Friday, August 25th 2017
AMD RX Vega Supply Issues to Persist At Least Until October - Digitimes
DigiTimes is reporting, through "sources from the upstream supply chain", that AMD's current shortage of RX Vega cards to distribute to the retail market will continue at least until October. The tech reporting site says that sources are pointing towards the package integration of HBM2 memory (from SK Hynix or Samsung Electronics) and the Vega GPU (manufactured on Global Foundries' 14 nm FinFet) as being at fault here, due to low yield rates for this packaging effort. However, some other sources point towards the issue being with the packaging process itself, done by Advanced Semiconductor Engineering (ASE) through use of SiP technology. Whichever one of these cases may be, it seems the problem lies with AMD's choice to use HBM2 on their Vega graphics architecture.
As a footnote to its story, DigiTimes is also reporting that according to industry sources, NVIDIA has, in light of RX Vega's performance, decided to postpone the launch of Volta-based GPUs towards the first quarter of 2018.
Sources:
Digitimes, via HardOCP
As a footnote to its story, DigiTimes is also reporting that according to industry sources, NVIDIA has, in light of RX Vega's performance, decided to postpone the launch of Volta-based GPUs towards the first quarter of 2018.
92 Comments on AMD RX Vega Supply Issues to Persist At Least Until October - Digitimes
Everyone has a right to voice their opinions, but not to openly offend, troll or abuse others. If you cannot simply follow the TPU Forum Guidelines and be respectful to others, then please do not bother posting. Infractions will be handed out to those that fail to adhere to this.
Nobody here is entitled to attack anyone else. Period.
:toast:
They could have had a DDR5 powered polaris chip with 4096 cores last year, with similar perf/watt to VEGA, without wasting all that capital. The worst 480 pulled maybe 180 watt, so double that for a 512 bit 4096 core polaris would still be 50-100 watt less then vega 64 likes to pull.
When will AMD management learn?
If it really is the interposer that's the problem, that means good Vega chips are being binned because of its failings, which is essentially the same as AMD burning their money. I'm sure the situation will improve by year end, but man, that's the exact opposite thing they (and especially RTG) need right now.
And then there's Volta, which now has an extra half a year to be tested and tweaked to ensure that it pushes out its full performance and impresses everyone maximally on launch day. If Vega had been competitive we would've seen Volta launched with GDDR5/X, but now NVIDIA has time to go back and ensure it works perfectly with GDDR6, and the memory manufacturers have time to get production of the new memory to decent levels. Of course, should GDDR6 flop, NVIDIA already has the finalised GDDR5/X designs for Volta taped out, so they have all their bases covered.
Navi's launch in a year's time is looking both very optimistic and very far away for AMD...
Thanks to AMD, NVidia is in no rush to release VOLTA.
I guess people are easily pleased.
tellingspeculating on 2 different theories. Usually you want your sources corroborating the story not complicating it by speculating.Let people believe what they want and see what happens. I'm damn happy I didn't wait.
I don't buy 'low yields' arguments anymore btw, having a yield issue this late in the game just doesn't happen. Yields don't just change for the worse, unless you've got drunk employees handling the silicon. What really happens is that HBM supply is funneled to the high margin markets first, which is super obvious and the story of HBM's short life up till today.
So here we are, AMD employs every stalling trick in the book to keep supply low and keep the product scarce to drive up its price to a manageable level. You can be sure that even Project 47 (Hitman, anyone?) is a dedicated attempt to increase margin on Vega cards.
As for the reference design's cons, when was the last time that a ref design from any maufacturer was GREAT for high-end GPUs that didn't need the custom ones to get the best out of it for performance-noise-thermals?
Fanboys making endless excuses for AMD isn't going to make AMD fix its s**t. Negative publicity and voting with your wallets will. No company should get a free pass on repeated failures, regardless of how small and underdog-y it is.
And I really doubt adding another ~50W+ for GDDR5 memory and its subsystem plus another potential bottleneck by reducing the memory bandwidth would actually help the card...