-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
** The AMD Navi Thread **
- Thread starter subbytna
- Start date
Because AMD won't be making Navi chips next year
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,379
DP pointed out navi isnt coming in 2018 as vega 20 is due out first, you seem to be back peddling or getting a bit confused.
Iirc glofo volume production of 7nm is due for 2h 2018, not 1h.
Last edited:
I just hope that they are able to finally scale down the arch and die sizes in order to produce a full range of GPUs from low end to high end, by cutting down the big die, a la Nvidia,
It would really make their driver development so much more efficient and potentially require less financial resources for driver development, and maybe more potential of having good day one drivers , they would have a full range of GPUs with the same software features, that would benefit from the same driver improvements at the same time.
None of this staggered releases of half ranges nonsense with half assed launch drivers. That is NVidia's force and AMD's weakness, they need to stop having so many different GCN iterations on the market at the same time.
Oh and I hope that Navi isn't GCN, but I have a bad feeling about that one.
It would really make their driver development so much more efficient and potentially require less financial resources for driver development, and maybe more potential of having good day one drivers , they would have a full range of GPUs with the same software features, that would benefit from the same driver improvements at the same time.
None of this staggered releases of half ranges nonsense with half assed launch drivers. That is NVidia's force and AMD's weakness, they need to stop having so many different GCN iterations on the market at the same time.
Oh and I hope that Navi isn't GCN, but I have a bad feeling about that one.
Last edited:
Vega 20 is on 14nm + it's just going to be a refresh like polaris. Not sure why people are expecting it to be another year line up. Navi can certainly slot in 2k18 on schedule. Road map shows Vega beginning 2k17 to end of 2k17 and navi beginning 2k18 to end of 2k18.
Vega seems to be held back by glofo and yield rates. Might be why AMD jumped to TSMC for navi.
Vega seems to be held back by glofo and yield rates. Might be why AMD jumped to TSMC for navi.
Yay Angerybird his here to spread his wisdom. DP has pointed exactly nothing but what he thinks could happen, but AMD are ordering 7nm parts from TSMC. BIG contract already signed you see.
So respin 14nm plus for Vega at Global. That would segest Navi will very likely be the 7nm TSMC part.
Must be a good reason for the TSMC deal. Has to be either cost, volume or performance.
Last edited by a moderator:
I had the impression that the 1:2 FP64 will mostly appeal to the pro market rather than gamers. A higher clocked HBM or a 3rd stack wouldn't hurt either.
So far we have one unconfirmed report (which cites "industry sources") they are using TSMC and points back to the analyst's day comments Lisa Su made which only talks about the flexibility to use other foundries now not that they have definite plans to. But AFAIK they still have to pay GF some money for every wafer they have made elsewhere so its unlikely they would go elsewhere unless absolutely forced to.
Doing a bit of digging it appears the source is probably the same as http://www.expreview.com/43087.html which claimed Ryzen CPUs would be switching from GF to TSMC 16FF+.
Caporegime
Its from the 4th biggest newspaper in Taiwan:
https://en.wikipedia.org/wiki/China_Times
From the article:
http://www.chinatimes.com/newspapers/20170808000038-260202
Basically,the reason is there stated in the article,they are only going to TSMC due to supply constraints. So they don't think that GF can supply Ryzen CPUs,APUs and AMD GPUs in enough quantity.
AMD is fully committed to GF,as thet have to with WSA,so it looks like they are dual sourcing.
AMD paid a $100 million charge so they could make products outside of GF:
http://images.anandtech.com/doci/10631/WSA5.png
That was in late 2016. However,remember,AMD tends to launch GPUs almost at the end of the year,and Raven Ridge is only launching late this year. So its most likely very late 2018(maybe November to December),as I cannot see AMD replacing Vega and Raven Ridge in under a year.
There are rumours of a "Pascal refresh" - if AMD is pushing to get to 7NM,I certainly expect Nvidia will too. We will need to wait and see.
https://en.wikipedia.org/wiki/China_Times
From the article:
http://www.chinatimes.com/newspapers/20170808000038-260202
Basically,the reason is there stated in the article,they are only going to TSMC due to supply constraints. So they don't think that GF can supply Ryzen CPUs,APUs and AMD GPUs in enough quantity.
AMD is fully committed to GF,as thet have to with WSA,so it looks like they are dual sourcing.
AMD paid a $100 million charge so they could make products outside of GF:
http://images.anandtech.com/doci/10631/WSA5.png
That was in late 2016. However,remember,AMD tends to launch GPUs almost at the end of the year,and Raven Ridge is only launching late this year. So its most likely very late 2018(maybe November to December),as I cannot see AMD replacing Vega and Raven Ridge in under a year.
There are rumours of a "Pascal refresh" - if AMD is pushing to get to 7NM,I certainly expect Nvidia will too. We will need to wait and see.
Last edited:
Associate
here we go again 
Any rumours heard that Navi will be multi chip using infinity fabric like Threadripper?
My mate says it's an option he read somewhere, but I'm 99% sure it was the beer talking and he's full of it. I can't find anything online to say it is, or isn't.
My mate says it's an option he read somewhere, but I'm 99% sure it was the beer talking and he's full of it. I can't find anything online to say it is, or isn't.
There is reason to suggest AMD is planning on going with MCM type designs in the future with Navi but it probably won't be the first incarnation that we see out the door - the problems with current mGPU techniques like SLI and CF can't be completely solved with something like IF and require some significant architecture changes to allow for high speed, low latency interconnects right into the various parts of the chips and not just between the chips and things like DX12 explicit multi-adapter haven't really taken off yet in any way that would mean such a GPU was a success (and wouldn't really require IF as the intention with it is that the developer utilises it in a way they know will make the most use of the individual GPUs hence the explicit part of the name).
To properly break down a GPU from the current monolithic designs into an architecture that could make use of multiple packages on a substrate requires advances in technology that aren't going to be available in time for the likely range of release dates for the first iteration of Navi.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,026
I also doubt that Infinity Fabric has the speed or the low enough latency at present, might need an iteration or two before it's fit for that purpose.
Soldato
Vega 20 had a chance to be 7nm I thought, the 2nd release being significant over the 1st for refinement which I think Vega probably needs.
Navi I think is going to be H1 2019 and so really I expect to be looking for a good 3rd party version early 2020 personally. Similarly I think Volta wont be this year but mostly parallel to Vega 20 but I saw a video saying Volta 2017
Yeah IF currently is a very general purpose interconnect - the stuff nVidia was talking about that would make headless modules workable in an MCM scenario was very dedicated.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 31,026
And AMD's definition of H1 at that! We're almost 2 years away from seeing that imo, poss 20 months if lucky.