• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

AMD and Hynix announce joint development of HBM memory stacks

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
sefsefsefsef said:
The stacked chips are bonded together, so they are like one thick chip, so heat goes through that chip just like it does any other chip. To get the power map of a stacked chip, you just sum up the power maps of all the individual layers-the thickness of each layer is so small that even when they're stacked on top of each other it's like they're one layer from a conventional chip. Bryan said in his talk that thermals are absolutely not a concern for 3D stacking.

It is not stacked DDRx. DDRx refers to the data rate of the *IO pins,* so what goes on inside the DRAM chip doesn't matter as long as it can keep up with the IO rate. HBM will have a layout and organization that will allow it to achieve 1TB/s of throughput (as per the slide above 1024 IOs x 1GHz each), compared to the 1-2 GB/s that a DDR3 chip provides.

HBM is a 2.5D technology, where there is a stack of memory *next to* the chip, connected by a silicon interposer. But in general, the various layers of a 3D stack do not need to be the same size. Smaller layers can go on top of larger layers.
Click to expand...


Excellent info, thanks...! I guess AMD, Intel, and Nvidia are going to have to build some better memory controllers. 😀
 
VirtualLarry said:
Will APUs get this technology before discrete cards? If so, it could cause quite an upheaval in the GPU market.
Click to expand...
I would almost guarantee a "no." APUs don't have the margins to absorb that cost.
VirtualLarry said:
Will APUs get this technology before discrete cards? If so, it could cause quite an upheaval in the GPU market.
Click to expand...
I would almost guarantee a "no." APUs don't have the margins to absorb that cost.
inf64 said:
In theory yes. In reality unlikely 🙁. It would make Carizzo's iGPU probably the fastest on the market and would erase all the lower end discrete GPUs from competition. Add a year or two on anything they claim and you will get approx. time frame for products.
Click to expand...
(Bad) Automobile analogy incoming:

Any car is faster than its competition if you strap a rocket to it. But rockets are expensive.

Same here. You could technically take off-package system RAM out of the equation, but for the time being, it's going to be quite expensive. APUs getting stacked memory is a matter of time, but I really don't think AMD will have it on their APU lineup until the generation after Carrizo at the earliest.

I'm fairly confident they'll have stacked memory before Nvidia though, and possibly before Intel's 14nm Xeon Phi.
 
Last edited:
Bryan Black started his talk by saying that Intel (where he used to work) shot down 3D stacking 10 years ago because it could add "10-15 cents per chip," but that AMD was excited enough about it that they would do it even if it added "200 dollars" per chip. He said it looks like when it finally arrives that it will add 2 or 3 dollars per chip.

Now, if this 2-3 dollars is including the price of the HBM, or just an additional integration cost on top of the price of the DRAM dice, I'm not sure. But my guess is that integrating HBM with an APU would cost AMD anywhere from $2 to a maximum of $20 per chip, which is totally reasonable for an APU.
 
sefsefsefsef said:
Bryan Black started his talk by saying that Intel (where he used to work) shot down 3D stacking 10 years ago because it could add "10-15 cents per chip," but that AMD was excited enough about it that they would do it even if it added "200 dollars" per chip. He said it looks like when it finally arrives that it will add 2 or 3 dollars per chip.

Now, if this 2-3 dollars is including the price of the HBM, or just an additional integration cost on top of the price of the DRAM dice, I'm not sure. But my guess is that integrating HBM with an APU would cost AMD anywhere from $2 to a maximum of $20 per chip, which is totally reasonable for an APU.
Click to expand...
And exactly how much would they be putting on there? There's no way they'd be able to do 4 or 8GB for $20. 2GB, maybe. But $2-3? We're talking about tiny amounts of RAM... I'd imagine performance would be rather limited in such a configuration.
 
Homeles said:
And exactly how much would they be putting on there? There's no way they'd be able to do 4 or 8GB for $20. 2GB, maybe. But $2-3? We're talking about tiny amounts of RAM... I'd imagine performance would be rather limited in such a configuration.
Click to expand...

That's a good point. Right now, a 4Gb chip of DDR3 is about $4, so getting 4GB worth of DDR3 chips would cost at least $32. This doesn't translate directly to costs in 3D packaged memories, though, and as soon as 3D memories are common I'm sure we'll be at least one DRAM node into the future by then, so prices will be lower. I wouldn't be surprised if APUs with 8GB of HBM cost $20-$30 more than their non-HBM counterparts when they debut. Eh, maybe $50 more, but 8GB is enough capacity to be the entire main memory, so you're saving cost elsewhere in the system.

One thing is for sure, though, that having 1TB/s of DRAM bandwidth will be excellent for the future of APUs.
 
I can't recall the source but HBM will cost $20 for 1 GB initially. Even that much would be really good for an APU.
 
Given that AMD also makes discrete graphics cards with high margins I'm sure we can expect stacked tech to show up there very early on into its lifecycle. Probably much sooner than APUs. It's a perfect match, since they already have to include memory on the graphics board itself and graphics applications need literally as much bandwidth as possible
 
sefsefsefsef said:
Bryan Black started his talk by saying that Intel (where he used to work) shot down 3D stacking 10 years ago because it could add "10-15 cents per chip," but that AMD was excited enough about it that they would do it even if it added "200 dollars" per chip. He said it looks like when it finally arrives that it will add 2 or 3 dollars per chip.

Now, if this 2-3 dollars is including the price of the HBM, or just an additional integration cost on top of the price of the DRAM dice, I'm not sure. But my guess is that integrating HBM with an APU would cost AMD anywhere from $2 to a maximum of $20 per chip, which is totally reasonable for an APU.
Click to expand...
That 2-3 dollars is propably diffrent / better / bigger memory controler and connectors between chip and stacked memory on board and all other changes that will allow CPU/APU to use such high bandwidth and take advantage of it.

Still it cannot come soon enough. This is kind of tech that can really make a diffrence. Of course it's price for end-user will be absolutely crucial in how fast it'll be adopted. I just hope production costs and volume in comparision with 'normal' DRAM will be reasonable, so that adoption can be fast once it hit the market.
 
Last edited:
bronxzv said:
where was it announced ?

I can't find it mentioned here
http://newsroom.intel.com/community...-next-generation-intelr-xeon-phi-tm-processor

the memory modules in slide 7 of the PDF look very thin

do you have a link to new information ?
Click to expand...

wpid-Knights-Landings-MCDRAM-660x660.jpg

Xeon-Phi-Knights-Landing-Architecture.png


MCDRAM=Memory Cube DRAM.
 
Last edited:
ehume said:
200W TDP. I don't see how to escape water cooling for that.
Click to expand...
Keep in mind, graphics cards have even higher TDPs. Since this is a PCI-E card, there's no issue with air cooling here.
sefsefsefsef said:
That's a good point. Right now, a 4Gb chip of DDR3 is about $4, so getting 4GB worth of DDR3 chips would cost at least $32. This doesn't translate directly to costs in 3D packaged memories, though, and as soon as 3D memories are common I'm sure we'll be at least one DRAM node into the future by then, so prices will be lower. I wouldn't be surprised if APUs with 8GB of HBM cost $20-$30 more than their non-HBM counterparts when they debut. Eh, maybe $50 more, but 8GB is enough capacity to be the entire main memory, so you're saving cost elsewhere in the system.

One thing is for sure, though, that having 1TB/s of DRAM bandwidth will be excellent for the future of APUs.
Click to expand...
Yeah, it'll be a quite disruptive technology. The sooner, the better.
 
ehume said:
200W TDP. I don't see how to escape water cooling for that.
Click to expand...


I'm confident I could of used an air cooler with my FX9370 if I left the clocks and voltage alone. With that being said, I went right to water, but people do occasionally use high end air coolers with these CPU's. They are 220 watt TDP.
 
If they make a 20nm gpu with 8 billion transistors and 8GB of 1 TB/s RAM, they better add 4 x86 cores to it. Then all they need to do is add a few usb ports and a couple sata ports and the video card becomes a complete system.
 
Homeles said:
Oh really? That's interesting.
Click to expand...

Yeah, a rack full of nothing but Phis is going to be pretty awesome from a compute/m^3 point of view. And there is a version with the network fabric integrated onto the package. It's going to be a HPC beast.

Xeon-Phi-Knights-Landing-Integrated-Fabric.png
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top