[citation][nom]jupiter optimus maximus[/nom]Hurray for AMD! (Why did its stock fall to $0.14? It should be the other way as it gives AMD a huge pie of the counsel market and games will be optimized specifically with AMD APU architecture.)[/citation]
That was because the AMD financial results ended Q4 in the red.What the wall street annalist missed is that AMD did make profit from both the CPU and GPU divisions (heck; even the ram modules did gain a little money).But the one time payments from layoffs, the Sea MIcro buyout and the GloFo liquidation; did eat the earnings.Interesting is that the CPU division is still by far the main business for AMD not the GPU division,Even if AMD is closer to Nvidia than to Intel.
[citation][nom]TeraMedia[/nom]The specs on that APU look fantastic, and perhaps a bit fantastical. An 8-core CPU AND a 78x0-level GPU on a single die? I suppose if they're using a low-power mobile-oriented CPU module (which theory the 1.6 GHz seems to support) with something like a 65W power cap, then there might be enough leftover cooling headroom to support a 78x0 GPU module on the same die.The memory BW looks amazing. Rather than bolting a GPU onto the CPU, it looks like they did the opposite and bolted a CPU onto the GPU and its memory bus. The only concern I have with this approach is memory latency. GPUs are not impacted as much by latency because the latency is effectively divided across the 1000-odd threads that are simultaneously running. But we've seen in recent THG articles that CPUs - especially AMD CPUs - are susceptible to reduced performance from high latency. AMD doesn't have the fast, high-volume L1 and L2 cache systems that Intel has, so they depend more on the memory performance. And if the memory is high-bandwidth but high latency, this can have an adverse effect on the performance of CPU-bound processing such as AI and other linear processing threads. I hope AMD included a good, large, fast L3 cache on this thing.[/citation]
Feel free to correct me (whit links) but the Jaguar is not a modular design. Is a true core by itself; without the fatness of the behemoths desktop products from INTEL/AMD .But keep some of the goodies like out of order execution and some form of branch prediction.Consoles binaries are optimized at compile time so... a 8 threads each one with perfect coded alignments is something worth to be seen in action.
Also the 1.6mhz is not from a under-clock; it was a target from the beginning; the performance of this design is optimal at that speed.The other target was a 7watts per core so a 8 core at 65watts are feasible.Even if you add the GPU side for another 50 watts, thats still WAY lower power consumption than a PC running the same game.Not to be ignored if you live in a country where the electricity bill is a PITA.
Also i would LOVE to know how the GDDR5 - APU was bridged. Do not worry about the latencies as long the raw speed is keep high enuff. Waiting 15 cycles is not big deal if the clock is fast, and GDDR5 is very fast.The need for larger caches diminish; if the main ram is this fast.
Looking at the big picture, premium DDR3 are expensive, DDR4 still far on the horizon,IF this is the first glance of the Unified memory architecture for the GDDR5 memory type, AMD overall performance will blow out of the water any NVIDIA/INTEL products on naturally closed gadgets like Tablets, Nettops,All-in-Ones and netbooks variants.