Supply of next-generation AMD Radeon R9 390-series graphics cards is expected to be very short initially. A media report suggests that the main reason behind supply constraints is limited availability of high-bandwidth memory made by SK Hynix.
Even though Advanced Micro Devices taped out its code-named “Fiji” graphics processor that powers the Radeon R9 390-series graphics cards in the middle of 2014 and currently its yields should be high, there is a major problem, which could hamper availability of such adapters: yields of high-bandwidth memory at SK Hynix. According to a report from TweakTown, there will not be enough memory available to fulfill AMD’s needs in the coming months.
The first-generation HBM (HBM1) stacks four DRAM dies with two 128-bit channels per die on a base logic die using through silicon vias (TSV), which results into a memory device with a 1024-bit interface. The first-gen HBM stacks are made using proven 29nm (which SK Hynix calls “2xnm”) process technology. High-bandwidth memory comes in HBM 4Hi stack (4 high stack) packages, which are tested by SK Hynix before shipments. Each HBM 4Hi stack is placed on an interposer next to the graphics processing unit, a process that has not been used for graphics cards before.
It is unclear whether yields of HBM memory is the only problem that affects availability of AMD’s Radeon R9 390-series graphics cards, or there is a combination of problems that involves HBM, usage of interposer, testing procedure of multi-die devices and so on.
AMD Radeon R9 390-series graphics adapters will be the first add-in-boards to utilize HBM memory. The new graphics processing unit from AMD reportedly packs 4096 stream processors/64 compute units, 256 texture mapping units and has a 4096-bit memory interface. The Radeon R9 390’s memory bus operates at up to 1.25Gb/s data-rate (1.25GHz effective DDR frequency) and delivers massive 640GB/s memory bandwidth.
AMD did not comment on the news-story.
Discuss on our Facebook page, HERE.
KitGuru Says: If AMD really has problems with production of its Radeon R9 390-series graphics cards, then it explains why the company wants to release the whole lineup of next-gen adapters at once. On the other hand, the company itself expects its GPU market share to increase after it releases its new graphics solution. If AMD cannot ship enough graphics cards, then its market share will drop, not increase.
Never mind the delay, if these really are 4GB max then it’s going to be worth skipping and waiting for 8GB+ Pascal. Not good for AMD…
i am a little confused here.
Wasnt the 390x a 8gb HMB and perhaps a 4gb HMB version ?
Can hardly trust nvidia on memory. 2014 and they couldn’t even manage proper 4GB on a $350 card. for years they were doing 2GB against amd’s higher bandwidth 3GB.
Not this again… Look the simple fact is that memory size is almost meaningless if you don’t have the processing power to support it. All of the tests that have been done with the 970 seem to suggest that the card really isn’t hampered by “only” 3.5gb as it really doesn’t have enough processing power to drive higher resolutions, unlike the 980. Sure memory can make a difference but the 3gb “for years” that you speak of was on cards that in all particle situations (aka gameplay) could of never used it.
I’d recommend that you check out this article: http://wccftech.com/fiji-xt-limited-4gb-memory/
I personally don’t believe the rumor as AMD would have to be brain dead to design a memory system limited to 4gb for a new architecture in 2015. High end cards are going to need more then 4gb and that’s a fact I’m certain their engineers are aware of.
The 290x can manage its 4GB fine and so could the 280x with its 3GB. If you want to claim the 970 is too weak to benefit from a full 4GB, I won’t stop you I guess. My point was simply that they are behind on memory
And, Nvidia would have to be brain dead to design a memory system limited to 2Gb on 128 Bit for it’s mainstream card in 2015, but…
I think this is more about nothing although if this rumor has any validity it hard being on the cutting edge of technology as it cuts both ways. The word back in Q4 2014 was that SK Hynix was ramping up availability of HBM memory to its partners.
As to the issue of 4Gb not being enough… as again “memory size is almost meaningless if you don’t have the processing power to support” And believe it or not 8GB of HBM will offer up to 1 Terabyte of bandwidth. It makes no sense for AMD to add extra memory if the GPU designs on 28nm physically prevents a die size that could exploit all that. AMD Fiji with 4096 SP’s, 50% of such 1Tb bandwidth would just be wasted.
◔❧❧◔❧❧◔I RECEIVED FIRST DRAFT OF $13000!@ak6:
,,,,
➨➨➨➨https://QuickerAdvicehina.biz/learner/kits….
In my opinion as someone who likes a stable 30+ fps minimum and idealy 60, neither the 290x or the 970 are anything better then 1440p cards. At that resolution, even with ultra textures, almost no games use more then 2-3gb of memory. Sure with Shadows of Mordor on 1440p the 970 starts to hit the 3.5gb cap, however it doesn’t seem to affect performance at all. With either of these cards anything higher then 1440p and sometimes even 1440p doesn’t produce good results as far as fps goes. Graphics in modern AAA games have pretty much peaked, so apart from higher resolutions, which these cards don’t have the power to support anyways, it’s doubtful we’ll see games which require more memory in the future. At 1080p, which is the resolution these cards will be used most at, anything over 3gb of memory will never be used and so it’s a meaningless number unless you start to do stuff like SLI or CrossFire, which is often not wort it anyways, and you’d really only want to do it on a top tier card like a 980.
I’d highly recommend checking out this video https://www.youtube.com/watch?v=v6k55epUBCE for comparisons on actual memory usage with a 970 in game and how it affects performance.
in many games the 290x at 4k ultra is way past your 30 fps minimum. This is not a situation where you can claim absolutes. Far cry 4 uses around 3.6GB on the 290x in at least one test at 1440p ultra. With minimum 39fps and average 50fps. The claim that they cannot use the full 4GB is silly because all you have to do is leave the vram consuming effects on and reduce other settings to improve fps. At which point the 4GB card bests the less than 4GB card.
Sounds like an apple launch, create supply issue rumors get all those pre-orders in.
(At over 700 a card, might be the only way to shift them)
Wasn’t it 6gb? Just joking
It wasn’t about whether or not 4GB was utilised. It was just that Nvidia omitted and outright misrepresented the capabilities of the graphics card. I’d rather they where upfront about it not find out later that ROPS Vram etc were not as advertised. Despite the success or failure of the class action resulting from that deception, it is still underhanded. We pay good money for these cards and its all in good faith.
As a previous owner of a GTX 970, I can tell you that whilst playing Dying Light and Shadow of Mordor I was using over 3.5GB Vram with texture quality set to high and I was forced to put settings on medium in dying light.
So, I definitely disagree with your point, that going over the 3.5GB Vram doesn’t affect performance at all. Despite getting high frames I was stuttering all over the place and in all honesty, I would rather have a card which can maintain solid 30 frames rather than a card that can produce 60 frames but with major dips in frames.
(I was playing on a 1080p monitor, and in moments when I was stuttering in game I was reaching the 3.5GB VRAM cap.)
Nah it will be ok LOL! They can just take old R7 260’s, R7 265’s, R9 285’s and R9 290’s and R9 290X’s and re-flash a 300 series bios on them, And then change the 2 to a 3! And boom no manufacturing problems!