JEDEC Updates Groundbreaking High Bandwidth Memory (HBM) Standard

[btarunr]

JEDEC Solid State Technology Association, the global leader in the development of standards for the microelectronics industry, today announced the publication of an update to JESD235 High Bandwidth Memory (HBM) DRAM standard. HBM DRAM is used in Graphics, High Performance Computing, Server, Networking and Client applications where peak bandwidth, bandwidth per watt, and capacity per area are valued metrics to a solution's success in the market. The standard was developed and updated with support from leading GPU and CPU developers to extend the system bandwidth growth curve beyond levels supported by traditional discrete packaged memory. JESD235A is available for free download from the JEDEC website.

JESD235A leverages Wide I/O and TSV technologies to support up to 8 GB per device at speeds up to 256 GB/s. This bandwidth is delivered across a 1024-bit wide device interface that is divided into 8 independent channels on each DRAM stack. The standard supports 2-high, 4-high and 8-high TSV stacks of DRAM at full bandwidth to allow systems flexibility on capacity requirements from 1 GB - 8 GB per stack.

Additional improvements in the recent update include a new pseudo channel architecture to improve effective bandwidth, and clarifications and enhancements to the test features. JESD235A also defines a new feature to alert controllers when DRAM temperatures have exceeded a level considered acceptable for reliable operation so that the controller can take appropriate steps to return the system to normal operation.

"GPUs and CPUs continue to drive demand for more memory bandwidth and capacity, amid increasing display resolutions and the growth in computing datasets. HBM provides a compelling solution to reduce the IO power and memory footprint for our most demanding applications," said Barry Wagner, JEDEC HBM Task Group Chairman.

View at TechPowerUp Main Site

 

[TheGuruStud]

I better never see crap DDRx, again. Cancel this ddr4 bullshit lol

 

[thebluebumblebee]

Come on AMD. Surprise the crap out of everyone. Bring out ZEN with 2-4GB HBM L4! BAM.

 

[Aquinus]

Come on AMD. Surprise the crap out of everyone. Bring out ZEN with 2-4GB HBM L4! BAM.

Screw that. Four stacks of 4 or 8GB would make for an amazing SoC that doesn't require external DRAM. Cache is already nutty fast and makes up for slower DRAM pretty well (hence why memory speeds north of 1600Mhz on DDR3 don't really have much impact.) HBM most likely will only serve to reduce the footprint of current devices.

I better never see crap DDRx, again. Cancel this ddr4 bullshit lol

DDR4 spec includes support for DRAM stacking. DDR4 can and will far outpace HBM when it comes to capacity. Personally, I think HBM is a memory level somewhere between last level cache and system memory. It's not like HBM can be a replacement in every circumstance but, could in many.

 

[mtcn77]

Come on AMD. Surprise the crap out of everyone. Bring out ZEN with 2-4GB HBM L4! BAM.

They kind of announced Zen will be "1 GB HBM."

 

[Legacy-ZA]

Well, all I can say is, that GPU's this year better have 8GB of ram, minimum, so it can be future proof. As in, lowest tier cards, being 8GB and higher end cards from 16GB to 32GB.

 

[happita]

HBM needs time to mature in order to be a viable replacement to comparable GDDR5 in terms of cost. We saw how it was priced with AMD's Fury line of cards. This time around since both camps are going to be implementing the newer HBM2 standard, I don't expect this next-gen will be any different. This of course doesn't mask the fact that forward progress is being made and I'm all for it. Barriers will be broke gentlemen.

 

[Mysteoa]

Well, all I can say is, that GPU's this year better have 8GB of ram, minimum, so it can be future proof. As in, lowest tier cards, being 8GB and higher end cards from 16GB to 32GB.

By the time you need 8gb for regular gaming the card will not be able to run games above 30 fps. There is no point in future proofing. Not to mention the consoles will delay that.

 

[Aquinus]

By the time you need 8gb for regular gaming the card will not be able to run games above 30 fps. There is no point in future proofing. Not to mention the consoles will delay that.

It's not about using 8GB IMHO, it's about using more than 4GB because you lose performance the moment you start using more VRAM than you actually have.

 

[ZoneDymo]

It's not about using 8GB IMHO, it's about using more than 4GB because you lose performance the moment you start using more VRAM than you actually have.

Not always the case, take GTA5, you can set the Vram usage higher then you actually have and run into no issues, if you go way overboard you still dont lose performance, you just have stuff popping in.

 

[Steve-007UK]

Not always the case, take GTA5, you can set the Vram usage higher then you actually have and run into no issues, if you go way overboard you still dont lose performance, you just have stuff popping in.

Play gta 5 long enough while over the ram limit and it WILL eventualy crap out seen it first hand when i pushed my HD7970 over the 3gig limit.

Game ran fine for about an hour then bogged right down till it felt like 1FPS.

 

[Aquinus]

Not always the case, take GTA5, you can set the Vram usage higher then you actually have and run into no issues, if you go way overboard you still dont lose performance, you just have stuff popping in.

I used to use 1.6GB of VRAM on my 6870s (1 GB GPUs,) before I replaced them with the 390. Needless to say, you start taking a performance hit to use more VRAM than you have. The question is, does that performance hit prevent you from maintaining 60 FPS? For me in a lot of cases that was a "no" but, I would also never see full GPU utilization on my 6870s as a result. There comes a point where the GPU has to wait for data longer than it has to actually do the computation so if the set of data being worked on can't fully reside in VRAM, a lot of swapping is going to occur between main memory and the GPU which always harms performance but may not impact your experience, at least, not right away.

 

[Steevo]

It's not about using 8GB IMHO, it's about using more than 4GB because you lose performance the moment you start using more VRAM than you actually have.

GTA4 and its desire for more than 152MB of vmem, even users with 700MB were suffering, while 1GB users were fine, it prevented a lot of the issues people complained about. So yeah, i see 8GB becoming the norm for high end video cards, it will allow for better high end textures that require less filtering and thus less GPU power to render, and we can save the power to overdraw the whole scene and scale it back down to standard resolutions eliminating more AA need as well.

 

[Aquinus]

GTA4 and its desire for more than 152MB of vmem, even users with 700MB were suffering, while 1GB users were fine, it prevented a lot of the issues people complained about. So yeah, i see 8GB becoming the norm for high end video cards, it will allow for better high end textures that require less filtering and thus less GPU power to render, and we can save the power to overdraw the whole scene and scale it back down to standard resolutions eliminating more AA need as well.

...and everything being able to reside in VRAM means less waiting for data that exists in system memory. It's the same argument for system memory, the moment you have to swap data out of physical memory because you've run out, you're going to see a performance loss. Memory is one of those things where you always want more than what you're going to use because if you have less, you're going to run into performance impediments. I didn't get a 390 thinking that I'm going to fill 8GB, I got it with the expectation that it won't be long before I exceed 4GB and in surround, even Elite: Dangerous consumes up to 4.2GB on a regular basis now. Now, that could be textures and data that isn't used all the time but, not fetching that from system memory is far better than the alternative which is relatively slow. The more waiting the GPU has to do, the more the frame rate will fluctuate because the amount of time to access memory outside of VRAM isn't as consistent as pulling it from VRAM itself, so frame time will vary and your experience is going to suffer.

Simply put, having more VRAM is never a bad thing. It can be overkill, it can be useless in a lot of cases but, it's never a bad thing. Not having enough VRAM or system memory is far worse than having too much.