Tuesday, May 23rd 2017

NVIDIA GeForce "Volta" Graphics Cards to Feature GDDR6 Memory According to SK Hynix Deal

NVIDIA's upcoming GeForce GTX graphics cards based on the "Volta" architecture, could feature GDDR6 memory, according to a supply deal SK Hynix struck with NVIDIA, resulting in the Korean memory manufacturer's stock price surging by 6 percent. It's not known if GDDR6 will be deployed on all SKUs, or if like GDDR5X, it will be exclusive to a handful high-end SKUs. The latest version of SK Hynix memory catalogue points to an 8 Gb (1 GB) GDDR6 memory chip supporting speeds of up to 14 Gbps at 1.35V, and up to 12 Gbps at 1.25V.

Considering NVIDIA already got GDDR5X to run at 11 Gbps, it could choose the faster option. Memory remains a cause for concern. If 8 Gb is the densest chip from SK Hynix, then the fabled "GV104" (GP104-successor), which could likely feature a 256-bit wide memory interface, will only feature up to 8 GB of memory, precluding the unlikely (and costly) option of piggy-backing chips to achieve 16 GB.
Source: Appuals
Add your own comment

55 Comments on NVIDIA GeForce "Volta" Graphics Cards to Feature GDDR6 Memory According to SK Hynix Deal

#1
T4C Fantasy
CPU & GPU DB Maintainer
i knew it would be volta!! haha
Posted on Reply
#2
Prima.Vera
The GTX 1180 (2080 ?) will most likely have a 256-bit WMI and 8GB of GDDR6(?), probably at 12 Gbps, while keeping the 14 Gbps for the Ti version...
It's cool, since the memory is not the performace bottlenecker anymore.
Posted on Reply
#3
T4C Fantasy
CPU & GPU DB Maintainer
Prima.Vera said:
The GTX 1180 (2080 ?) will most likely have a 256-bit WMI and 8GB of GDDR6(?), probably at 12 Gbps, while keeping the 14 Gbps for the Ti version...
It's cool, since the memory is not the performace bottlenecker anymore.
Yea thats how i have it in db 12gbps for 1180 14 for 1185 16 can be for next gen

its 100% 11 series too
Posted on Reply
#4
eidairaman1
The Exiled Airman
So is this going to be exclusivity against AMD or will Hynix allow them to use it as well?
Posted on Reply
#5
bubbleawsome
Hm what if they go slower (12Gbps) and use a 384-bit bus for 12GB
Posted on Reply
#6
Nkd
T4C Fantasy said:
i knew it would be volta!! haha
I think the article is guessing about volta portion. Nvidia next cards are apparently based of Turing. Who knows if its based of volta I am sure some features are but rumor is might be just an evolution of pascal since there is no competition by amd this year. We all knew it was GDDR6 on the next high end cards but the volta portion on this write up is just speculation.
Posted on Reply
#7
T4C Fantasy
CPU & GPU DB Maintainer
Nkd said:
I think the article is guessing about volta portion. Nvidia next cards are apparently based of Turing. Who knows if its based of volta I am sure some features are but rumor is might be just an evolution of pascal since there is no competition by amd this year. We all knew it was GDDR6 on the next high end cards but the volta portion on this write up is just speculation.
only Volta is on Nvidias timeline, they wouldnt waste an architecture for one chip. they have already cut it down and Turing is speculated, only gamers nexus heard it was real but probably for mining.

literally every site thinks its turing based on nexus nothing else.
Posted on Reply
#8
Slizzo
bubbleawsome said:
Hm what if they go slower (12Gbps) and use a 384-bit bus for 12GB
Upping bus width is expensive; I don't think that NVIDIA would put that in their high end GPUs (GV104). More likely GDDR6 on a 256bit bus. 12gbps chips on a 256bit bus will net 384GB/s, which is respectable. If they go with 14Gbps on 256bit bus it'll net 448GB/s.
Posted on Reply
#10
Kaapstad
Prima.Vera said:
The GTX 1180 (2080 ?) will most likely have a 256-bit WMI and 8GB of GDDR6(?), probably at 12 Gbps, while keeping the 14 Gbps for the Ti version...
It's cool, since the memory is not the performace bottlenecker anymore.
Memory is the bottleneck for Volta

It really throttles the Titan Volta and even OCing the HBM2 by 25% to 30% is not enough.

Having said that GDDR6 for gaming may be more suitable than HBM2
Posted on Reply
#11
the54thvoid
Kaapstad said:
Memory is the bottleneck for Volta

It really throttles the Titan Volta and even OCing the HBM2 by 25% to 30% is not enough.

Having said that GDDR6 for gaming may be more suitable than HBM2
How does memory throttle Volta? If you look at games, Titan V isn't much better than 1080ti up to 1080p, gets a little better at 1440p, then starts pulling away at 4k.
If anything, the higher resolution should hurt a memory starved chip. Just like how HBM helped Fury X at higher resolution, when it stumbled at 1080p.
Saying Titan V is throttled by its memory requires some significant analytical proof (given it uses HBM2).
Posted on Reply
#12
Vya Domus
Prima.Vera said:

It's cool, since the memory is not the performace bottlenecker anymore.
Memory bandwidth is one of the primary concerns for any GPU design.
Posted on Reply
#13
eidairaman1
The Exiled Airman
Vya Domus said:
Memory bandwidth is one of the primary concerns for any GPU design.
Yup. Im gonna pm you with a theory to not clutter this thread up
Posted on Reply
#14
Kaapstad
the54thvoid said:
How does memory throttle Volta? If you look at games, Titan V isn't much better than 1080ti up to 1080p, gets a little better at 1440p, then starts pulling away at 4k.
If anything, the higher resolution should hurt a memory starved chip. Just like how HBM helped Fury X at higher resolution, when it stumbled at 1080p.
Saying Titan V is throttled by its memory requires some significant analytical proof (given it uses HBM2).
1080p throttles a lot of top end cards due to the CPU.

Yes the Titan V does pull away at 2160p but that does not mean that the memory is not throttling it.

The Titan V is a brute of a card for performance in gaming but it would benefit from faster memory.

Ultimately HBM will be the answer but it will most likely be version 4 or 5 where the clockspeed is high enough to keep up with fps.
Posted on Reply
#15
RejZoR
Slizzo said:
Upping bus width is expensive; I don't think that NVIDIA would put that in their high end GPUs (GV104). More likely GDDR6 on a 256bit bus. 12gbps chips on a 256bit bus will net 384GB/s, which is respectable. If they go with 14Gbps on 256bit bus it'll net 448GB/s.
Whole point of using high speed memory is to avoid using wider memory bus. It's cheaper to slam slightly more expensive and faster VRAM than making chip with wider memory controller.
Posted on Reply
#16
M2B
It's gonna be very disappointing if the new flagship uses only 8GB of memory @12GHz

but 12GB/16GB @14GHz should be fine.
Posted on Reply
#17
Upgrayedd
T4C Fantasy said:
only Volta is on Nvidias timeline, they wouldnt waste an architecture for one chip. they have already cut it down and Turing is speculated, only gamers nexus heard it was real but probably for mining.

literally every site thinks its turing based on nexus nothing else.
Yeah Turing, Ampere and Volta lol I keep thinking what's next? Schrodinger?
Posted on Reply
#18
Caring1
Upgrayedd said:
Yeah Turing, Ampere and Volta lol I keep thinking what's next? Schrodinger?
You forgot Tesla ….
Posted on Reply
#19
agent_x007
With Schrodinger as GPU codename, there is a chance it will be like Duke Nukem Forever :D
It's out there, but it's not.

Einstein ?
Planck ?

Tesla = G80/GT200
Posted on Reply
#21
Upgrayedd
Caring1 said:
You forgot Tesla ….
Wouldn't that be like saying I forgot GeForce?
Posted on Reply
#22
Imsochobo
M2B said:
It's gonna be very disappointing if the new flagship uses only 8GB of memory @12GHz

but 12GB/16GB @14GHz should be fine.
I do not expect Nvidia to massively improve but respectable improvements and much awaited.
I expect AMD to massively improve in gaming performance (like it could get worse? :) vega56 is the only great card in their entire lineup with hardware vs cost vs performance (MSRP!!!))

If so life won't be so bad, competition reduces prices :)

GDDR6 just isn't a good enough improvement and they have to increase width to make up for it which drives up the cost and then we're not getting bang for our buck.
Posted on Reply
#23
MT66
T4C Fantasy said:
only Volta is on Nvidias timeline, they wouldnt waste an architecture for one chip. they have already cut it down and Turing is speculated, only gamers nexus heard it was real but probably for mining.

literally every site thinks its turing based on nexus nothing else.
I don't think Pascal was after Maxwell on a older roadmap either. Its subject to change apparently.
Posted on Reply
#24
bug
Vya Domus said:
Memory bandwidth is one of the primary concerns for any GPU design.
And it's only going to get funnier when, because of HDR, we move to 10 bits per channel. 25% more bandwidth required right off the bat.
Posted on Reply
#25
jabbadap
Kaapstad said:
1080p throttles a lot of top end cards due to the CPU.

Yes the Titan V does pull away at 2160p but that does not mean that the memory is not throttling it.

The Titan V is a brute of a card for performance in gaming but it would benefit from faster memory.

Ultimately HBM will be the answer but it will most likely be version 4 or 5 where the clockspeed is high enough to keep up with fps.
Well Titan V has one HBM2 disabled, which cut that memory bandwidth to 3/4 from the maximum available. But is Titan V more ROPs limited or bw limited?

By raw bandwidth next gen. ti/titan with 384bit gddr6 bus will have more(16*384/8 = 768 GB/s vs Titan V:s 653 GB/s). But yeah brink of 7nm I wonder what are nvidia's plans. I.E. do they make only gv104 with 12nm and release it now(summer) and shrink it to 7nm call it Turing/Ampere 6-9 months later with full release from low to high end(GX108, GX107, GX106, GX104, GX102, X= A or T).
Posted on Reply
Add your own comment