Samsung begins mass production of 4GB HBM2 stacks

[zMeul]

source: http://news.samsung.com/global/samsung-begins-mass-producing-worlds-fastest-dram-based-on-newest-high-bandwidth-memory-hbm-interface

  

Samsung Electronics announced that it has begun mass producing the industry’s first 4-gigabyte (GB) DRAM package based on the second-generation High Bandwidth Memory (HBM2) interface, for use in high performance computing (HPC), advanced graphics and network systems, as well as enterprise servers. Samsung’s new HBM solution will offer unprecedented DRAM performance – more than seven times faster than the current DRAM performance limit, allowing faster responsiveness for high-end computing tasks including parallel computing, graphics rendering and machine learning.

 

“By mass producing next-generation HBM2 DRAM, we can contribute much more to the rapid adoption of next-generation HPC systems by global IT companies,” said Sewon Chun, senior vice president, Memory Marketing, Samsung Electronics. “Also, in using our 3D memory technology here, we can more proactively cope with the multifaceted needs of global IT, while at the same time strengthening the foundation for future growth of the DRAM market.”

 

the 4GB stack is manufactured in 20nm and was achieved by stacking four 8 gigabit cores on top of a buffer die; they are vertically interconnected by TSV (Through Silicon Vias) holes and microbumps - over 5000 per single 8gb die

 

 

Samsung's DRM package offers 256GBps of bandwidth - twice as much as HBM1 and ~7 times more than a single 4Gb GDDR5 chip

 

Samsung also plans to introduce 8GB HBM2 stacks, specifically to be used in GPU production, during this year - no concrete date

 

---

 

if nVidia was planning to source 8GB HBM2 from Samsung, looks like they'll have to wait a bit longer

but this doesn't mean they can't use multiple 4GB stacks - depending on how exactly they engineered the memory controller inside the GPU

[ozziestig]

Well it looks like the next generation of GPUs are nearly in full production.

[don_svetlio]

Soooo, Nvidia 32GB TItan not coming?

[KamiKatze]

I'll not buy a GPU until both teams have released their full next lineup, no matter how good the technology of the first cards will be.

[marldorthegreat]

Would it not be pretty cool if top end zen cpu came with 4/8 gbs of ram ON BOARD. OR mayber the zen apus.

 

I'll not buy a GPU until both teams have released their full next lineup, no matter how good the technology of the first cards will be.

 

I see your logic, but im going for amd. My plan is 2 nvidia gpus, 2 amd, 2 nvidia etc.

[Prysin]

does this mean, we maybe can get 4GB HBM2 AMD APUs??? (since they werent specifically aiming these for GPUs, unlike the 8GB versions)

if so... hell yes!!!

[marldorthegreat]

Soooo, Nvidia 32GB TItan not coming?

Probably wont be 32gb, most likely 16 gb. (4x4gb)

[Lethal Seraph]

Oh yes! Come to papa.

[patrickjp93]

Soooo, Nvidia 32GB TItan not coming?

They have 8GB stacks coming too.

[Bouzoo]

Soooo, Nvidia 32GB TItan not coming?

 

I'm quite sure they are coming later on, same with GM204 and 200. 

[Doobeedoo]

Awesome. Can't wait to see how much top cards will have VRAM and also maybe APU with ton of it ?

[iTonyy]

Nice. Can't wait for Pascal

[TheRandomness]

Slight typo in the op, HBM misspelt as HMB. Fixd

[Notional]

does this mean, we maybe can get 4GB HBM2 AMD APUs??? (since they werent specifically aiming these for GPUs, unlike the 8GB versions)

if so... hell yes!!!

 

Well afaik, HK Hynix should have 8GB stacks. SO AMD can pretty much do what they want. NVidia, maybe not so much.

[dalekphalm]

@zMeul, NVIDIA could certainly still use 8GB of HBM2 with this - remember, AMD used 4x 1GB HBM 1.0 stacks. So if Samsung has 4GB stacks ready to use, that means you could do up to 4x 4GB = 16GB HBM2 onboard. Assuming NVIDIA is capable of producing similar technology as AMD, which we have no reason to doubt yet.

 

Probably wont be 32gb, most likely 16 gb. (4x4gb)

That was my thought as well - and there's no reason that they won't have 8GB stacks ready later this year, so we could still see 32GB HBM2 GPU's (Tesla's/FirePro's) at the tail end of the year.

[zMeul]

@zMeul, NVIDIA could certainly still use 8GB of HBM2 with this - remember, AMD used 4x 1GB HBM 1.0 stacks. So if Samsung has 4GB stacks ready to use, that means you could do up to 4x 4GB = 16GB HBM2 onboard. Assuming NVIDIA is capable of producing similar technology as AMD, which we have no reason to doubt yet.

I know, this is why I said they could use multiple 4GB stacks

the weird part of the 8GB stacks is that Samsung were adamant them being used in GPUs

 

In addition, Samsung plans to produce an 8GB HBM2 DRAM package within this year. By specifying 8GB HBM2 DRAM in graphics cards, designers will be able to enjoy a space savings of more than 95 percent, compared to using GDDR5 DRAM, offering more optimal solutions for compact devices that require high-level graphics computing capabilities.

[Starelementpoke]

does this mean, we maybe can get 4GB HBM2 AMD APUs??? (since they werent specifically aiming these for GPUs, unlike the 8GB versions)

if so... hell yes!!!

That'd be interesting to see.

[dalekphalm]

I know, this is why I said they could use multiple 4GB stacks

the weird part of the 8GB stacks is that Samsung were adamant them being used in GPUs

 

Well I have no doubt we'll see both FirePro and Quadro workstation GPU's that use 32GB of HBM2 - I would think that's pretty much a given at this point.

[beach_boy98]

So now we need the core technology to be able to take advantage as well as the api's

[azurezeed]

Yes bring the new technology, my body is ready (tho my wallet isn't)

[Steven9595]

I'll not buy a GPU until both teams have released their full next lineup, no matter how good the technology of the first cards will be.

I'm hoping the jump is big enough to justify me upgrading from a 970 :c 

[spartaman64]

i wonder if this would be somewhat awkward for nvidia as they were in a big legal battle with samsung but then again samsung does make a lot of parts for the iphone so they would probably be fine