The AMD Execution Thread [2007 - 2017]

[kalelovil]

I still have my doubts, I don't see it being that much further away from what they have now. What would be the use of a next gen next gen lol. Simple put, why would AMD try a next gen big part, and not use the same IP for the small part? Usually with new processes, and what not they would do it with the small part first.... As if anything goes wrong the need for a sustained lower power and higher performance would be a risk.

They might have planned for Polaris to be out significantly earlier.

I am reminded of the HD 6xxx generation (plus a node shrink).
68xx: Re-balanced HD 58xx derivative, removed FP64, unchanged ISA (?), faster tesselation, expanded UVD and display engines.
69xx: Larger next-generation design released a few months later

An interested section from Anandtech's HD 68xx review:

However it’s worth noting that internally AMD was throwing around 2 designs for Barts: a 16 SIMD (1280 SP) 16 ROP design, and a 14 SIMD (1120 SP) 32 ROP design that they ultimately went with. The 14/32 design was faster, but only by 2%.

 

[Razor1]

Well the barts designs weren't much different from each other were they?

The issue here is the time to create another architecture with things like frequency and voltage in mind, those will change design drastically, and given an extra 6 months, ain't going to do the trick.....

And if AMD has the budget to do two different design teams with two drastically different architectures in such a short amount of time, I would be surprised. We have not seen that from nV or Intel who have much bigger R&D ever.

 

[eastmen]

Well the barts designs weren't much different from each other were they?

The issue here is the time to create another architecture with things like frequency and voltage in mind, those will change design drastically, and given an extra 6 months, ain't going to do the trick.....

And if AMD has the budget to do two different design teams with two drastically different architectures in such a short amount of time, I would be surprised. We have not seen that from nV or Intel who have much bigger R&D ever.

Polaris could have been a 22nm design pushed down to 14nm when that node hit trouble. Vega could have always been the 14nm

 

[Razor1]

Hmm why would they use the Polaris design then? If Vega is much better at what ever its going to do an architecture level for power and performance, there would be no need to Polaris's design to be shifted to 14nm.

 

[Kaotik]

Hmm why would they use the Polaris design then? If Vega is much better at what ever its going to do an architecture level for power and performance, there would be no need to Polaris's design to be shifted to 14nm.

Because it's probably faster to convert 22nm design to 14nm than to create whole new chip based on new architecture

 

[AlNom]

20

 

[Alexko]

Polaris could have been a 22nm design pushed down to 14nm when that node hit trouble. Vega could have always been the 14nm

That thought has crossed my mind as well. @Dave Baumann, is this something you can comment on?

 

[Razor1]

Because it's probably faster to convert 22nm design to 14nm than to create whole new chip based on new architecture

Going from TSMC 20nm to 14nm GF, that would require quite a bit of work..... I don't see how time scaling and savings fit into that.

You don't make back up plans based off of failed node designs and push them to new nodes, cause you already have been working on 14nm designs well before the failure of the 20nm node was know.

Well then they would have had to have thought their 14nm midrange chip would take longer to design and bring to manufacturing, so they dropped that? Just isn't logical, unless they have planned for an entire Vega line up, and now dropped its midrange version or going to do a quick replacement of their midrange with in a year. Which would just be not cost effective either way.

At this point Vega's only delay (well not delay but release schedule) is because of HBM2 and their exclusivity contract with Hynix.

 

[Anarchist4000]

Going from TSMC 20nm to 14nm GF, that would require quite a bit of work..... I don't see how time scaling and savings fit into that.

You don't make back up plans based off of failed node designs and push them to new nodes, cause you already have been working on 14nm designs well before the failure of the 20nm node was know.

Well then they would have had to have thought their 14nm midrange chip would take longer to design and bring to manufacturing, so they dropped that? Just isn't logical, unless they have planned for an entire Vega line up, and now dropped its midrange version or going to do a quick replacement of their midrange with in a year. Which would just be not cost effective either way.

At this point Vega's only delay (well not delay but release schedule) is because of HBM2 and their exclusivity contract with Hynix.

Actually substantially different designs wouldn't be implausible because of the memory that is getting used. Maybe Vega is exclusively HBM and somewhat locked to higher tier parts with costs for the time being. The available bandwidth between GDDR5X and HBM wouldn't be insignificant if a certain amount of memory (8/12/16GB) was required. Denser HBM chips may be required before pushing down to midrange products. May have very well been more cost effective to use than writing off the 20nm design and creating a 14nm for GDDR5 with a projected short shelf life before HBM prices decrease. An 8GB HBM2 480 would probably be a bit too much bandwidth and costs for the time being.

 

[Razor1]

They already wrote of the 20nm design though

To transition to 14nm from a already designed chip for another node is a lot of money in the order of 20 million + without going into mass production, and to do that with 2 designs and not use one or double back and replace one quickly, will cost more than 20 million + not to mention the possible risk of loss from having two designs overlapping with stock amounts.

So in theory you aren't hedging your bets, you are increasing your bottom line risk.

 

[Anarchist4000]

They already wrote of the 20nm design though

To transition to 14nm from a already designed chip for another node is a lot of money in the order of 20 million + without going into mass production, and to do that with 2 designs and not use one or double back and replace one quickly, will cost more than 20 million + not to mention the possible risk of loss from having two designs overlapping with stock amounts.

So in theory you aren't hedging your bets, you are increasing your bottom line risk.

What I'm saying is that one of the designs may not have been feasible, at this time, and there would have been significant architectural changes required. Polaris having the same IP as Fiji/Tonga may not have been a coincidence. Maybe GDDR isn't supported with Vega in exchange for a bunch of interconnects for interfacing with MCMs. The voltages and cache configurations could change radically.

 

[Kaotik]

What I'm saying is that one of the designs may not have been feasible, at this time, and there would have been significant architectural changes required. Polaris having the same IP as Fiji/Tonga may not have been a coincidence. Maybe GDDR isn't supported with Vega in exchange for a bunch of interconnects for interfacing with MCMs. The voltages and cache configurations could change radically.

"Vega generation" can't really require HBM, since it's coming to Raven Ridge too (Zen APU, not the HPC one), which will no doubt deal with DDR4 shared with the CPU-portion. But then, the GPUs use several different IP blocks and gfx ip is just one of them

 

[eastmen]

Hmm why would they use the Polaris design then? If Vega is much better at what ever its going to do an architecture level for power and performance, there would be no need to Polaris's design to be shifted to 14nm.

Time frames ? Looks like Vega wont hit till at least late fall from rumors. So they needed something before then. Also if you look at AMD's line up Vega seems to be the higher end , even a cut down Vega would only go so far down the product line. Enter Polaris and you have a product stack down to the $100 price point. its smaller than vega so it will always be cheaper to produce. Vega can have the $300 and up spot for AMD

 

[eastmen]

They already wrote of the 20nm design though

To transition to 14nm from a already designed chip for another node is a lot of money in the order of 20 million + without going into mass production, and to do that with 2 designs and not use one or double back and replace one quickly, will cost more than 20 million + not to mention the possible risk of loss from having two designs overlapping with stock amounts.

So in theory you aren't hedging your bets, you are increasing your bottom line risk.

Yea but look at their product stack. Hawaii and the previous chips wont sell at the $100-$300 range. They run hot and use a ton of power compared to NVidia's older cards , compared to the newer ones its not even a contest .

So AMD always needed something more efficient to replace Hawaii with . Enter Polaris , its 500m tranistors smaller , uses half the memory bus width and uses almost half the power of Hawaii.

AMD would have known the problems of the other node so they could have made changes years ago to their plans. Vega may have always been the 14nm chip and sole 14nm chip. But without a 20/22nm part to replace Hawaii AMD knew they wouldn't have anything for the low end. 20+m sounds like a lot of money esp if they didn't make 20/22nm masks or run any test chips for Polaris on the previous node.

 

[Razor1]

Time frames ? Looks like Vega wont hit till at least late fall from rumors. So they needed something before then. Also if you look at AMD's line up Vega seems to be the higher end , even a cut down Vega would only go so far down the product line. Enter Polaris and you have a product stack down to the $100 price point. its smaller than vega so it will always be cheaper to produce. Vega can have the $300 and up spot for AMD

Vega, being less than 350 even at 400 I think is going to be hard if they are using HBM2. Well if they were thinking that, then there will be no midrange Vega and that is what I was saying to begin with. Vega and Polaris was always the plan for 14nm, Polaris wasn't or isn't a transition from 20nm to 14nm.

 

[Razor1]

Yea but look at their product stack. Hawaii and the previous chips wont sell at the $100-$300 range. They run hot and use a ton of power compared to NVidia's older cards , compared to the newer ones its not even a contest .

So AMD always needed something more efficient to replace Hawaii with . Enter Polaris , its 500m tranistors smaller , uses half the memory bus width and uses almost half the power of Hawaii.

AMD would have known the problems of the other node so they could have made changes years ago to their plans. Vega may have always been the 14nm chip and sole 14nm chip. But without a 20/22nm part to replace Hawaii AMD knew they wouldn't have anything for the low end. 20+m sounds like a lot of money esp if they didn't make 20/22nm masks or run any test chips for Polaris on the previous node.

20 million + is what AMD wrote off for their failed 20nm transition. Now that might not even be the full price to get to mass production, it could be just the amount used before they figured out 20nm was a fail.

 

[Anarchist4000]

"Vega generation" can't really require HBM, since it's coming to Raven Ridge too (Zen APU, not the HPC one), which will no doubt deal with DDR4 shared with the CPU-portion. But then, the GPUs use several different IP blocks and gfx ip is just one of them

Require HBM no, GMI links, maybe. For an APU it would be cheap as your GPU would use system memory in most cases. For a discrete, low end card using the link would likely be an unnecessary expense. Linking multiple chips, dual GPU, etc they might still be necessary. Point being, some sort of fabric in the place of the memory controller likely isn't warranted on a mid-range product. Ideally low/mid tier products would be replaced by correspondingly powerful APUs. Throw in a single stack of HBM that can benefit the CPU portion as well for performance variants.

 

[eastmen]

Vega, being less than 350 even at 400 I think is going to be hard if they are using HBM2. Well if they were thinking that, then there will be no midrange Vega and that is what I was saying to begin with. Vega and Polaris was always the plan for 14nm, Polaris wasn't or isn't a transition from 20nm to 14nm.

well that's my point even small vega is going to be to expensive to replace the low end with. So Polaris certainly can be a transition

20 million + is what AMD wrote off for their failed 20nm transition. Now that might not even be the full price to get to mass production, it could be just the amount used before they figured out 20nm was a fail.

Right but that money is gone . How much would it cost to take a chip designed for that 20nm process and move it to 14nm vs designing a whole new chip ?

 

[kalelovil]

AMD would have known the problems of the other node so they could have made changes years ago to their plans. Vega may have always been the 14nm chip and sole 14nm chip. But without a 20/22nm part to replace Hawaii AMD knew they wouldn't have anything for the low end. 20+m sounds like a lot of money esp if they didn't make 20/22nm masks or run any test chips for Polaris on the previous node.

That might help explain why Polaris achieves the same transistors/mm2 gain from the supposedly more dense GF 14nm process as Pascal gets from the TSMC 16nm process.
(Or it's just a result of AMD's lower R&D budget; less man-hours to devote to the process transition)

 

[Razor1]

well that's my point even small vega is going to be to expensive to replace the low end with. So Polaris certainly can be a transition

Right but that money is gone . How much would it cost to take a chip designed for that 20nm process and move it to 14nm vs designing a whole new chip ?

Thats the whole thing, AMD is cash strapped they won't be doing both......