It is official AMD skips 20nm and jumps to 14nm

Published by

Click here to post a comment for It is official AMD skips 20nm and jumps to 14nm on our message forum
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
Come on your on 7970 upgrade to 390x and you get huge leap in performance. At some point you can safely upgrade in your case it seems a good deal 7970 to 390x is huge step.
That's the plan, I've just set 2x performance rule for simple reason. I am quite happy with 7970 performance on 1080p screen. 1440p screen has 1.78x more pixels, so To move there I want double performance to gain few fps in process too (some reserve).
https://forums.guru3d.com/data/avatars/m/242/242371.jpg
I agree vg24a3. I would buy an AMD card in a shot to help them out but only if performance was there. I don't understand these ridiculous people that are brand centric.
I'm the same. The last three GPUs I've had were Nvidia, but the three before that were ATI/AMD. I always go with what's best for me in general terms, and at the moment it's Nvidia, but if AMD bring out a card I like ill jump on it. There's no point being loyal to X company.
https://forums.guru3d.com/data/avatars/m/235/235344.jpg
Puts the whole limited supply at launch for the 300 series in perspective.
https://forums.guru3d.com/data/avatars/m/116/116345.jpg
That's the plan, I've just set 2x performance rule for simple reason. I am quite happy with 7970 performance on 1080p screen. 1440p screen has 1.78x more pixels, so To move there I want double performance to gain few fps in process too (some reserve).
You mean 1.33x, how did you get that number 1.78x lol?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Puts the whole limited supply at launch for the 300 series in perspective.
Why?
data/avatar/default/avatar27.webp
you mean 1.33x, how did you get that number 1.78x lol?
(2560*1440) / (1920*1080) = ~1.78.
https://forums.guru3d.com/data/avatars/m/56/56686.jpg
wait if there 20nm are massive fail how is switching to 14nm gona be any less of problem?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
wait if there 20nm are massive fail how is switching to 14nm gona be any less of problem?
Because 20nm is planar, 14nm is multigate.
https://forums.guru3d.com/data/avatars/m/55/55855.jpg
Be out before the 390s at this rate :P
https://forums.guru3d.com/data/avatars/m/235/235344.jpg
Why?
If the yields at 20nm are as terrible as made out to be, and the 300 series being at 20nm, would it not be logical to conclude that the supply at launch would be less than was hoped for?
https://forums.guru3d.com/data/avatars/m/56/56686.jpg
Because 20nm is planar, 14nm is multigate.
i dont see that matter much, it still getting smaller and small things are the easier problem will happen.
https://forums.guru3d.com/data/avatars/m/231/231640.jpg
Well considering they are going back to a x86 architecture I am really looking forward to this. If the new CPU's can perform like the Phenom2's did when they came out then AMD will be back into the performance competition. But I just hope they make CPU's and not APU's only.
https://forums.guru3d.com/data/avatars/m/128/128096.jpg
Well considering they are going back to a x86 architecture I am really looking forward to this. If the new CPU's can perform like the Phenom2's did when they came out then AMD will be back into the performance competition. But I just hope they make CPU's and not APU's only.
They never left x86 besides trying to venture into some ARM combinations.
data/avatar/default/avatar10.webp
wait if there 20nm are massive fail how is switching to 14nm gona be any less of problem?
Because the 20nm nodes in question was from TSMC; and the 14nm is from Samsung / Global Fundries ... TSMC have just fail again to deliver...so everyone is switching to Samsung Fundry for the 14nm.. Qualcomm have break his contract with TSMC for 16nm.. Apple same, Nvidia it seems too, and AMD was surely allready have do the choice to move to Glofo/Samsung ( who are partner and use the same fabrication and node process )
https://forums.guru3d.com/data/avatars/m/238/238382.jpg
Will there ever be new AM3 Phenoms? or FX's? What is the roadmap for AMD CPU?
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
Will there ever be new AM3 Phenoms? or FX's? What is the roadmap for AMD CPU?
There is no longer development on AM3, FM2+ will get something of minor importance to customers. (I hope they'll push something with full HSA 1.0, so people can develop and learn on it properly before Zen comes. Otherwise we'll have only HSA 1.0 in notebooks.) Zen plans are quite grandiose and transistor hungry, but it should be OK since it was from start planned to be 14nm FF. If you want to know what to expect from Zen: - Improved IPC over FX CPUs/APUs by at least 20% - Power efficiency improvement from 14nm and that additional Voltage fluctuation control introduced with Carrizo (We should see how much more can carrizo do over kaveri FX-8800p vs FX-7600p ~ both 4 cores and 512SP) - Because it will be big expensive core anyway, we can expect ddr4/HBM controller (potentially low latency which is one of weak points of AMD CPUs/APUs), rumor says quad channel ddr4 and HBM at same time - So 100% new socket I expect it to match intel in total performance, I am not sure about power efficiency with normal non HSA applications as AMD invests to many technologies around and unless they are used it's just dead weight.
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
If the yields at 20nm are as terrible as made out to be, and the 300 series being at 20nm, would it not be logical to conclude that the supply at launch would be less than was hoped for?
The 300 series is rumored to be 28nm, not 20. The supply for the 390 is going to be worse because the yields for the interposer are going to be low.
i dont see that matter much, it still getting smaller and small things are the easier problem will happen.
Uh it matters a whole bunch. It's a completely different transistor design. There is no manufacturing issue with 20nm, the issue is with the performance scaling. There's so much gate leakage on 20nm planar that it essentially renders the power scaling from the shorter length useless. By raising the channel above the substrate they effectively increase the the volume channel and have better control of voltage and gain additional switching speed.
Because the 20nm nodes in question was from TSMC; and the 14nm is from Samsung / Global Fundries ... TSMC have just fail again to deliver...so everyone is switching to Samsung Fundry for the 14nm.. Qualcomm have break his contract with TSMC for 16nm.. Apple same, Nvidia it seems too, and AMD was surely allready have do the choice to move to Glofo/Samsung ( who are partner and use the same fabrication and node process )
TSMC's 20nm node was never intended for GPU production. They essentially built it for Apple. 16nmFF is also for phones, the only GPU ready process they have coming up is 16nm FF+, which will be in mass production in November. The difference between 16nm FF+ and Samsung's 14nm FF is so minor it's not even worth mentioning. The interconnects on both are 20nm and they both use 78nm gate pitch, 64nm metal pitch. Samsung should be able to get slightly better power/switching performance but in terms of density they are exactly the same.
https://forums.guru3d.com/data/avatars/m/128/128096.jpg
Again, performance is rather secondary issue today, chips have been "fast enough" for most people since the turn of the decade. The issue is that they need to get a competitive mobile CPU and chipset already in volume. They can have a faster and competitive otherwise chip with Intel on the desktop and it still won't make them the money they need. http://cdn.inquisitr.com/wp-content/uploads/2010/06/Forrester-Tablet-540x329.jpeg The desktop market is flat if not outright declining, the laptop market continues to grow. Apple is by the far the most successful here, and AMD has no presence on the CPU side of the equation (which is in every single unit, unlike the GPU's). As far as I see they have to choices, try to compete in mobile or try to compete in servers. Still heavily investing in desktops and trying to fight Intel for an ever declining market share and slimmer margins is not going to save them.
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
They need efficiency. Simple as that, they need an all in one package that can do it all with reasonable price, performance, power consumption, and availability. AMD is trying with their APU's with apps that are both CPU and GPU accelerated at least, and Intel is doing better in their iGPU section as well now. Neither one are perfect right now, but Intel is showing more promise with the performance and power consumption with the Core M. Issue with Core M is availability and price.
https://forums.guru3d.com/data/avatars/m/248/248627.jpg
The only reason apu's fail in laptops is because no one gives them a chance. I would take a cheaper a10 with dual graphics over an i7 based laptop. I7 will win in cpu but no way in hell will it keep up graphicaly which i find most laptops lacking.