Originally Posted by Seronx
Will be on AM3+ and will have 12 cores.
I want to believe but what happened to RevC and RevD? Are you saying AMD will sell Warsaw chips for AM3+? Do not mess with me, that would be a dream come true.
My thoughts on AM3+ dying are that AMD released 220w CPUs and made motherboard manufacturers adapt. But why? So they could force mother board manufacturers to update their boards for a niche product hardly anyone is buying before the platform is phased out?
IMO AMD is getting people and motherboard manufacturers ready for some sort of beastly chip that is expensive. As I've mentioned before, FX 9590 is a "price conditioning" chip, just like Titan was. Very over priced and not worth the money, but when something comes out later for more than the previous high end but is cheaper than the price conditioning product, it looks much better.
GTX 780 would have people rolling over angry for $650 high end GPUs. Instead people turned around and went, "wow, a Titan LE for $650! THATS AWESOME!!!"
If AMD released a successor to FX 9590 that was cheaper and faster, the successor would look pretty awesome compared to FX 9590, even at $400. Compare that to FX 8350 at $200 and then doubling in price for high end, and then everyone is mad that AMD's "high end" doubled in price. With FX 9590, a $400 AMD CPU would now be "something much faster than $900 FX 9590 at less than half the price", which sounds way better than "AMD high end doubles in price!"
Also, as I said at S|A today, AMD can not fit a solid GPU and CPU in even a 220w socket. A high end GPU can be 250w+. How can AMD fit 250w GPU with 130w CPU in a single socket?
They simply can't. They will still need a CPU + HSA GPU solution if they want to do more than sell $150 APUs for the rest of AMD's life. Not to mention, as I said at S|A, GPGPU is only good for float. What will AMD do for multi-thread integer workloads? Just tell people to pound sand?
AMD already is trying to squeeze their way into professional market (FirePros in Mac Pro, OpenCL Adobe, etc). No way will any APU come remotely close in performance to one or two Tahitis in GPGPU for years. But by then a big GPU would still destroy it.
I think AMD will stay mum about things and then release a large upgrade. Perhaps new platform with DDR4 and HTX/PCIe slots where HTX can be used with AMD GPUs to enable HSA features. Now their high end platform is big AMD CPU, AMD GPU accelerator for HSA, AMD GPU for graphics rendering.
Imagine that cash involved in that, why would AMD toss this aside? Specially when CUDA is dying and people are running from it? And that PhysX is pretty dead end unless Nvidia gets it to run on GCN for consoles. Do you see that happening? Nvidia will throw tons of money at companies for PC PhysX exclusive far before they bite the bullet on that one.
FX CPUs are 30% of their sales from recent news. AMD's server market share is abysmal, less than 5%. If AMD is going to continue in such a weak market with updated products, why do you think they're going to drop their basically second largest CPU market outside of APUs?
APU might be 5x faster than just CPU, but it's still a mid-range GPU. Nvidia can come along, make a monster 500mm^2+ die for GPGPU that's good at OpenCL and then scoff at most of AMD's APUs. Meanwhile AMD sells APUs for $150 tops and Nvidia sells Quadros for thousands of dollars.
Why would AMD do this to themselves? They are touting HSA as the future, not APUs. There will be higher end HSA solutions than APUs, I do feel that AMD is starting from the cheaper price point and working their way up, however. It is much easier to get people to buy HSA gear under the guise of "cheap gaming computer" as opposed to "please buy this FirePro for $1500, HSA is coming soon for it, we swear"
Also, on Tom' someone contacted Roy or something at AMD about FX, and the response was "keep the faith", not "i can't say anything" or whatever, but a positive response that doesn't confirm anything.
So overall, APU only strategy is missing a few points:
1. Mid-range at best for anything because of power and heat constraints of a single chip
2. AMD completely discarding multi-threaded integer workloads
3. Throwing loyal customers who like FX 8000 for whatever reason under the bus while they wait for HSA applications to replace what exists
APU only future is not feasible, AMD needs big CPU + big dGPU to compete with Quadro, Phi, etc.