Hynix teases new High-End GPU with GDDR6 in early 2018

[Samwell]

SK Hynix Inc. (or ‘the Company’, www.skhynix.com)today introduced the world’s fastest 2Znm 8Gb(Gigabit) GDDR6(Graphics DDR6) DRAM. The product operates with an I/O data rate of 16Gbps(Gigabits per second) per pin, which is the industry’s fastest. With a forthcoming high-end graphics card of 384-bit I/Os, this DRAM processes up to 768GB(Gigabytes) of graphics data per second. SK Hynix has been planning to mass produce the product for a client to release high-end graphics card by early 2018 equipped with high performance GDDR6 DRAMs.

http://www.skhynix.com/eng/pr/pressReleaseView.do?seq=2086&offset=1
As we know that AMD is using HBM2 at least in the high end and Nvidia will pretty surely use HBM2 for GV100, i think this is the first real sign of big consumer volta. 384Bit interface with 16Gbps ram yielding in 768Gb/s sounds pretty nice.

 

[psolord]

GDDR6 my heiny(x). j/k

 

[IllogicalGlory]

Sounds like the Q3 Volta rumor isn't holding much water. Well, maybe we'll see GV107 in Q3 with GDDR5/X. Nothing high performance, though.

 

[DooKey]

Sounds like the Q3 Volta rumor isn't holding much water. Well, maybe we'll see GV107 in Q3 with GDDR5/X. Nothing high performance, though.

The same thing was said for GDDR5X and Pascal last summer. Keep holding out hope for your favorite company. I'm not saying GDDR6 will be out soon, but confirmation bias has happened before in the same context.

 

[IllogicalGlory]

The same thing was said for GDDR5X and Pascal last summer. Keep holding out hope for your favorite company. I'm not saying GDDR6 will be out soon, but confirmation bias has happened before in the same context.

What the heck is up with you? Why does discourse on this forum always degrade to taunts within the first couple of posts?

Sorry, but pretty sure we were literally just told, by the company that makes it, that they aren't putting GDDR6 in anything until early 2018. The GDDR5X thing was all speculation about supply and manufacturing capability. We knew it was available at the time of Pascal launch. I guess we could see an HBM or GDDR5X GV104, but that seems pretty unlikely to me.

Releasing Volta for the consumer market with in a few months seems like a bad move anyway. Pascal needs no replacement considering how feeble the pushback from AMD has been and they're selling it at huge margins. However, I could very easily see a Volta Tegra or HPC card happening in 2017.

Also, not everyone has a favorite company. I guess that can be hard to understand for some.

 

[DooKey]

What the heck is up with you? Why does discourse on this forum always degrade to taunts within the first couple of posts?

Sorry, but pretty sure we were literally just told, by the company that makes it, that they aren't putting GDDR6 in anything until early 2018. The GDDR5X thing was all speculation about supply and manufacturing capability. We knew it was available at the time of Pascal launch. I guess we could see an HBM or GDDR5X GV104, but that seems pretty unlikely to me.

Releasing Volta for the consumer market with in a few months seems like a bad move anyway. Pascal needs no replacement considering how feeble the pushback from AMD has been and they're selling it at huge margins. However, I could very easily see a Volta Tegra or HPC card happening in 2017.

Also, not everyone has a favorite company. I guess that can be hard to understand for some.

Like I said, everyone including the company making GDDR5X was saying Fall for mass production and not to expect anything. Guess what, it came out in NV cards in the summer. NO, most everyone said no GDDR5x in the summer of 2016, don't try to remake history.

 

[IllogicalGlory]

Like I said, everyone including the company making GDDR5X was saying Fall for mass production and not to expect anything. Guess what, it came out in NV cards in the summer. NO, most everyone said no GDDR5x in the summer of 2016, don't try to remake history.

Fine. I don't actually remember that. I'm not trying to remake history, and why you'd assume that...?

Nonetheless, you can't deny that this statement is making the the Q3 release of Volta look pretty unlikely. I'd think the word of a company involved in manufacturing the product in question would take higher priority than a rumor.

And once again, don't understand that taunting and the passive-aggressive tone of your posts. Is it possible to talk about things without that? Does hearing someone say something that casts doubt about a rumor you obviously hold pretty dear tick you off that much?

 

[Samwell]

Like I said, everyone including the company making GDDR5X was saying Fall for mass production and not to expect anything. Guess what, it came out in NV cards in the summer. NO, most everyone said no GDDR5x in the summer of 2016, don't try to remake history.

At the very beginning yes, but then the rumour websites were saying fall and were ignoring what micron was saying, when micron pushed it forward. Micron was saying mass production/availability middle of the year. They left the details unclear at the beginning and people made fall out of what they said. In march Micron even said, that there will be GDDR5X products launched till end of may in their conference call and the shareholders are the last people you lie to, but wtftech and others first would still write about a june or later launch.

 

[Shmee]

Cool it, both of you. Personal callouts are not allowed.

 

[SpaceBeer]

Micron will have GDDR6 by the end of the year
http://www.anandtech.com/show/11100...ce-roadmap-updates-forecasts-and-ceo-retiring

And we know nVidia has good relations with Micron (first to implement GDDR5X), while AMD has with SK Hynix (first to implement HBM)

 

[tamz_msc]

AMD has people working on GDDR6 as well.
https://in.linkedin.com/in/daehyun-jun-bb135a22

 

[itsmydamnation]

AMD has people working on GDDR6 as well.
https://in.linkedin.com/in/daehyun-jun-bb135a22

That guy has a pretty insane amount of exp with just ram interfaces.....lol

 

[raghu78]

768 GB/s using 16 Gbps GDDR6 running on a 384 bit memory bus. Wow. That aligns well with what a GV102 would generally sport. Since we already know Volta is scheduled for launch in late 2017/early 2018 this pretty much confirms Volta GV102 for Q1 2018.

We could also see AMD bring Navi based GDDR6 based products in 2018. I think GDDR6 will be the next ubiquitous graphics memory standard for consumer graphics. HBM2 will remain the choice for professional graphics and HPC.

 

[Krteq]

Hmm, all GDDR (and HBM) standards, starting with GDDR3, have been developed in cooperation of SK Hynix and ATi/AMD... except GDDR5X, which is only a standard extension (just a prefetch was doubled). In addition, all these new GDDR memory standards have been firstly adopted by ATi/AMD.

IMHO, GDDR6 is not an exception in this case.

 

[tviceman]

Hmm, all GDDR (and HBM) standards, starting with GDDR3, have been developed in cooperation of SK Hynix and ATi/AMD... except GDDR5X, which is only a standard extension (just a prefetch was doubled). In addition, all these new GDDR memory standards have been firstly adopted by ATi/AMD.

IMHO, GDDR6 is not an exception in this case.

You can label GDDR5X however you like to fit your narrative, but HBM2 was adopted first by Nvidia by nearly a year ahead of AMD. AMD was also always first to adopt new nodes, until this year. Times change. AMD's release cadence has fallen behind Nvidia's and manufacturers of new techs will not sit back and wait for someone to catch up if they can sell to someone else.

 

[Krteq]

You can label GDDR5X however you like to fit your narrative...

Well, Micron itself (Micron forwarded a technical design of GDDR5X standard to JEDEC for evaluation and standardization) is calling a GDDR5X as an enhancement to GDDR5 standard

While this speed improvement has been widely embraced by the industry, some aspects of GDDR5 SGRAM have become bottlenecks, resulting in data rate saturation at 8 Gb/s. In other words, GDDR5 SGRAM has hit the memory wall.

Micron addressed these bottlenecks with what now is called GDDR5X SGRAM. As the name suggests, GDDR5X should be considered a speed-enhanced derivative of GDDR5 rather than a radical new DRAM standard. This approach was chosen to let users leverage their previous investment in the GDDR5 memory ecosystem and enable a fast and low-risk transition from GDDR5.

Micron - TN-ED-02: GDDR5X: The Next-Generation Graphics DRAM Introduction (PDF)

...but HBM2 was adopted first by Nvidia by nearly a year ahead of AMD...

Yes, you are right.

nV was first on HPC server market with HBM2 fitted on P100 mezzanine modules (Vega with HBM2 has been also delivered to LiquidSky at the beginning of this year), but I think AMD will be first with HBM2 in gaming segment.

 

[tviceman]

Well, Micron itself (Micron forwarded a technical design of GDDR5X standard to JEDEC for evaluation and standardization) is calling a GDDR5X as an enhancement to GDDR5 standard
Micron - TN-ED-02: GDDR5X: The Next-Generation Graphics DRAM Introduction (PDF)

Yes, you are right.

nV was first on HPC server market with HBM2 fitted on P100 mezzanine modules (Vega with HBM2 has been also delivered to LiquidSky at the beginning of this year), but I think AMD will be first with HBM2 in gaming segment.

Yep, I could just entirely repeat what I said. I did not qualify anything with ifs, buts, and asterisks, and I also addressed the large elephant in the room of AMD falling behind Nvidia in release cadence. The chances of AMD being first to gddr6 are miniscule.

 

[raghu78]

One of the things we see from Nvidia's relentless pace of execution is they are now the first to new graphics memory standards. AMD which historically were the first to a new graphics memory standard like GDDR5 or HBM has fallen behind Nvidia when moving to GDDR5X or HBM2. Nvidia now seems to further that lead with being first to GDDR6 by a significant time to market lead. Even though Navi is expected to sport a next gen memory I doubt we will see it in 2018. I think Navi is a 2019 product and will be manufactured at GF 7nm. Volta GV102 should be a damn fast GPU with a massive 768 GB/s bandwidth. Its even more telling that Nvidia are beating AMD on raw bandwidth and have vastly superior color compression leading to impressive efficiency and excellent bandwidth utilization.

 

[crisium]

Quite right, Titan Xp 384-bit GDDR5X already has more bandwidth than the still unreleased HBM2 Vega (as rumoured). Unless AMD color compression has now not only caught up, but outright passed Nvidia, it's just one more disadvantage.

 

[Bouowmx]

GDDR6 'uneiled' in GTC: https://www.hardwareluxx.de/index.p...atz-auf-zukuenftigen-nvidia-grafikkarten.html
https://videocardz.com/69304/sk-hynix-showcases-first-gddr6-wafer-at-gtc-2017

 

[raghu78]

With Hynix revealing GDDR6 wafers at Nvidia GTC 2017 we have confirmation that big Volta will sport GDDR6 and have the highest bandwidth seen in a GPU in early 2018. Kudos to Nvidia for pushing next gen graphics memory technology. I think Volta will bring a repeat of the domination we saw during Maxwell generation.

 

[w3rd]

You can label GDDR5X however you like to fit your narrative, but HBM2 was adopted first by Nvidia by nearly a year ahead of AMD. AMD was also always first to adopt new nodes, until this year. Times change. AMD's release cadence has fallen behind Nvidia's and manufacturers of new techs will not sit back and wait for someone to catch up if they can sell to someone else.

HBM2 was only adopted by nVidia, because AMD allowed it. As they help co-develop the technology and have exclusive rights to HBM2, which is why AMD is in no hurry with Vega and why the 1080ti & Titan Xp were released so sudden. Also why nVidia has to seek GDDR5x, because AMD locked them out of HBM2 with exclusive rights.

Nvidia is allowed to use HBM2 in small quantities in the enterprise market. That's the deal.

 

[96Firebird]

HBM2 was only adopted by nVidia, because AMD allowed it. As they help co-develop the technology and have exclusive rights to HBM2, which is why AMD is in no hurry with Vega and why the 1080ti & Titan Xp were released so sudden. Also why nVidia has to seek GDDR5x, because AMD locked them out of HBM2 with exclusive rights.

Nvidia is allowed to use HBM2 in small quantities in the enterprise market. That's the deal.

You're completely wrong and have no idea what you're talking about.

 

[Arachnotronic]

HBM2 was only adopted by nVidia, because AMD allowed it. As they help co-develop the technology and have exclusive rights to HBM2, which is why AMD is in no hurry with Vega and why the 1080ti & Titan Xp were released so sudden. Also why nVidia has to seek GDDR5x, because AMD locked them out of HBM2 with exclusive rights.

Nvidia is allowed to use HBM2 in small quantities in the enterprise market. That's the deal.

This is not correct.

 

[raghu78]

HBM2 was only adopted by nVidia, because AMD allowed it. As they help co-develop the technology and have exclusive rights to HBM2, which is why AMD is in no hurry with Vega and why the 1080ti & Titan Xp were released so sudden. Also why nVidia has to seek GDDR5x, because AMD locked them out of HBM2 with exclusive rights.

Nvidia is allowed to use HBM2 in small quantities in the enterprise market. That's the deal.

AMD and Hynix co-developed HBM and submitted the technology to JEDEC to be adopted as an industry standard. Once JEDEC adopted the technology its not under AMD's control. Its under JEDEC's control. Samsung and Hynix are manufacturing HBM2 graphics memory. Nvidia has wisely chosen GDDR5X for consumer Pascal and GDDR6 for consumer Volta as these graphics memory technologies are easily manufacturable in high volume and do not have cost / yield / volume issues which HBM/HBM2 face. GP100 and most likely GV100 are HBM2 based and sell in ultra high margin datacenter products and also sell in much lower volumes than consumer Pascal or Volta.