And that is OK. I think someone with a strong background in finance should be making decisions at AMD. Look at Nintendo. They completely flopped with the Wii U but every single console and game they sell at a profit. That means Nintendo is actually making $ selling each console and corresponding games, while MS's division keeps bleeding $ with Xbox. Sometimes it's better to have low market share and no popularity but sell products at a profit to survive until the next generation where you will have a chance to regroup.
Even if AMD's market share goes to 5%, as long as their GPU division is making profits, it's better than to have 35% market share and lose millions. I think AMD really just needs to survive until 1H of 2017 because I don't see their financial woes changing at all over the next 4 quarters. AMD has stated that they will have 2 new revenue streams, with a combined value of $1 billion. I believe these are design wins with MediaTek and new Nintendo console products. However, these won't even show up until late 2016. Right now, why waste hundreds of millions of dollars on redesigning the entire GCN stack on the ancient 28nm node when AMD knows that the real new generation is 14nm/16nm node? Would you redesign the whole stack knowing how short-lived this generation will be? Nope. NV is in a different position since they worked on Maxwell for 3-4 years which means GK206, 204 and 200 all received the attention they deserved. I can't possibly see AMD doing that level of SKU re-design this round. They should just focus on R9 380/380X/390/390X and re-badge everything below. Just throw R9 270 as R9 360 for $99, R9 270X as an R9 360X at $119, R9 285 as R9 370 at $149, 2048 SP Tonga XT 4GB as R9 370X at $199 and be done with it. I would love to be wrong but I don't see AMD doing anything dramatic on the low end.
I think considering just HOW late AMD is with this round, they better preserve as much cash flow as possible for 14nm/16nm against Pascal. I pretty much call this generation a write-off for them. They are almost a year late to 750/750Ti and soon will be a year late to 970/980, and 6 months late to a 960. Might as well just sell whatever they can at profits and establish strong image with R9 380-390 series. Maybe they can even throw R9 280X 3GB for $149 and wreck total havoc on the entire market at that level.
complete rubbish. If AMD is losing money at 25% market share how the heck are they going to make any money at 5%. AMD needs a new GPU stack to gain back market share and stop the bleeding. AMD is profitable as long as they have a market share of 35 - 40%. AMD will strive to provide competitive products and get back to having 35 - 40% market share. AMD will focus on the notebook GPU market where they have the lowest share. I am sure that R9 380X aka FijiXT will sport HBM as the benefit of HBM is most needed in notebook GPUs where AMD is power constrained. Nvidia with their superior memory bandwidth efficiency have been able to avoid the need for HBM with the GM204.
btw your theory of rebadges is horrible. Tonga is a proof of concept GPU. Its not a balanced GPU as it does not improve perf /sq mm over Tahiti which launched 32 months earlier. Its not a cost effective GPU at 360 sq mm. Selling a 360 sq mm GPU at USD 149 - 199 is not good for margins as the competitor has a 230 sq mm GPU selliing in the same price range. AMD designed Tonga to serve 2 purproses - as a test bed for architectural improvements and to be able to sell a 256 bit GPU with Tahiti like performance for Apple's notebook requirements as PCI-E MxM is restricted to GPUs with 256 bit memory bus.
http://en.wikipedia.org/wiki/Mobile_PCI_Express_Module#2nd_generation_configurations_.28MXM_3.29
AMD needs new chips specifically targetted at good perf/watt and perf/ sq mm at the right price points. Thats why I strongly believe AMD needs new chips across the stack, More importantly these chips should be competitive with Maxwell at perf/sq mm as thats how AMD can improve margins.
R9 390X Bermuda XT - 4096 sp, 4 shader, 4 raster engines and 4 tesselation engines, 64 or 128 ROPs, 8 ACE, 8 GB HBM, <= 550 sq mm
R9 380X Fiji XT - 3072 sp, 4 shader, 4 raster and 4 tesselation engines, 64 ROPs, 4 or 8 ACE, 4 GB HBM, <=400 sq mm
R9 370X Trinidad XT - 1536 sp, 2 shader, 2 raster engines, 2 tesselation engines, 32 ROPs, 2 ACE, 4 GB HBM, <= 250 sq mm
This GPU stack will allow AMD to realize good margins and I expect this to be their next gen GPU stack.
btw the majority of GPU volume in 2016 will be on the 28nm node. Both the GPU vendors are unlikely to ship their first 16/14nm GPUs before mid-2016. Even then the 16/14nm ramp will take many quarters and the overall volume of 28nm GPUs sold in 2016 will be much higher than 16nm / 14nm. I am guessing the ratio will be 3:1 for the total year. For H1 2016 28nm will be almost 90 -100% of GPU volume. By Q1 2017 I expect 16/14nm GPUs to overtake 28nm GPUs in volume. So you want AMD to bleed another 18 months with an outdated GPU stack. What an illogical statement.
You have to realize that there is a race for 16/14nm capacity and not everyone is going to get sufficient volume in 2016.
http://www.anandtech.com/show/9180/...results-strong-q2-but-lower-forecast-for-2015
"Qualcomm actually had negative cash flow for the quarter. The already mentioned fine paid to China accounted for some of it, and
Qualcomm also performed a prepayment of $950 million to secure long-term capacity from one of their suppliers. "
Neither Nvidia nor AMD can afford to pay that kind of money for securing long term leading edge capacity :biggrin: Apple, Qualcomm will get the lion's share of TSMC/Samsung 16/14nm wafers for 2016. The rest like AMD, Nvidia and Mediatek will be fighting for leftovers. In fact TSMC in one of their earning call last year stated that 16FF and 16FF+ combined will overtake 28nm capacity only in 2017.
Facebook
Twitter