AMD engineer signals upcoming GPUs with GDDR6 tech

[HEXUS]

Principal member of AMD's technical staff has since 'corrected' his LinkedIn profile.

Read more.

[Bagpuss]

I think it's safe to say AMD want to move away from HBM as quickly as possible.

It's been a complete disaster for their GPU's.

Good technology but it just wasn't ready for mainstream in the quantity needed at the price required for the home GPU market, and ultimately, not much better than the latest GDDR offerings.

[DanceswithUnix]

I think it's safe to say AMD want to move away from HBM as quickly as possible.

And yet Intel are giving it a go, and Nvidia use it in their high end server parts.

TBH I think AMD are quite capable of shooting themselves in the foot without needing the help of HBM

[MrJim]

And yet Intel are giving it a go, and Nvidia use it in their high end server parts.

TBH I think AMD are quite capable of shooting themselves in the foot without needing the help of HBM

I think HBM2 makes most sense in an APU - the alternative is using DDR4 system RAM, and the bandwidth limitation that entails. There just wouldn't be space on the CPU package for multiple GDDR dies.

[TooNice]

Last time I briefly looked up GPU, I saw a lot of posters expecting big thing from HBM2, some even saying that they are holding back for it. No longer then case then?

[lumireleon]

the problem of HBM is not the supply or fabrication tech but an issue of demand, the demand is just not good enough.

[DanceswithUnix]

Last time I briefly looked up GPU, I saw a lot of posters expecting big thing from HBM2, some even saying that they are holding back for it. No longer then case then?

Vega has the same die size and memory bandwidth as a 1080ti, but not the same performance. The bandwidth is there though, so on the surface the HBM2 seems to be doing its bit, but it isn't a golden ticket.

the problem of HBM is not the supply or fabrication tech but an issue of demand, the demand is just not good enough.

I quite fancy a Vega 56, but can't find one at a sensible price. Sounds like price inflation through lack of supply to me. Am hoping custom cooler boards turn up soon.

[3dcandy]

the problem of HBM is not the supply or fabrication tech but an issue of demand, the demand is just not good enough.

I think the demand is there it all points to yields and supply still being poor. Remember with Vega the chips supplied to AMD and board partners is combined with hbm2, so even if AMD have plenty of Vega dies they can't do anything without the hbm2 module

[lumireleon]

........Nvidia sabotaged HBM by not implementing the chips on high end GTX 10- series. Samsung developed lower bandwidth cheaper versions of HBM2 but up to date AMD,Intel or Nvidia has showed no interest.

[shaithis]

I still say it's down to pricing due to a combination of demand and number of different production lines required currently.

When the industry is only producing 1 or 2 types of RAM, prices are low. At the moment we have a ton of different chips, fragmenting a market that is being bogged down by mobile devices.

[Dragonstongue]

I think it's safe to say AMD want to move away from HBM as quickly as possible.

It's been a complete disaster for their GPU's.

Good technology but it just wasn't ready for mainstream in the quantity needed at the price required for the home GPU market, and ultimately, not much better than the latest GDDR offerings.

that is hilarious to be saying seeing as it was AMD that brought forth to market GDDR3-4-5 HBM-HBM2 (to the masses)
why would they get away from HBM2 (the most recent version) that allowed them to keep power and temperatures in check (much more so than GDDR5 or 5x would have allowed them)

The only reason why Ngreedia went the route of GDDR5 and 5x was they were ABLE to do so while keeping power and thermals in check (much less complex chip design)

either way, to "claim" they want to get away from HBM as fast as possible because it has been a complete disaster for their gpu is asinine to say the least. they had to "clock it down" to keep power and thermals in check, not that they could not run it as fast as the spec allowed it to be run at.

Was Raja that made the call to make Vega the way it was made, pretty sure AMD has to work its arse off to coincide to keep performance where they want it to be having to fight every step of the way with devs automatically giving their direct support to "present looking" designs that Nvidia throws on the shelf and fanboys buy hand over fist.

anyways.....HBM is a very smart "design choice" as it uses much less power at WAY higher speeds than GDDR is capable of, there are "kinks to work out" this has been proven of course, going along this line, might as well say that Ngreedia should be doing everything they possibly can to distance themselves from using shoddy component selection on their primary and secondary capacitor selection (85 and 105c instead of 110 and 125c like they should be using) but hey, what do I know, I just LOOOOVVVEEE pissing my hard earned money away on a massively overvaluated company (Ngreedia....Nvidia)