[Kitguru ] AMD’s partners cannot get enough Radeon R9 Fury graphics cards –

[soccerballtux]

I'd be surprised if anyone was surprised by this. I would be even more surprised if they've even produced more than 10000 Fiji chips as of today.

that would surprise me too. There just aren't enough newegg reviews, and it's been out of stock every time I've looked

 

[x3sphere]

Man, the availability situation is really disappointing.

Newegg had the Sapphire card for like an hour, haven't seen stock anywhere else in the US. AMD seems to be having major supply issues, I guess the price is right for the cards because they're going to sell out regardless. I personally have not seen supply issues to this degree for any GPU launch ever.

 

[soccerballtux]

As God as my witness, it still doesn't show up for me. Is my Nvidia GPU blocking it from my searches? Haha.

lol

 

[LTC8K6]

IIRC, Pascal is said to use "3D memory", which isn't necessarily HBM.

 

[dark zero]

At first it was claimed that Fury Pro won't be any good, hence, only two AIBs (ASUS & Sapphire) have SKUs. Reviews done and dusted, the vast majority are positive.

Now its only 2 AIBs have SKUs due to lack of yields. On a cut-down part. Lack of yields on a cut down part.

It would make more sense if they speculate that HBM yields are rubbish and is limiting the entire chain of supply so we don't get good volume of SKUs using it.

While we're on this trend of speculative random stuff, let's throw this one to the mix.

According the usual rumor click bait:
http://wccftech.com/amd-working-entire-range-hbm-gpus-follow-fiji-fury-lineup/

They are suggesting that AMD is going to play very dirty with HBM2. They will leverage their close relationship with SK Hynix to get first dibs on HBM2 chips, all of it, before NV is allowed to get their hands on them. This means they will have first to market next-gen Artic Islands + HBM2 GPUs.

Actually AMD could ge tthe advantage and is not playing so dirty than expected since they are taking the bullet of the HBM1 failure storage and production...

BTW... I hear that there are lower and lower supplies of GTX 970 and 980... how truth is that?
If that so... seems that this and the next year could be very dark ages for gamers in GDDR5 and HBM1/2... Intel could get advantage of this.

 

[soccerballtux]

I wonder if it would have made more sense to make 390 line carry the Fury cards? Then when Fury is out of stock, people would just get the 380.

 

[ocre]

Actually AMD could ge tthe advantage and is not playing so dirty than expected since they are taking the bullet of the HBM1 failure storage and production...

BTW... I hear that there are lower and lower supplies of GTX 970 and 980... how truth is that?
If that so... seems that this and the next year could be very dark ages for gamers in GDDR5 and HBM1/2... Intel could get advantage of this.

Would fury be possible without HBM? Doesn't the traditional gddr5 memory, bus and memory controller use a lot of power? possibly even precious die space? Some where saying 50watts for chips less complex than Hawaii and its 512bit bus.

Would fury had been possible without HBM? It might not be biting the bullet as much as it could just be the double edged sword. Either way, hopefully this is all just temporary and the supply situation improves as time goes by.

If nvidia volume is much higher on the upper tier segments (and it would be hard to think it isn't), perhaps HBM 1 just wasn't feasible at all for them.
This would mean that both of them made some tough decisions a good while back. And so far, nvidia's is paying off better. But long term, this could play out very different. Who knows

 

[boozzer]

At first it was claimed that Fury Pro won't be any good, hence, only two AIBs (ASUS & Sapphire) have SKUs. Reviews done and dusted, the vast majority are positive.

Now its only 2 AIBs have SKUs due to lack of yields. On a cut-down part. Lack of yields on a cut down part.

It would make more sense if they speculate that HBM yields are rubbish and is limiting the entire chain of supply so we don't get good volume of SKUs using it.

While we're on this trend of speculative random stuff, let's throw this one to the mix.

According the usual rumor click bait:
http://wccftech.com/amd-working-entire-range-hbm-gpus-follow-fiji-fury-lineup/

They are suggesting that AMD is going to play very dirty with HBM2. They will leverage their close relationship with SK Hynix to get first dibs on HBM2 chips, all of it, before NV is allowed to get their hands on them. This means they will have first to market next-gen Artic Islands + HBM2 GPUs.

I don't see a problem with this at all. it is about time amd started to play dirty too. I 100% wish this was true. :twisted:

 

[soccerballtux]

I don't think 512bit buses are as big of a limitation as people think. Basically zero games are bandwidth constrained.

 

[Elixer]

For the availability of the Fury line of cards...
http://www.nowinstock.net/computers/videocards/amd/r9fury/
http://www.nowinstock.net/computers/videocards/amd/r9furyx/

Looks like there is a SAPPHIRE 100379SR, and a SAPPHIRE 100379OCSR now...

 

[3DVagabond]

That would imply that the story in the OP is correct, then. Not that any of us should be surprised. It is extremely normal to have delays, when introducing any new technology. The "supply is lower than demand" problem is the most normal one of them all to have, with any new technology. It honestly would have been a much bigger surprise to me if AMD had pulled this off without any delays or supply problems, since I doubt that either Intel or nVidia could have pulled it off without those same problems.

While Australia (Where Silverforce lives) is a large country it's pop is only ~24mil. and it's pretty far out of the way. I would assume that their allocations are fairly small.

FWIW there are 3 models shown in stock in NZ. They run ~$800 USD +15% tax Using the same conversion GTX 980 ti is around $650 USD +15% tax, and in stock.

This is the first time I've seen nVidia cheaper than comparable AMD product here.

 

[Silverforce11]

It's not just AU. I heard from e-tailers in the SE Asia region as well, they simply don't have much and its all sold out instantly.

From reading UK, EU forums, it seems same thing there.

So definitely its an issue of yield, not sure which, there were rumors regarding HBM yields are horrible, as well as Asetek not delivering (note these guys make almost every single CPU AIO as it is, and they had a huge order for servers).

 

[n0x1ous]

s well as Asetek not delivering (note these guys make almost every single CPU AIO as it is, and they had a huge order for servers).

? what does Asetek have to do with Fury supply and yields?

 

[MrTeal]

It's not just AU. I heard from e-tailers in the SE Asia region as well, they simply don't have much and its all sold out instantly.

From reading UK, EU forums, it seems same thing there.

So definitely its an issue of yield, not sure which, there were rumors regarding HBM yields are horrible, as well as Asetek not delivering (note these guys make almost every single CPU AIO as it is, and they had a huge order for servers).

There are a bunch of different CPU AOI manufacturers. Coolermaster makes some including the FuryX coolers. CoolIt also makes a bunch, including some of the most popular like the Corsair H80i and H100i. Outside those, there's smaller manufacturers like Swiftech. Asetek being constrained would have no effect on AMD.

 

[Sabrewings]

All AMD has to have is a contract for hynix chips and that would shut out nvidia if nobody else was making them.

You really think Samsung, Micron, Kingston, Winbond, Transcend, Mushkin, etc have simply been sitting around for a couple years with a standard for a new type of memory out?

I don't see a problem with this at all. it is about time amd started to play dirty too. I 100% wish this was true. :twisted:

Wish all you want. It's not possible since AMD and SK Hynix made it a JEDEC standard.

 

[Stuka87]

You really think Samsung, Micron, Kingston, Winbond, Transcend, Mushkin, etc have simply been sitting around for a couple years with a standard for a new type of memory out?



Wish all you want. It's not possible since AMD and SK Hynix made it a JEDEC standard.

Kingston, Winbond, Transcend, and Mushkin do not make memory. They just put their labels on it.

Micron, Samsung, Hynix, and Toshiba and the four companies that actually manufacture memory chips.

And yes HBM is a JEDEC standard, but thats relatively recent. Hynix has a huge head start on manufacturing it. The others will catch up, but its not an overnight process.

 

[Sabrewings]

Kingston, Winbond, Transcend, and Mushkin do not make memory. They just put their labels on it.

You're right. They do design some products but rely on the others for actual fab work.

Micron, Samsung, Hynix, and Toshiba and the four companies that actually manufacture memory chips.

To my knowledge Toshiba is not into DRAM, only NAND.

And yes HBM is a JEDEC standard, but thats relatively recent. Hynix has a huge head start on manufacturing it. The others will catch up, but its not an overnight process.

No, not overnight, but the HBM standard was published in October 2013. They've had plenty of time, and quite more so by the time Pascal will be launching.

 

[Stuka87]

You're right. They do design some products but rely on the others for actual fab work.



To my knowledge Toshiba is not into DRAM, only NAND.



No, not overnight, but the HBM standard was published in October 2013. They've had plenty of time, and quite more so by the time Pascal will be launching.

HBM1 standard was posted in 2013. HBM2 is what nVidia is wanting to use for Pascal. I do not have a JEDEC login, so I cannot see if the HBM2 stuff is already included in that, or when it was added if it is.

I have no doubt it will be there at some point, just not sure when that point is.

 

[Sabrewings]

HBM1 standard was posted in 2013. HBM2 is what nVidia is wanting to use for Pascal. I do not have a JEDEC login, so I cannot see if the HBM2 stuff is already included in that, or when it was added if it is.

I have no doubt it will be there at some point, just not sure when that point is.

Everything I can find on HBM2 points to it being a density improvement and nothing more. Thus, the standard (signaling and form factor) doesn't require an update.

I could be wrong, but I can't find anything that points to HBM2 being a separate standard or incompatible with the HBM1 standard. Just a density increase, and probably clock speeds if history is any indication.

Edit: Increased stack height as well, but that still shouldn't affect the standard.

 

[Genx87]

Wrong.
Look at what RAMBUS did.
So, yes, AMD could have patents on lots of things, and, make it very difficult for nvidia to do what AMD has done. Not so much for Intel, since they share technology.

And look where RAMBUS memory is today.

But the comparison isnt apples to apples. RDRAM was a proprietary standard licensed by Intel. In the mid 1990s inserted some of their patents into memory standards via JEDEC then lampooned the competition via lawsuits. In the end they lost. RDRAM hasnt been supported in 12 years in the PC world. If AMD or Hynix tried something similar they would also be the long term loser.

 

[Genx87]

always with this simplistic market share argument. It really does not matter to Hynix. If AMD buys up all their Chip capacity, even if amd was number 500000000, its money in their pockets. This silly market share is less so amd can't do anything sentiment is for the simple minded.

Plus they probably should be thankful AMD created it with them and they are the first to offer it (possibly only to offer for a year or more). It doesnt have to be a matter of exclusivity. All AMD has to have is a contract for hynix chips and that would shut out nvidia if nobody else was making them. Similar things probably happened with actual GPU chips due to TSMC capacity, in the past

It is like Nvidia wouldnt have sourced a second or third manufacturer. If AMD is willing to pay a premium to consume every scrap of capacity Hynix will oblige. I have a feeling AMD cant, and other manufacturers will be offering similar parts. Even if initial supply is constrained we are talking months, not years. Eventually it will be a total non-issue. Knowing AMD they would use whatever advantage they have from a potential shortage to gain 3 points of market share.

 

[werepossum]

What exactly did they do that is analogous to what's going on? They flirted with JEDEC for years but ended up pulling away because they didn't agree with the patent rules. Do you have an example of something they had a patent on, that was approved by JEDEC, and that they were allowed to enforce that patent on?

FYI:

https://en.wikipedia.org/wiki/Rambus

If they don't allow their standard to be licensed in a fair and non-discriminatory way, it's not allowed to be a standard. So, if Hynix/AMD (assuming they have an approved patent, as an application does not equal a patent) do not allow other memory manufacturers to produce HBM, the standard is revoked. This would allow a similar but different enough technology for another patent to become the standard and Hynix/AMD would be up a creek.

Interesting, thanks.

It is like Nvidia wouldnt have sourced a second or third manufacturer. If AMD is willing to pay a premium to consume every scrap of capacity Hynix will oblige. I have a feeling AMD cant, and other manufacturers will be offering similar parts. Even if initial supply is constrained we are talking months, not years. Eventually it will be a total non-issue. Knowing AMD they would use whatever advantage they have from a potential shortage to gain 3 points of market share.

Could it be that the reason NVidia is delaying adopting HBM is that they simply don't need it at the moment? The GTX980/980Ti with GDDR5 are pretty competitive with the Fury/Fury X with less risk and better margins, and oftimes being last to market offers the chance to do it right with less risk.

 

[n0x1ous]

Could it be that the reason NVidia is delaying adopting HBM is that they simply don't need it at the moment? The GTX980/980Ti with GDDR5 are pretty competitive with the Fury/Fury X with less risk and better margins, and oftimes being last to market offers the chance to do it right with less risk.

I think the capacity issue was probably a major concern as well

 

[Sabrewings]

Could it be that the reason NVidia is delaying adopting HBM is that they simply don't need it at the moment? The GTX980/980Ti with GDDR5 are pretty competitive with the Fury/Fury X with less risk and better margins, and oftimes being last to market offers the chance to do it right with less risk.

More bandwidth always helps, but I don't think the pros even came close to outweighing the cons. In my personal experience, the 980 Ti doesn't really show much gains with memory overclocking, so it isn't bandwidth constrained. Having the extra capacity is more useful. Just today I was up to 4.5GB flying around in Star Citizen.

 

[werepossum]

I think the capacity issue was probably a major concern as well

More bandwidth always helps, but I don't think the pros even came close to outweighing the cons. In my personal experience, the 980 Ti doesn't really show much gains with memory overclocking, so it isn't bandwidth constrained. Having the extra capacity is more useful. Just today I was up to 4.5GB flying around in Star Citizen.

The idea being that anything justifying the cost of HBM needs more memory than is currently feasible in HBM from cost, technical or availability standpoints? That makes sense - moving information in to and out of very high speed memory isn't necessarily better than just providing more memory, especially if you can do it while maintaining cost, heat and power advantages. Let the other guy have the bragging rights for being first and thereby dodge the learning curve issues.

Thanks, guys.