HBM is a 3D Stacked Memory That Promises 9 Faster Performance Than GDDR5. Will Be Featured in Pirate Islands AMD R9 390X & 380X Cards.

[CarlosRex]

HBM was originally developed in tandem by both SK Hynix and AMD to replace GDDR5 as the new standard for bandwidth hungry processors such as GPUs.  GDDR5 performance scaling has slowed down dramatically and grown exponentially more expensive in the last few years. Faster GDDR5 modules and wider memory interfaces were only going to take us so far, a new standard had to replace the aging technology.

 

AMD Will release cards with this new memory next year.

The first graphics products to feature HBM will be AMD’s 20nm Pirate Island Rx 300 series GPUs in 2015.  In addition to all the new architectural improvements coming with the Pirate Islands graphics architecture. The Rx 300 series is expected to be the first to feature TSMC’s 20nm manufacturing technology a full shrink from the current 28nm technology. And also be the first to be equipped with HBM.

 

Nvidia will use it a year later in 2016.

Nvidia will be using 3D stacked memory on its GPUs starting with Pascal in 2016, a full year after AMD. Volta was Nvidia’s scheduled GPU for 2015 that was supposed to use the Hybrid Memory Cube or HMC for short which was s 3D stacked memory technology competing with HBM. However after HMC’s development fell behind the roadmap it put a lot of pressure on Nvidia to look for an alternative technology, which they did. What’s painfully ironic however is that Nvidia ended up using the AMD and SK Hynix co-developed HBM with Pascal. Which the company confirmed earlier this year. Nvidia very rarely uses a technology developed by their competition if ever.

AMD bet on HBM early on, while Nvidia bet on the competing HMC standard. It seems AMD bet on the right technology and they stand to gain an entire year’s worth of exclusive access to the technology. Nvidia will simply have to make due with GDDR5 for 2015.

 

Full Article

 

[Nineshadow]

I wonder if it will actually make a difference.

[virtualTune]

about time. they talked about this months ago. According to Nvidia, they will feature 3D stacked ram on Pascal...so a little while down the road ^^

 

Please do not use black text. Many users, like me, are using the dark theme! 

[Speedbird]

dude.. that's too damn fast...

[Jobbe03]

There's absolutely no way it'll be 9 times faster

[Archangel1994]

Only memory bandwidth is totally NOT the issue right now. Memory speeds/bandwidth are fine as they are

[Enderman]

Memory bandwidth is barely a bottleneck...so this will make next to 0 difference in fps.

But lower power consumption is always a plus, especially for AMD

[TERAFLOP]

Looks like the Hybrid Memory Cube will join XDR memory in the list of promising but abandoned memory techs.

[Levent]

I wonder if it will actually make a difference.

i gain like 1 fps when I overclock my 1300mhz gddr5 to 1550mhz so.

[TechFan@ic]

This reminds me of 2008 when AMD released the first card with GDDR5 memory the HD 4870. Nvidia had to use a memory interface that was twice as wide as the one on the 4870 to maintain similar bandwidth with the GTX 280. Which inflated the die size, power and most importantly cost.

Memory bandwidth is currently the most important bottleneck for APUs. Which is where I bet AMD is going to take this next.

[dlf]

Never heard of HBM till now.

[EmmaLong]

AMD cards always had more bandwidth than the nvidia equivalents. Which is why they performed better at higher resolutions and anti aliasing.

The HD 7950 for example beat the GTX 680 in most 2560x1440 benchmarks at the time. And it was positioned to compete with the GTX 660 Ti. Being 200$ less expensive than the 680.

[Opcode]

HBM has been available for a while now. It's nice to see AMD finally using the end product for intended purposes. All we need now is them to stack some HBM on-chip with the next generation of APU's. Come on AMD give me Carrizo with Excavator and 640 GCN 1.2 streams paired with HBM.

[patrickjp93]

This should be merged with the 390x/380x thread. This is a better explanation of HBM itself, but we already had the context it was coming on the 380x/390x.

[Kingofpants]

This is the kind of thing that will let you run more and Higher refresh rate high resolution monitors without really bad fps.

[Snickerzz]

uh, isnt the max speed of 16x gen3 lane 32GB/s?

[bcredeur97]

hmm i would like to have an HBM PCIe memory card for an APU...

[Opcode]

uh, isnt the max speed of 16x gen3 lane 32GB/s?

Indeed it is, tho the bandwidth of GPU memory is critical to how fast the GPU itself can operate. The issue isn't how fast the CPU and GPU can communicate, it's how fast the memory is to allow the GPU itself to do what it needs to do.

 

hmm i would like to have an HBM PCIe memory card for an APU...

I would prefer on package instead of an add-on card. Just stack 128 MB+ of it on-chip and then the CPU cores can utilize it as L3 as well (win/win).

[patrickjp93]

uh, isnt the max speed of 16x gen3 lane 32GB/s?

This is why Skylake E comes with PCIe 4.0. Both Hybrid Memory Cube and HBM have too much bandwidth for current busses.

[Trik'Stari]

Will it better allow me to run star citizen on ultra? at 1080, 1440, or 4k? at or above 60Hz? if so good, if not, who cares.

[Real_PhillBert]

Wow, AMD clearly has some big plans for the 300 series cards. If they can deliver on half of this it will make for an impressive product.

[patrickjp93]

Will it better allow me to run star citizen on ultra? at 1080, 1440, or 4k? at or above 60Hz? if so good, if not, who cares.

It will certainly help in the endeavor, but other GPU tech has to upgrade with it. That said, I doubt Star Citizen will be as taxing as the devs claim.

[flibberdipper]

I really don't find a point in going with EVEN HIGHER bandwidth memory. Hell, Nvidia still uses a 256-bit bus on the 980. Granted, it is pretty much the single-GPU king, but still.

 

LOL, like those 4 MB AGP cards for intel onboard GPUs ?

I have two of those. PC66 CL3 goodness.

[Opcode]

I really don't find a point in going with EVEN HIGHER bandwidth memory. Hell, Nvidia still uses a 256-bit bus on the 980. Granted, it is pretty much the single-GPU king, but still.

 

I have two of those. PC66 CL3 goodness.

The reason for that is both AMD and Nvidia has adopted a new technology known as lossless pixel compression. This is also why the R9 285 has only a 256-bit bus and out performs the R9 280. I personally am hoping we see GCN 1.2 implemented into AMD's next generation APU's. As this technology allows for up to 40% better memory bandwidth.

[flibberdipper]

The reason for that is both AMD and Nvidia has adopted a new technology known as lossless pixel compression. This is also why the R9 285 has only a 256-bit bus and out performs the R9 280. I personally am hoping we see GCN 1.2 implemented into AMD's next generation APU's. As this technology allows for up to 40% better memory bandwidth.

Everyone knows more bits is more better.