 TF
2017 - Vega 64 (Air) @ Boost Clock - 14nm - 12.67 TF
2019 - Radeon VII @ Boost Clock - 7nm - 13.82 TF

Consoles based on AMD GPU
2013 - PS4 - 28nm -1.8 TF
2016 - PS4 Pro - 14nm - 4.2 TF
2017 - Xbox One X - 14nm - 6 TF

A console based on a SoC containing an AMD GPU has never had more than 1/2 the total TF capability of the top-end GPU from AMD at any time during this period. Keep that in mind when setting your expectations for the performance of next-gen consoles.

Yea lol.
Given the price points we are working with I’m often baffled at the call for 10+ TF as being reasonable.

Newer features will define the generation not power output. Variable rate shading, mesh shaders/primitive, GPU side draw calls, RT, ML res, animation, physics and denoising. Etc etc.

6-7TF with this feature set is going to drive much larger gains in image quality and performance than simply more flops.

That being said, without a doubt they will cram in as much power as they can, but should not do so at the cost of these new features.

 

[mrcorbo]

Ok this math could be way off but:

If we take the Original Xbox One (363mm^2) and Xbox One X (359mm^2) as a reference of what launch next-gen die size may be.

And subtract 75mm^2 for the cpu. We are left with 285mm^2 for the GPU.

Vega VII is 331mm^2 for a 64CU(60+4 disabled). So about 5.17mm^2 for each CU.

285/5.17 gives about 55 CU's. You could round that up to 56 I guess.

56CU's would probably give 52 active CU's?

So basically for a chip size the same as launch Xbox One and Xbox One X would give a 8 core Zen + 52 active CU's.

If they manage better clocks ie 1.3GHz vs Xbox One X's 1.17GHZ.... it would give 8.652 Tflops.

Not overly impressive but it doesn't take into account significant improvements from zen, gpu architecture and memory and also assumes they stick with about a 360mm^2 chip.

Why does this look familiar?

 

[troyan]

https://www.pcworld.com/article/3332205/amd/amd-ceo-lisa-su-interview-ryzen-raytracing-radeon.html

“I think ray tracing is an important technology, and it’s something we’re working on as well, both from a hardware and software standpoint,” Su said. “The most important thing, and that’s why we talk so much about the development community, is technology for technology’s sake is okay, but technology done together with partners who are fully engaged is really important.”

Nvidia has received some criticism from enthusiasts concerning the price of its RTX cards and the relative of lack of game support at present. Su indicated that building a development ecosystem was important.

Later, Su expanded on her thought. “I don’t think we should say that we are ‘waiting,’” Su said, in response to this reporter’s question. “We are deep in development, and that development is concurrent between hardware and software.”

“The consumer doesn’t see a lot of benefit today because the other parts of the ecosystem are not ready,” Su added. [iroboto: DirectML is not ready for instance]

So, basically she waits that other do the job? Wow.
Her statements dont make any sense.

 

[iroboto]

So, basically she waits that other do the job? Wow.
Her statements dont make any sense.

I think a while back they were asked about RT when the MI60 was announced. And the spokesman said that they were waiting.

I think after going on stage, there is some regret in saying that they were waiting. And that they have actually been deep in development for RT.

I think they don’t want to outright say to investors that they aren’t ready yet and it may not be for a little bit longer.