AMD Vega Architeture Speculation

One huge point is that AMD has a target to aim at...

Of course, they need to aim higher as 6 months after, nVidia will also start thinking of the next steps...
 
And in even more WTF?


http://techreport.com/news/30600/amd-takes-a-335m-one-time-charge-for-more-sourcing-flexibility



As AMD tells it, the new agreement does the following things:
Covers a 5-year period, spanning from calendar year 2016 through 2020;
Establishes a comprehensive framework for technology collaboration between AMD and GF for the 7nm technology node, building on the success of the 14nm node;
Provides AMD with the flexibility to manufacture certain products with another wafer foundry;
Sets annual wafer purchase targets from 2016 through the end of 2020, fixed wafer prices for 2016, and a framework for yearly wafer pricing.


In exchange for the new agreement, AMD is giving GlobalFoundries the following consideration:
Make a $100 million cash payment to GF, paid in installments beginning in Q4 2016 through Q3 2017.
Make quarterly payments to GF beginning in 2017 based on the volume of certain wafers purchased from another wafer foundry.
Grant to West Coast Hitech L.P., a wholly-owned subsidiary of the Mubadala Development Company PJSC, a warrant to purchase 75 million shares of AMD common stock at a purchase price of $5.98 per share. The warrant may be exercised in whole or in part prior to February 29, 2020. The warrant is only exercisable to the extent that Mubadala or its subsidiaries do not beneficially own, either directly or indirectly, an aggregate of more than 19.99 percent of AMD's outstanding capital stock after the exercise.
Click to expand...



And the tech report thought of the same thing at the end of the article as I was thinking:



so it might also be looking to establish a partnership with TSMC to use that foundry's 16-nm FinFET process. That technology has already been deployed to great effect in Nvidia's Pascal cards. We suppose we'll just have to wait for official word from AMD for more info about its sourcing strategy.
Click to expand...


The main reason for this is that the R480x is a nice card no doubt, and only uses 150 watts so the power consumption is reasonable, but when you compare with it with the GTX1080, the latter is only rated at a 180 watt TDP at stock clocks and significantly faster for the extra 30 watts it uses.


I am wondering if TSMC's 16nm process is inherently more power efficient when compared with Global's 14nm process?
 
Global Foundries seems like kind of a train wreck. I get the sense that AMD has been wanting to cut their ties with them for a while now, but they're stuck with them due to the agreement that was part of the spinoff.

That being said, I still don't think you should read too much into the power consumption figures with totally different architecture. I'd also like to see how Perf/Watt stacks up in DX12 games where GCN seems to do considerably better.
 
Nagorak said:
Global Foundries seems like kind of a train wreck. I get the sense that AMD has been wanting to cut their ties with them for a while now, but they're stuck with them due to the agreement that was part of the spinoff.

That being said, I still don't think you should read too much into the power consumption figures with totally different architecture. I'd also like to see how Perf/Watt stacks up in DX12 games where GCN seems to do considerably better.
Click to expand...


If it was just the architecture differences alone it wouldn't be hard to ignore it, but the chip is packing another 2.3 billion transistors more than Polaris 10 and running between 1.6 to 1.7 Ghz clock speeds.......About 500 Mhz faster than Polaris for those 30 watts of power.


Add the performance gap between both to put extra salt into the wound so to speak.......Easily well over 50% faster for the GTX1080.


All for that extra 30 watts.....:hmm:
 
I would think amd was going to use Samsung a lot more before TSMC

with TSMC making apple iPhone 7 chips is the reason I think nv took so long to get enough 1080 out the door and most likely part of the price increase also

can TSMC do vega also ?
 
bill dennison said:
I would think amd was going to use Samsung a lot more before TSMC

with TSMC making apple iPhone 7 chips is the reason I think nv took so long to get enough 1080 out the door and most likely part of the price increase also

can TSMC do vega also ?
Click to expand...


At least from a power consumption point of view it really gives the impression that TSMC's fab process really is better that global foundries, so there's at least that.


Physical size wise, the process is however a little bit larger than global's, so for any given transistor budget the end GPU is slightly larger which means there's less dies coming out of each wafer, even if the yeilds are perfect......The not so good part going to TSMC.



Samsung is another option and their process really is 14nm like global's, but got no clue how the power consumption is with them for something like higher end GPU's such as Vega.
 
shadow001 said:
At least from a power consumption point of view it really gives the impression that TSMC's fab process really is better that global foundries, so there's at least that.


Physical size wise, the process is however a little bit larger than global's, so for any given transistor budget the end GPU is slightly larger which means there's less dies coming out of each wafer, even if the yeilds are perfect......The not so good part going to TSMC.



Samsung is another option and their process really is 14nm like global's, but got no clue how the power consumption is with them for something like higher end GPU's such as Vega.
Click to expand...


as polaris was a pipe cleaner we really have no clue what vega will be on the same process

horse power wise I think vaga will be very close to a titan x pascal maybe 10% slower or hope 20% faster and I hope near half the price
given the time frame( most likely March ) I can't see them shooting low for the gtx 1080 little vega maybe but big vega should have no problem matching or beating titan x

power draw I don't care as long as it is under 350 watts per card
 
Last edited:
GloFlo does take some time to refine their process but the 28nm GloFlo process is about the best you can do. I expect Glo-Flo will improve dramatically over time and Vega should do better power to transistor count I do believe. Also if Polaris gets another revision for an update I think it will also improve significantly as well - at least a 100-200mhz bump that is. So AMD could have Vega launch and an updated Polaris next year followed by lower skews of Vega later in the year. It can really dramatically shift performance per price bracket around. I expect Nvidia to push up Volta as much as possible too, if they don't I think AMD will have a big window for 6 months or more.
 
bill dennison said:
as polaris was a pipe cleaner we really have no clue what vega will be on the same process

horse power wise I think vaga will be very close to a titan x pascal maybe 10% slower or hope 20% faster and I hope near half the price
given the time frame( most likely March ) I can't see them shooting low for the gtx 1080 little vega maybe but big vega should have no problem matching or beating titan x

power draw I don't care as long as it is under 350 watts per card
Click to expand...

We've all been here before bill with AMD and they've never lived up to the hype yet. Remember the Fury being an overclockers wet dream? It can't overclock for s***. I've given up making predictions about Vega's supposed performance because it'll just bite you in the ass and just heightens everyone's expectations. If any version of Vega is competitive with a 1080 and has HBM2 and is competitively priced then I'll get one. If not I'll just stick with my Fury and will most likely upgrade my mobo & cpu instead. Whether that's Zen we'll just have to wait and see!
 
noko said:
GloFlo does take some time to refine their process but the 28nm GloFlo process is about the best you can do. I expect Glo-Flo will improve dramatically over time and Vega should do better power to transistor count I do believe. Also if Polaris gets another revision for an update I think it will also improve significantly as well - at least a 100-200mhz bump that is. So AMD could have Vega launch and an updated Polaris next year followed by lower skews of Vega later in the year. It can really dramatically shift performance per price bracket around. I expect Nvidia to push up Volta as much as possible too, if they don't I think AMD will have a big window for 6 months or more.
Click to expand...



Time to market is everything in this business, and if TSMC got their 16nm process in better shape overall and especially when it comes to power consumption, and did so with a 6 ~ 9 month lead over Global being able to do the same, then it's obvious Nvidia is laughing all the way to the bank with the high prices on what still isn't their highest end card like the GTX1080.



Add that an even faster version of the Titan X can be released, since it is only 471mm^ so it's still smaller than big maxwell seen in the GTX980Ti, or big Fermi seen in the GTX780Ti series, and that as you mentioned Volta's release has been pushed up to mid next year, and whatever Vega is and even if it's faster than anything currently released right now ( including Titan X ), might not remain so for long.


Nvidia aren't sitting with their thumbs up their asses and waiting to see what AMD releases before reacting, that's for sure.
 
atomBomb said:
Wish GF won't fail them again. If it was TSMC I would have very high hopes for VEGA. Think Vega with GPU 2ghz clock + HBM2, now its probably 1200mhz like Polaris.
Click to expand...

I think your concern is valid, given GloFo's track record. Perhaps they've chosen to dual source like Apple did with their last round of iPhone SOCs (A9) which used both Samsung and TSMC.
 
atomBomb said:
Wish GF won't fail them again. If it was TSMC I would have very high hopes for VEGA. Think Vega with GPU 2ghz clock + HBM2, now its probably 1200mhz like Polaris.
Click to expand...

Engineering samples of Zen from GF are at 3ghz, when launched that will most likely be higher. GF are making those too.

Design of Polaris also affects max clock speed, limiting component. Vega could go a hell a lot faster but I don't think AMD concentrated on maxing out speed like Nvidia.

So guessing, prediction based on little to nothing ;), Vega will be a 1350mhz chip, 4096 shader part. Will it have 8gb or 16gb is the question? Two 4hi stacks of HBM2 for 8gb and ~500gb/s bandwidth using the fast HBM2 or 16gb of the slow HBM2 for ~760gb/s using four 4hi stacks? HBM2 can be stacked 2hi, 4hi and 8hi. You could have one stack or two (not sure about three) and four stacks.

Could be the fastest Vega will be 16gb and the slower one 8gb. Kinda fun waiting on how this will pan.
 
noko said:
Engineering samples of Zen from GF are at 3ghz, when launched that will most likely be higher. GF are making those too.

Design of Polaris also affects max clock speed, limiting component. Vega could go a hell a lot faster but I don't think AMD concentrated on maxing out speed like Nvidia.
Click to expand...

Do you mean Polaris' design affects its clockspeed or do you mean APUs that feature Polaris IP have a clockspeed limitation?
 
shadow001 said:
Time to market is everything in this business, and if TSMC got their 16nm process in better shape overall and especially when it comes to power consumption, and did so with a 6 ~ 9 month lead over Global being able to do the same, then it's obvious Nvidia is laughing all the way to the bank with the high prices on what still isn't their highest end card like the GTX1080.



Add that an even faster version of the Titan X can be released, since it is only 471mm^ so it's still smaller than big maxwell seen in the GTX980Ti, or big Fermi seen in the GTX780Ti series, and that as you mentioned Volta's release has been pushed up to mid next year, and whatever Vega is and even if it's faster than anything currently released right now ( including Titan X ), might not remain so for long.


Nvidia aren't sitting with their thumbs up their asses and waiting to see what AMD releases before reacting, that's for sure.
Click to expand...

So what you're really saying is your happy to pay exorbitant prices for mid-range cards :lol:? Nvidia are milking you to the hilt and if you're happy to be shafted that's great. We're all different otherwise it would be a bit boring!
 
LordHawkwind said:
So what you're really saying is your happy to pay exorbitant prices for mid-range cards :lol:? Nvidia are milking you to the hilt and if you're happy to be shafted that's great. We're all different otherwise it would be a bit boring!
Click to expand...



I didn't buy the mid range cards to begin with, but the Titan-X's where the GPU's are packing 12 billion transistors and pack 12 GB of ram and 1/2 a terabyte of memory bandwidth........Doesn't sound midrange to me, and do show me what AMD has to compete with this mid range card?
 
You must log in or register to reply here.
Back
Top