SK Hynix 8GHz 3D stackable GDDR5 memory
Click here to post a comment for SK Hynix 8GHz 3D stackable GDDR5 memory on our message forum
icedman
Im very excited to see what AMD brings to the table this time around. Would like to see this used as edram for the apus one day too would kick a$$ if they could stick 1 4gb chip merged in the die.
jimdove
does this mean this time next year the new GPUs will come with hideous amounts of Vram?
ricardonuno1980
I'm very interested 3D VRAM GDDR5 with HBM!! 😀 But... I'm congrats to AMD-ATi. 🙂 4096-bit bus!?!! :eek2: 😀
GTX 980/970 have the bit too low bandwidth memory because currently NVIDIA hasn't said 3D VRAM! 😕 :thumbdown
Lane
Or TOP PIM "Troughput-oriented programmable processing in memory "http://www.dongpingzhang.com/wordpress/wp-content/uploads/2012/04/TOP-PIM_v6_public.pdf
fantaskarsef
I'm curious: 4096bit BUS, does it mean internal bus, as in G-RAM to GPU, or does it mean this thing could service a 4096bit BUS over PCIe?
I'd want to know how that works out in regard of PCIe lanes, maybe this card will be the first to really gain something above 8 lanes PCIe 3.0 ( =16 lanes PCIe 2.0).
Anybody got a clue or guess?
Ukdude21
Presumably Nvidia will not benefit from this technology any time soon as Amd partnered with Hynix. When Nvidia develops any form of advancement for their GPU's they do not share the technology with Amd and I think Amd should let Nvidia suffer the same now.
If amd holds the exclusive for this technology it will leave Nvidia playing Catch up for a while.
Denial
PASCAL is already slated to use 3D stacked RAM. So yeah.
-Tj-
Its a 3d stack thing, each 1gb stack has 1024bit bus x4 = 4096bit bus and 4gb vram, also each stack has 128gb/s bandwidth x4 = ~512gb/s+
fantaskarsef
If those 512 gb/s mean gigabit/second, that should be less than e.g. a single PCIe 3.0 lane with over 7 terrabits/s? Or am I making a fool off myself? :bonk:
-Tj-
No it is Gigabytes/s 😉
In one leak it said up to 640gb/s
fantaskarsef
Wow, that still beats 15+ GB/s (PCIe 3.0 x16, if wikipedia is right).
Ukdude21
All I am going to say...
All I am going to say is read number 7 on the list of this website. "http://streamcomputing.eu/blog/2014-08-05/7-things-nvidia-doesnt-want-know/"
Noisiv
7 things NVIDIA doesn’t want you to know
2. Profit per GPU is huge
Nice.
Truly a cuckoo blog 🙂
Ukdude21
Yeah obviously a cuckoo blog lol... not!
For a moment there I thought I had a reply from Denial after reading number 7 on the blog I posted but guess not lol
Noisiv
shhh... Profit per GPU is huge... don't ask don't tell
https://www.youtube.com/watch?v=KuZBqicI-I8
Ukdude21
Haha XD
siriq
Nvidia will see it earlier than you think. http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers
Denial
What does number 7 on the blog you posted have anything to do with what I said about 3D stacked ram? You mean number 6?
This blog is garbage. His first point is wrong, he even admits he's wrong. The rest are obvious "per profit GPU is huge" geewiz, who would have thought that a company who spends the majority of it's revenue on R&D is going to have huge profit margins on it's products. AMD & Hynix created 3D stacked memory? Really? 3D packaging has been around for at least a decade, it was outlined in my digital signals class at RIT in 2007. Samsung already produces 3D DD4 modules. Yeah maybe both companies worked together to finalize the HBM standard. But wtf does that have to do with Nvidia? Should everytime AMD/Nvidia use OpenCL they mention Apple as they were one the the biggest contributors?
And even point 7 is wrong and dumb and the author is an idiot. They outlined virtual unified address space in 2011.
http://on-demand.gputechconf.com/gtc-express/2011/presentations/cuda_webinars_GPUDirect_uva.pdf
God forbid they left out a word in a slide and he cried a river about it.
Like look I hate propriety tech as much as anyone else. I also hate Nvidia's marketing department. That being said advocating for AMD to do exactly what Nvidia is not only hypocritical but its' a bull**** way to produce a standard.
Ukdude21
Wow calm down... what a lengthy reply -.- Insulting the author of that blog is not necessary. Yes you are right on certain things e.g. I do hate propriety tech and I hate how NVidia operates in terms of marketing. How will NVidia learn? if they have access to all or any tech Amd develops which you have a refuted about 3d stack memory... Up until now NVidia would not of wanted any tech that Amd specifically had anything to do with.
Sometimes the best medicine is your own and that is why I was advocating Amd to practice the same methods as Nvidia.
It is obvious though that Amd has been collaborating very closely with Hynix in regards to developing this tech.
Hugh profits hurts the consumer wallet I think that is why that is mentioned...
Sorry your right point 6 on that blog... I wonder what 3d stacked memory will be used by NVidia? Hynix I honestly don't know but I am interested.
Noisiv
Please,,, it's painful hearing something stupid as this from apparently normal guy.
There are no huuuuge profits. They net income ~100-200M on $1.2B of revenue.
They do all right, and yet they are #1, creme de la creme, absolutely the best in the world at what they do.
You think price should be BoM + small fee?
They should f***ing subsidize it so you can buy baddest meanest GPU, as if your life depends on it?
GPU is not water, you DON'T need one.
Or NV should offer their GPUs lower than what people are willing to pay?
Dafug... I don't hear many sub-saharan gamers crying over supposed high GPU prices.
It's always some 1st/2nd world crybaby who has to have the absolute best, and is willing to abandon as simple and universally accepted concept as free trade in getting there.
Ask Intel about huuuuge profit margin. You get a piece of silicon, no PCB, no heatsink, no GBs of VRAM for $200-1000.
Ask anyone in the industry, even financially semi-competent and money losing AMD, ask them about R&D needed to come up with this silicon
