^ The problem with x86 now is that to get even a 10% IPC improvement , the needed branch prediction, faster/larger cache, improved pipelines require a lot more transistors, which means shrinking the node is the best step.
Last i checked, a 8 'module' PD CPU had more transistors than Intels 8 'core' processor,
with a iGPU . And an HD4000 takes a considerable amount of die space (about 30-40% ?)
My point : for pure x86 performance, AMD is using more transistors and much more power (50% more ?) than Intel, to get similar perf in imultithreaded integer apps, and quite less in FP apps, and a
lot less in single threaded apps.
To reach intels IPC, AMD would need to add quite a lot more logic, which means quite a lot transistors. Assuming AMD has to work within the physial laws, a node shrink is their best option.
1) 8350 at much higher clocks consumes a little less power than the FX 8150 in idle and load states, so by deductive logic is the 8350 is clocked at 3.3ghz it will consume considerably less power than the FX8150 and the 1100T even comparible to Intels Hexcore parts. This is without much in way of die shrinks even on the exact same process. This to me suggests that Intels reckless drive to 10nm will not really yield the gains intended and if it is Intels endeavor to leave DT and
Your argument is facetious. A 8350 at 3.33ghz will be extremely shitty performing. Reason : longer pipelines. So :
1. Initial latency is high.
2. In case of branch prediction miss, the whole pipeline is flushed and refilled @ 3.3 GHZ. So super slowness.
Long pipelines are very good for throughput if :
1. Your clock speeds are high enough to compensate for a) initial delay. b) pipeline flush'
2. Excellent branch prediction.
#1 means increased power consumption increases, as P increases with the square of the frequency. So wither you remain stuck on the current node, or you can move to a smaller node. Intels "reckless drive to 10nm" is actually logical and practical.
#2 needs extensive CS
research, which needs money AND scientists AND engineers.
Would you care to benchmark mere Intel hexcore with a super 8 core AMD Piledriver @ 3.33 GHZ ?
go to mobility, even at that level they will not be able to compete with Samsung, Qualcom, Nvidia etc in mobility, so again die shrinks don't bring performance gains it is just Intel trying to drive its FABs into a universal 14nm/10nm process without diversifying it depending on the market.
Intels problem in mobile space is not technical, but the old mindset of management. They just cant 'allow' the consumers to get a fully functional chip. They have to trim features, and quote ridiculously high prices.
Plus, there is the architectural difference between x86 and ARM to consider. AMD is failing worse than Intel in the smartphone/Tablet market. So this means
1. x86 is not the best arch for low power.
2. There are no x86 software for mobiles. So users dont need a x86 CPU.
3. AMd's Temash is basically banking on its iGPU perf to win customers. And GPU is API driven, not related to x86.
4. To become competitive with ARM in both power and performance, x86 needs to be manufactured at a much smaller node.
5. If intel makes ARM SoC's at their 14nm node, do you think Sammy/QC/Apple have a fcuking chance ? No. But intel cant do that, becuase the market is already biased in mobility (which is ARM dominated). If they themselves make ARM, x86 has no chance in the mobile market.
Your argument sounds like its OK for TSMC and GloFo to move to a smaller node, but if Intel does it, its somehow wrong. Plus, what Intel has already done by moving to 22nm FinFET
alone , the
consoritum of IBM, GloFO, Sammy have not yet reached.
Intel already has 14nm working chips, which will have both low frequency version for mobiles, and high frequency version for DT. The high freq chips are much more harder to research and fabriate.
The consoritum is working to make only mobile chips, which are typically low freq.
I sense a major case of butthurt because AMD cant do for atleast 1 year what Intel has already done one year back.
Still, i applaud AMD for their perseverance and bringing HSA to the market. But you cant appreciate Intel for anything.
Now with AMD making some potent stuff in notebook, ultrabook and tablet forms, I think it will start becoming a problem for Intel who are still very dependant upon AMD and Nvidia for any value.
Intel has about 80% of the notebook market.
And 100% of the Ultrabook market.
Last i checked, AMD's ultrathin concept hasnt taken off the ground.
Tablets ? x86 has about 1% of the tablet market. Let AMD and Intel fight for that. Nobody can claim victory here. I suspect x86 cant get a bigger share.
Intel's iGPU will be nothing more than a iGPU with absolutely now Heterogeneous support
The advantage off a node win is that you can keep throwing more and more transistors at the iGPU to gain perf. Your arch may be shitty, but you will still get good performance.
adapting their parts to hybrid processing, by excavator we will have a fully fledged x86 processor working in tandem with a iGPU core packing some serious grunt existing in a space of harmony with each other, by then HSA will be more prominant.
I really wish HSA gets more prominant. But the software devs need AMD's support to use HSA. Its easy to get parallel processing from a embarassingly parallel problem (like video, rendering). Its the other more common, general softwares, which are difficult to parallelize. Unless AMD can do some automagic here, HSA will remain gimmick at best.
AMD is also releasing SSD's soon which will recieve support on all AMD chipsets to work in unison with AMD RAM disks for faster pre caching.
(This not aimed at you, but at another member
) :
How is this different from the "monopolistic, bloody, capitalist,manipulative, cheating, bastard" Intel ?