[TT] Pascal rumored to use GDDR5X..

[lilltesaito]

It depends how many cards have been developed for it. I dont think either AMD or NVidia did so for mainstream cards.

Would it not be better to just stick with DDR5 instead of moving to higher power DDR5X? Seems like a waste of money for the slower cards to have newer tech that will not help it.

I feel like there is a huge misunderstanding going on here. Or someone is changing goal post here.

 

[Headfoot]

I wonder what the cost difference would be between 256-bit GDDR5X @ 4GB vs single HBM2 stack @ 4GB. I could see midrange to low-midrange cards using just a single stack. IIRC it is 2Gbps per pin @ 500 mhz for HBM2, double that of hbm1, and we've seen people overclock hbm1 pretty far. A single 4GB @600mhz HBM2 stack should provide ~305-310 GB/s bandwidth, @500mhz ~256 GB/s bandwidth. ~300 GB/s + memory compression tech could allow for midrange/lowmidrange to not get bottlenecked by bandwidth. For comparsion, the 980 runs 256-bit 7ghz GDDR5 netting 224 GB/s plus memory compression tech and that seems pretty well balanced. Having just a single stack would cut down on cost compared to 4 dies. That comparison would be interesting but I doubt anyone here is at liberty to share that price info even if they had it

 

[Gikaseixas]

GDDR5X is only here because Micron knows HBM is so cost ineffective currently and for a foreseeable future

Source or it's just your opinion

 

[ShintaiDK]

I wonder what the cost difference would be between 256-bit GDDR5X @ 4GB vs single HBM2 stack @ 4GB. I could see midrange to low-midrange cards using just a single stack. IIRC it is 2Gbps per pin @ 500 mhz for HBM2, double that of hbm1, and we've seen people overclock hbm1 pretty far. A single 4GB @600mhz HBM2 stack should provide ~305-310 GB/s bandwidth, @500mhz ~256 GB/s bandwidth. ~300 GB/s + memory compression tech could allow for midrange/lowmidrange to not get bottlenecked by bandwidth. For comparsion, the 980 runs 256-bit 7ghz GDDR5 netting 224 GB/s plus memory compression tech and that seems pretty well balanced. Having just a single stack would cut down on cost compared to 4 dies. That comparison would be interesting but I doubt anyone here is at liberty to share that price info even if they had it

A single stack HBM2 would be between 204.8 and 256GB/sec. While a 256bit GDDR5X would be between 320 and 448GB/sec. So you need at least 2 stacks. You would also need 2 stacks anyway for 8GB. Since 4GB will not sell as 970/980 cards.

 

[3DVagabond]

Why on earth does the age of the tech matter? I want whatever best fits my needs. I couldn't careless how old the tech is. I bought a plasma TV last year despite it being an EOL technology and older than LED TV's because it's simply a better TV. It's hard to comprehend how someone would base their purchasing decisions solely on how old the tech is. That's absurd.

Good for you. I think paying top dollar for equipment that is aging tech when later superior technology is available is asinine. If I'm going to buy older inferior technology, I'm not paying top dollar for it. That's getting taken advantage of.

If you are OK with what you buy, I'm not really concerned with it. I hope you got a deal though on your EOL TV. If not, whoever you bought it from saw you coming.

 

[sm625]

I'm going to laugh when the next generation nvidia card with GDDR5X ends up being faster than the next AMD card, with 8GB of HBM2! Ok I will laugh and cry at the same time.

 

[AtenRa]

I really dont think its only about the cost but most probably it has to do more about low availability.

 

[AtenRa]

I'm going to laugh when the next generation nvidia card with GDDR5X ends up being faster than the next AMD card, with 8GB of HBM2! Ok I will laugh and cry at the same time.

GTX280 with GDDR-3 was faster than HD4870 with GDDR-5.

I dont remember anyone laughing with AMD being slower but the majority were laughing for the GTX280 price cut.

 

[PPB]

I really dont think its only about the cost but most probably it has to do more about low availability.

Its about yields. You have your normal die yields, right coming from the wafer. Then you have your yields AFTER the interposer and HBM ICs are gotten together. Apparently the methodology used right now is kind of lowering a bit too much the GPU die+Interposer+HBM ICs final yields. Also this more of a integrated solution makes die harvesting totally diferent. What would you do with a package that has 1 HBM IC gone bad?

As I said, the techonology is available, but it is in the stacking process that things have changed for the OEMs.

Lower yielding GPUs translate into reduced margins, and we know Nvidia will never be OK with lower margins. They would rather shove another generation of GDDR5/X cards. Its like Apple's bendgate, now chipgate, etc. That never stopped them from having 3 blocks full of sheep waiting for their shiny new product.

 

[Atreidin]

I'm going to laugh when the next generation nvidia card with GDDR5X ends up being faster than the next AMD card, with 8GB of HBM2! Ok I will laugh and cry at the same time.

That's really weird.

 

[dark zero]

I'm going to laugh when the next generation nvidia card with GDDR5X ends up being faster than the next AMD card, with 8GB of HBM2! Ok I will laugh and cry at the same time.

Maybe you will laugh because AMD ended to be the clowns that didn't won anything.

But you will cry since HBM2 will be a massive failure and nVIDIA will follow that destiny?

If that so... Both companies might die in the long run and Intel with HMC will be the future

 

[Azix]

GPU speed is still more important than VRAM speed as long as VRAM speed is decent. What to watch out for in this case would be power consumption, die size and w/e else would be affected. At least for the near future these are the major concerns with VRAM tech.

I don't think nvidia will be faster next year. Their big advantage in dx11 is gone. The playing field is more even now.

 

[Techhog]

It feels like some people here think that, if a new type of memory doesn't dramatically increase real-world performance on its own, it's a failure. Well... I guess that people are entitled to having opinions, even if the opinion is a stupid one...

 

[frozentundra123456]

All this speculation that somehow magically AMD is going to have HBM2 while nVidia doesnt, giving AMD a big advantage. It *could* happen I suppose, since AMD had a part in developing HBM, but it seems unlikely to me that a company with 20% market share is going to be able to strike deals with the manufacturers to corner the market on HBM2 while nVidia is stuck on GDDR5X. Even then, I am not sure how much HBM will benefit desktop cards. I could see how it would be compelling in mobile though, with the power and space savings. But again, AMD always seems to be introducing some innovative technology, lets see if they can execute it properly and leverage that into actually, you know, making money.

 

[Tecnoworld]

Frozen, remember that amd makes the apus for the last gen consoles. That should be a big help in their budget.

 

[tential]

The biggest improvement hbm brought was overall card size. That's what excites me. If we can get good quiet aio coolers that's be even better so that I can get all heat out of my case.

I can't wait for nvidia to follow because then companies will actually make mitx cases for these card sizes since amd is too much of a niche player to really drive sales enough to make a niche case like that.

 

[happy medium]

The biggest improvement hbm brought was overall card size.

Yea your right just like this HBM 1 gtx970 Mini....oh no nevermind it had GDDR5

Mabe it was this tiny HBM Fury card?

Point is it doesn't need to have HBM to be small or have to be small because of HBM.

 

[rgallant]

Yea your right just like this HBM 1 gtx970 Mini....oh no nevermind it had GDDR5

all 3.5 gb ?

 

[happy medium]

all 3.5 gb ?

+500mb! and the best selling 300$ card in 2014/15, amazing!

 

[tential]

Yea your right just like this HBM 1 gtx970 Mini....oh no nevermind it had GDDR5

Please explain to me how that is relevant to my comment?

HBM brought an overall reduction in card size. Is that a false statement?

 

[happy medium]

HBM brought an overall reduction in card size. Is that a false statement?

yes it is, see picture above I posted of gtx970.

You might have to elaborate about what "overall reduction in card size means".
Are you saying that without HBM cards can not be made smaller?
If so, your wrong.

You need to rebait your hook.

 

[ThatBuzzkiller]

yes it is, see picture above I posted of gtx970.

You might have to elaborate about what "overall reduction in card size means".
Are you saying that without HBM cards can not be made smaller?
If so, your wrong.

You need to rebait your hook.

You proved that video cards can be made more small without HBM but you did not disprove that HBM does brings in savings when it came to PCB size ...

 

[IllogicalGlory]

+500mb! and the best selling 300$ card in 2014/15, amazing!

Transformers 4: highest grossing movie in 2014, amazing!

"Most popular" is not synonymous with "best". I'd say that particular fact, like the one I posted, is more disappointing than something worth bragging about. People paid more for a similar experience, while supporting a company with dishonest businesses practices rather than paying less and punishing companies for lying about their products. NVIDIA didn't even apologize for that scandal, and they continue to advertise those half (or eighth )-truths on geforce.com.

 

[frozentundra123456]

Frozen, remember that amd makes the apus for the last gen consoles. That should be a big help in their budget.

Really? Is that what drives those huge profits they make every quarter?

 

[5150Joker]

If HBM 2 is in short supply, why wouldn't they use GDDR5X for their midrange designs? Since when is video card size a priority for anyone besides the very very tiny niche of people that build very specific HTPCs? Man, some of these arguments are really reaching. If AMD is smart, they'll follow in NVIDIA's footsteps or keep seeing production shortfalls and declining market share. I mean, what are they at right now, 18% and falling? Should be interesting to see what the conference call tomorrow reveals.