WYP
News Guru
It is now rumored that Nvidia will be using GDDR5X memory with some of their upcoming Pascal GPUs, offering double the speeds of GDDR5 memory.
Read more on GDDR5X memory.
Read more on GDDR5X memory.
Nvidia to use GDDR5X memory with next generation GPUs
- Thread starter WYP
- Start date
TheF34RChannel
New member
Did I miss something? I've read that micron created it, but found no link to Nvidia whatsoever?
WYP
News Guru
The rumor is that Nvidia will be using it, apart from that no information is really available, so i explained exactly what the memory was.
GDDR5X is significantly less complex to move to than HBM as it is so similar to GDDR5, making it a great option for them to use until HBM can be manufactured in required quantities and all manufacturing issues for placing it on GPUs are resolved.
TheF34RChannel
New member
WYP
News Guru
It is likely that next generation GPUs will use HBM or GDDR5/x, with high end GPUs using the HBM and lower end GPUs using more traditional GDDR5.
It looks like HBM is in short supply, and why use the HBM on lower end stuff if it can be done cheaper with GDDR5 until production ramps up and matures.
AngryGoldfish
Old N Gold
tolagarf
Member
As others have mentioned there seems to be a shortage for HBM, so we can probably conclude that HBM 2.0 will be limited as well. This means only the professional and enthusiast (Titan etc) models of graphics cards are likely to be using HBM 2.0 in 2016.
Lets face it HBM in it's initial form didn't really impress anyway, certainly a lot of people expected more. So if they can almost double the bandwidth with GDDR5X, I say go for it! Makes no difference to me what it's named or how fancy it looks, as long as it kicks ass
Lets face it HBM in it's initial form didn't really impress anyway, certainly a lot of people expected more. So if they can almost double the bandwidth with GDDR5X, I say go for it! Makes no difference to me what it's named or how fancy it looks, as long as it kicks ass
NeverBackDown
AMD Enthusiast
Double the speeds just means double the power consumption and therefore heat. GDDR5 is already power hungry as it is so no every manufacture is going to have to have active cooling on memory and VRMs to keep it all cool. HBM is still ahead of it so don't really see why they would opt for this when it consumes more power when it's more than likely only going to go in lower range cards where it's all about power consumption.
HBM 2.0 shouldn't be as constrained for supply as HBM1 was. Now that it is JEDEC standard anyone can produce the design and not just SK.
HBM 2.0 shouldn't be as constrained for supply as HBM1 was. Now that it is JEDEC standard anyone can produce the design and not just SK.
TheF34RChannel
New member
And the rumour is accepted at face value and regarded as fact. And probably all because of some random guy spoke his mind in an obscure forum post somewhere. Oh dear... 
Could it be true? Sure, just as much as it might not be true. I'll wait for confirmation on whatever will be on what card
Ps: I didn't mean you WYP.
Could it be true? Sure, just as much as it might not be true. I'll wait for confirmation on whatever will be on what card
Ps: I didn't mean you WYP.
NeverBackDown
AMD Enthusiast
And the rumour is accepted at face value and regarded as fact. And probably all because of some random guy spoke his mind in an obscure forum post somewhere. Oh dear...
Could it be true? Sure, just as much as it might not be true. I'll wait for confirmation on whatever will be on what card
Ps: I didn't mean you WYP.
If it's enough of a rumor to get an article from more than one website then it it has a chance of being true. It's kinda like everyone saying Pascal will have 16GB HBM2 as a standard when that is obviously more of a fanboy wet dream. It would only appear on a Titan imo for the epeen
This rumor also contradicts the fact everyone is claiming Sk and Samsung will produce HBM2 for Nvidia.. now there's supply problems when we get another manufacturer producing it so they need to change to this? All rumors and that's why there's nothing being confirmed.
It would be interesting to see if Nvidia don't go down HBM2 while AMD will. It would mean more supply for AMD and since HBM2 is still faster and also will consume much less power it would give them quite an edge on memory and at high resolutions once again
AngryGoldfish
Old N Gold
This would be fabulous. Something tells me, though, that if AMD can get enough HBM2 chips, so could nVidia, and if they're not, maybe only AMD's flagship 'Fury X' and nVidia 'Titan' will have HBM2, with limited supply, and with lower tier GPU's still only coming with GDDR5 or HBM1. Is HBM1 limited to 4GB or was 8GB possible if supply could keep up?
NeverBackDown
AMD Enthusiast
It's limited to 4gb. I don't think hbm supply will be a huge issue. With Samsung joining in to produce them there should be just enough. Samsung are pretty great at producing memory so I have faith in them coping with high demand. SK are also pretty good so we should be fine tbh
AngryGoldfish
Old N Gold
I'm not as interested in buying AMD's flagship, to be honest. The 'Fury' (non-X) equivalent should be enough for 1440p gaming in 2016 and 2017. But if it still only comes with 4GB of HBM1, that might put me off a little. Hopefully they have their 290 (non-X) and Fury (non-X) versions of Arctic Islands with HBM2 and not just their highest end card.
NeverBackDown
AMD Enthusiast
4GB is fine for 4k even now. It won't be in a couple years but as of now it's fine. HBM2 can go upto 32GB so the next gen cards will probably all have them in the -mid high end and lower end versions might stick with HBM1 just because it'll more than likely be cheaper.
AngryGoldfish
Old N Gold
4GB is fine, yeah, but that doesn't stop the fact that I want more for my next card.
In all seriousness, while 4GB won't throttle a GPU that should theoretically be as powerful as an overclocked 980ti, it will limit the performance slightly. In games like GTA V at 1440p, a 980ti can draw over 4GB of VRAM. The same for AC Unity. With my Fury I only draw 3.3GB of VRAM, but that's because the GPU is simply not powerful enough to run the settings that draw so much VRAM. If I were to gain a big performance with the next AMD 'second-to-top' flagship, I'll be able to push the settings higher. While 4GB would only limit performance by a little, I'd prefer that not to happen. Also, VR might eventually draw me in.NeverBackDown
AMD Enthusiast
4GB is fine, yeah, but that doesn't stop the fact that I want more for my next card.
Next gen cards will guaranteed to have more than 4GB of vram as I already said. You kinda repeated yourself?:huh:
Also games will use as much vram as it can when there is enough available. That's why games running on a TX will consume a lot of vram because there is 12GB of it. If it's limited to 4GB it'll draw a lot less. Battlefield is pretty good example at doing this. However like I said already though, next gen GPUs will more than likely get large memory increases as resolutions rise and they both intend to be the better VR team, both are going to require more memory
The_Mard
New member
Heyyo,
True... but for that initial HBM showcase? That could be blamed more on the Fury than the HBM. The GPU could have been beefier, it could have also had better overclocking abilities on the Fury X... but it definitely didn't. The aircooled Fury was pretty darn good though yeilding results close to the Fury X despite being cheaper by quite a bit... my guess too is the drivers fell short. I mean, look at the R9 290X... it has only gotten faster over time as the drivers became better optimized... probably the same with with the Fury series.
But yeah, it did feel like AMD could have done just a little more and deliver that new GPU performance king like they were promising with their charts that supposedly beats out the GTX 980 Ti.
As others have mentioned there seems to be a shortage for HBM, so we can probably conclude that HBM 2.0 will be limited as well. This means only the professional and enthusiast (Titan etc) models of graphics cards are likely to be using HBM 2.0 in 2016.
Lets face it HBM in it's initial form didn't really impress anyway, certainly a lot of people expected more. So if they can almost double the bandwidth with GDDR5X, I say go for it! Makes no difference to me what it's named or how fancy it looks, as long as it kicks ass
True... but for that initial HBM showcase? That could be blamed more on the Fury than the HBM. The GPU could have been beefier, it could have also had better overclocking abilities on the Fury X... but it definitely didn't. The aircooled Fury was pretty darn good though yeilding results close to the Fury X despite being cheaper by quite a bit... my guess too is the drivers fell short. I mean, look at the R9 290X... it has only gotten faster over time as the drivers became better optimized... probably the same with with the Fury series.
But yeah, it did feel like AMD could have done just a little more and deliver that new GPU performance king like they were promising with their charts that supposedly beats out the GTX 980 Ti.
4GB HBM is fine right now since it has good color compression techniques... but there's no guarantee titles in 2016 won't push the VRAM envelope further though with Dx12 and Vulkan... if Vulkan even comes out in a different time frame... we are talking about Khronos Group hahaha.
AngryGoldfish
Old N Gold
Ahh, I'm just chatting. It's no worries. I'm not trying to preach or anything. I imagine next gen cards will have more than 4GB of VRAM.
And yeah, I've seen AC Unity draw a massive amount of VRAM on a Titan X, but it performs perfectly fine with 6GB or even 4GB. The same thing for SoM. A Titan Black will draw 6GB of VRAM while the 980 performs better whilst drawing much less.
Similar threads
