![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
When you know that the GDDR6/GDDR5X bus will never be wide enough to offer the same or better bandwidth while you try to construe otherwise by constantly hyping "HIGHER PER PIN BW WOOOO" then yes it is.
Based on what? HBM is wide and fast remember. And latency is a tricky thing when issues like error correction come into play. GDDR in general is known for being fairly high latency.
But the total bus bandwidth for GDDR5X/6 and HBM will be very different. That is where you're being facetious and how you're attempting to misrepresent things. As the end user all that will matter is total bus bandwidth. Per pin bus specs is technical minutia.
And some things get lots cheaper over time. Plenty of historical examples of that too. And there is plenty reason to believe that HBM will be able to scale down in costs.
Have you already forgotten the Iris Pro? It wasn't high performing vs a dGPU but for the mobile segment it was quite impressive for a iGPU vs other discrete mobile solutions. The eDRAM cache it used wasn't even that big of a chunk of memory either. Again the future won't be exactly like the past.
. HBM advantages: smaller package, less latency, great bandwidth , lower power per a given bandwidth, makes AMD look good .
When NVidia has HBM on their gaming cards the naysayers here will say HBM was the second coming in which NVidia bless the world with
How much higher though? If you're going to assume HBM costs won't ever drop much then sure it won't ever make sense but there is no reason to believe that will be true. There also isn't any reason to believe the iGPU won't be able to use the extra bandwidth. They're already highly bandwidth starved and have been since ever. AMD and Intel are already perfectly willing to dedicate ever more die space to the iGPU too.
The first consumer GPU's to use HBM came out in mid 2015 and cost lots more than they do now. They're not hard to find either like they were at launch. You don't think we've seen huge improvements over a relatively short amount of time here?
Of course. It was probably in development for a long time before it started getting press on enthusiast sites or ratified by JEDEC.
Ask Nvidia. The first thought that comes to mind is profit margins though. They're like Intel in that they'll do the minimum to improve their products if they can keep their profit margins or increase them. Any business is these days.
1Tbps on a 512 bit bus and 512Gbps on a 256 bit bus. We both know you'll probably never see a 512 bit GDDR6 bus though so stop being factious. 256 bit or 384 bit are the most likely implementations. And that is real fast now but in 2018? Against HBM2 or HBM3? Or even a cost oriented HBM? Come on.
Which is pretty fast so what is your point?
Why don't you compare that to the 2Tbps of HBM2 or 4Tbps+ (will do more than a 8 stack remember) of HBM3? How do per pin bandwidth numbers matter so much when the peak bandwidth numbers are so different again?
But those have all improved massively just over the last year and half. You have to show that those can't improve at all or very little to make your point.
What? Tons of people bought it. Intel stopped making it, people didn't stop buying it. Especially in enthusiast circles when it was found the eDRAM cache improved general gaming performance a nice amount even if you weren't using the iGPU.
If we keep seeing improvements continue at their current rate? Sure.
Which has nothing to do with what I said. Its quite possible for AMD to get costs down lots and improve manufacturing volumes but still have financial issues due to mediocre sales + high development costs. You know this. You have to show that manufacturing costs and production volumes haven't improved much if at all.
Development time has nothing to do with production ramping or product announcements. You know this too.
Did you not read this part? "Any business is these days." If you're not going to read my posts, or worse, selectively read them then what the heck is the point of posting at all? This is getting ridiculous!
Which has what to do with costs coming down quite a bit on Fury/X + improved availability of those same cards? Since when is lowered costs + improved availability not a sign of improvements in manufacturing and lowered production costs?
It does with any other fabrication process. There are some unique steps involved for HBM but some of the more expensive parts (the interposer itself) will get cheaper over reasonable time scales.
OK but how does this matter much? One was a iteration/evolution of a existing tech and the other was something entirely new. And we already know what is coming for the next few years. Its highly unlikely something new will pop out of nowhere. Or that GDDR has much more scaling room left. If you want to go into fantasy land and do 'what ifs' based on nothing then sure you can create any scenario you want. But then what is the point? I can do the same and we can go back in forth and it'll all be BS.
No it doesn't. Of course there is need there. By and large the PC community doesn't want to buy dGPU's. Only enthusiasts do that for the most part. And there have been persistent rumors AMD will do a HBM APU using Zen for Raven Ridge in late 2017.
Who knows.. perhaps they just use AMD as a patent shield. Or they could be using GCN. Intel did mention they would support FreeSync while back, I wonder if they knew about the deal for awhile. But yeah I would like to know the answer to your question as well.
Same folks are here bashing AMD as in every other AMD thread, video or cpu.
To my understanding patents for current technology are going to be extended and AMD teams will be working on tech to be used in future Intel products. I do not think it is an either/or situation.
