Where did you see that the dGPU market is doomed ??? the entry/low dGPU segment is dead due to the AMDs and Intels APUs but the rest (from $60.00 and up) is doing just fine.
It's pretty daft logic, agreed. If AMD really pushed the boundaries on APU performance then they could shut NVidia out of certain markets entirely. Instead they dithered, and now Intel is beating them to it with the Iris Pro.
AMD should accept that the dGPU market is doomed, and try as hard as they possibly can to hold on to their IGP advantage and push it as fast as they can.
Where did you see that the dGPU market is doomed ??? the entry/low dGPU segment is dead due to the AMDs and Intels APUs but the rest (from $60.00 and up) is doing just fine.
APUs are devouring the dGPU market from the feet up. More segments of the dGPU market get made irrelevant with each generation of new APUs.
APUs will never get more performance than $60-70.00 dGPU market.
dGPU segment will not have any problem above that price point.
APUs will never get more performance than $60-70.00 dGPU market.
dGPU segment will not have any problem above that price point.
NTMBK is right in regards to pricing, and if AMD redesigns their memory controller, which rumor has it they are for Excavator (or quite possibly Steamroller too), then we'll see that iGPU performance mark creep into $70-85 dGPU territory too.
iGPUs will eat into dGPU sales up until die size constraint limits. You'd think once we get to stacked memory modules that it'd be all over, since after all the memory bandwidth issue would be largely resolved, but I don't think that'll happen. dGPUs will continue to stick around for one primary reason; separate die. iGPU performance will be limited by die size, transistor budgets, cooling, and thermal budgets; basically we should be looking at iGPUs eating dGPU marketshare up to 50-60w not $50-60.
Beyond the 100w mark, dGPUs will have no competition in the near future.
That train left long ago. 60$ dGPUs dont have a chance against IGPs. 80-90$ maybe. But thats just a matter of time before they lose to IGP as well.
dGPUs are simply slowing dying dinosaurs.
Okay then, which one of these gives better performance than an APU? http://www.newegg.com/Product/Produc...50%20-%20%2475 (And no, open box and refurbed don't count.)
That depends on the APU. Why do all here insist on only taking top of the line APUs into account? The HD2500/HD7480D is anaemic compared to almost anything else. Haswell isn't really doing anything for the really low end. (we get 4(!) more shader clusters... yay...)
Unless you have a magical way of cramming a Hawaii/GK110 level GPU + Haswell into a 60-80W TDP or solving the memory bandwidth problem, I'm going to have to disagree with you. Even the absolute highest speed DDR4 (4266MHz) announced is only good for 68GB/s in dual channel mode. My old 5770 (bought in 2009) has more bandwidth then that and still performs better then ANY APU. It will only get worse when 4K takes off.
You forget volume.
Oh, please. You are aware that something like 95% of PCs have shipped with an IGP for the last 15 years. Discrete has been a very low volume market for 10+ years and for those 10 years people have doomed discrete GPUs. Hasn't happened yet.
I remember this exact same discussion when the GMA900 was launched in 2004.
Unless AMD will produce a 200W APU, i dont see how the APUs will compete against $100> dGPUs.
Unless AMD will produce a 200W APU, i dont see how the APUs will compete against $100> dGPUs.
You have to ask yourself the question, when is the ROI too low for dGPUs to make business sense. At that point you wont see new dGPUs.
Don't AMD and Nvidia make more money from the professional market?
Don't AMD and Nvidia make more money from the professional market?
The bolded part is simply not true.
Nobody doomed dGPUs for 10 years either. And with Broadwell coming with up to GT4 with 2Tflops. Its obvious that IGPs keeps lowering the delta to dGPUs. The 3 year+ node cycle for dGPus now is not helping either. Add stacked memory and its really looking bleak for the dGPU.
I will believe Broadwell GT4 can do 2Tflops when I see it. That'll require between HD7850 and HD7870 level performance, and I just don't see that happening. Lets say they double the number of EU's from 40 to 80. That times 4 equals 320 shaders more-or-less (rough simplification). Even the 6800K has 384, the HD7850 has 1024 and the HD7870 1280. Then you have a very nasty bandwidth problem even with 2133MHz(34.1GB/s)/2400MHz(38.4GB/s) DDR3. I haven't seen that many notebooks with 2133 and higher RAM, but if we keep to desktop the HD7850/70 has 153.6GB/s available...
About stacked memory its still a while off for actual shipping products, and no one is stopping AMD/NV from using it with discrete GPUs. I even think I saw a roadmap somewhere where Maxwell's successor (Volta or whatever it is called) used it.
I dont think AMD make much money there. They are down to ~300M$ revenue per quarter on graphics, and it keeps shrinking. And HPC market is not getting easier with the Xeon Phi. For nVidia its a little different, but again, get certified drivers for IGPs and the market erodes. Just look at AMDs FirePro APU for example: